Physical AI is artificial intelligence acting right where production happens: on the assembly line, inside the electrical panel, and on the machinery, not just in back offices. For a manufacturing company, this means deploying AI close to factory data (at the edge), enabling it to interface with a digital twin of the product or process, and overcoming friction between OT and IT systems. Let’s dive into the details.

What Is Physical AI (and why is everyone talking about It now?)

Physical AI represents the new frontier of artificial intelligence: no longer just systems that respond to text prompts in a chat, but systems that perceive, interpret, and act within the physical world through cameras, sensors, PLCs, and industrial machinery. Recent years have seen significant buzz around generative AI and agentic AI applied to text, documents, and back-office workflows. Physical AI applies this exact principle, giving AI the capacity to act, rather than merely answer, to the factory floor, where decisions confront physical realities: mechanical tolerances, vibrations, lighting conditions, and component wear.

For a manufacturer, Physical AI is far from an abstract concept. It is the distinction between a model that identifies a defect in a manually uploaded photo and a system that detects that same defect in real time on the line as the part passes under a camera. This shift, from manually loaded data to field-captured data, is the core of Physical AI.

From the cloud to the factory floor: why AI needs to be near the machine

Until now, most enterprise AI applications have resided in the cloud: language models, dashboards, and analytical tools accessible via a web browser. Physical AI changes the rules of the game because it must respond within the timeframes of the physical process it observes, milliseconds for a robotic arm, seconds for a visual inspection. It cannot afford the latency of round-trip communications with a remote data center.

Added to this are practical constraints unique to industrial environments: factory floor connectivity is rarely as reliable as office Wi-Fi, production data is often sensitive or bound by customer non-disclosure agreements, and streaming continuous high-definition video feeds from dozens of cameras to the cloud is cost-prohibitive. Consequently, industrial Physical AI relies heavily on edge computing: processing occurs as close to the data source as possible, while the cloud remains reserved for heavy model training, historical data aggregation, and multi-facility analytics.

The digital twin: The “common language” between design and reality

A critical second component of Physical AI is the digital twin: a dynamic digital representation of a product, facility, or process that is continuously updated to mirror physical reality. It is not a static simulation, but a live baseline against which Physical AI measures deviations: comparing what should be happening according to design specifications with what is actually taking place inside the machine, panel, or plant.

Revelis has long developed solutions in this domain. With Ei-TWIN, for example, we applied the digital twin paradigm to the energy sector, integrating real-time IoT data to optimize consumption and energy exchange within a pharmaceutical plant and an energy community. Physical AI generalizes this principle to broader industrial contexts: wherever a design exists (a CAD file, a schematic, a specification sheet) alongside a physical build, automated comparison between the two yields actionable value.

OT/IT integration: where friction arises (and gets solved)

Implementing Physical AI on the factory floor means tackling a long-standing industrial challenge: the gap between the OT world (Operational Technology, PLCs, SCADA systems, sensors, line cameras) and the IT world (ERP systems, data warehouses, cloud applications). These two domains operate with different protocols, lifecycles, and cultures, historically communicating poorly.

To function, Physical AI must act as a bridge between these layers: capturing and interpreting real-time OT data, normalizing it, and serving it in a format that IT systems (and personnel) can use for decision-making. We addressed this challenge through a different angle in the InCoP project, where environmental and biometric field data collected at a bakery (for Target S.p.A.) were processed using machine learning and reasoning to optimize shift scheduling. There too, value stemmed from enabling field devices to communicate with management systems. In Physical AI applied to quality and production, that same principle connects cameras, PLCs, and CAD files.

MHS: an emerging standard for connecting AI to physical hardware

The challenge of machine protocol heterogeneity is significant enough that on August 27, 2026, Anthropic released a research preview of the Model Hardware Standard (MHS): a specification designed to connect AI agents to physical laboratory and production instruments, aiming to reduce integration times from weeks or months down to hours or minutes. 

Technically, MHS introduces a standardized driver layer between the operating system and the hardware device, built on simple read/write primitives (e.g., “read temperature”, “set temperature”), automatic device discovery, and natural language tags documenting capabilities and safety bounds. It is designed to be model-agnostic and accessible via standard protocols like the Model Context Protocol (MCP), command line interfaces, or APIs.

Initial results shared by Anthropic alongside research partners demonstrate its potential:

• QuEra Computing (quantum computing): An AI agent reduced laser recalibration times from 150 seconds to 6 seconds while boosting the success rate from 58% to 99.3%.
• Carnegie Mellon University: Integration of dilution series for dose-response curve tracking went from weeks to roughly 8 hours, executing three times faster.
• Genentech: An agent automated a BCA assay for protease analysis, autonomously optimizing liquid flow parameters and identifying errors caused by air bubbles.
• University of Washington: The system remotely coordinated qPCR and instrument monitoring from a unified dashboard, featuring real-time tracking of amplification curves.
• HHMI Janelia Research Campus: Unified seven proprietary software programs for two-photon zebrafish microscopy, enabling real-time data analysis during acquisition.
• Tetsuwan Scientific: Managed a qPCR workflow for water pollution monitoring at San Pedro Creek, achieving 12% higher predictive accuracy than the manufacturer’s published specifications.

These milestones confirm that Physical AI, when backed by a unified hardware standard, can drastically reduce integration times across diverse domains, from pharmaceutical research to environmental monitoring.

However, existing limitations remain clearly defined: Anthropic notes that Claude lacks reliable “physical intuition”—struggling to predict collisions, inertia, or structural instability. High-risk decisions still require human approval, and safety roadmaps (hard stops, emergency shutoffs, error handling) are actively being refined prior to any open-source release. This aligns directly with the core philosophy behind Physical AI at Revelis: AI accelerates execution, but human oversight remains an essential layer. MHS serves as a clear indicator of industry direction, representing a potential abstraction layer for the OT gateways Revelis builds when integrating new machinery into PlugAIn.

PlugAIn: how Revelis operationalizes physical AI on the shop floor

To make Physical AI practical for manufacturing companies, Revelis relies on PlugAIn, our proprietary platform that models production processes using BPM (Business Process Management) and Big Data Analytics components. The core approach involves collecting and processing data generated by machinery, sensors, and line cameras to deliver actionable intelligence to decision-makers—whether an algorithm or a human operator—integrating edge processing and a comparison engine between design (the digital twin) and field reality.

PlugAIn combines three complementary layers of Physical AI:

1. Computer Vision to perceive the physical state of components.
2. Reasoning Engine to compare that state against the expected design model.
3. Orchestration Layer to decide, based on the output, whether to proceed automatically or involve a human operator.

This structural principle, AI proposes, humans decide on high-value tasks,  mirrors Revelis’s agentic AI solutions on the IT side, here extended to the physical environment of the factory floor.

A real-world case study: Aquila, physical AI for quality control (AI-REGIO Project)

The best way to understand the impact of Physical AI is through a real-world application. Within the European project AI-REGIO (Horizon 2020, grant agreement No. 952003), Revelis developed Aquila, an Automatic Quality Inspection system validated at Elettrocablaggi, an Italian manufacturer of electrical panels. It serves as a direct example of Physical AI: cameras observing physical products, a CAD model representing design specifications, and an automated engine comparing the two worlds.

Before Aquila, quality control for electrical panels was entirely manual. An operator visually cross-referenced each manufactured panel against engineering schematics, taking an average of 340 seconds per panel with an error rate of roughly 1 in every 250 inspected panels, typical metrics for a repetitive, labor-intensive process prone to human fatigue. This is precisely where thoughtfully applied Physical AI creates impact.

The Aquila solution combines two complementary AI techniques:

• Computer Vision: A Mask R-CNN object detection neural network identifies installed components and their precise spatial positioning from photos of physical panels.
• Symbolic Reasoning: The CAD design file is transformed into a logical representation, effectively the electrical panel’s digital twin. An automated reasoning engine based on Answer Set Programming (the DLV engine) compares the two graph representations (the vision-captured state vs. the target design) and generates a compliance report flagging missing, extraneous, or misaligned components.

Results recorded during testing highlight the power of quality-focused Physical AI:

• The object detection model achieved an Area Under the Precision-Recall Curve (AUC) of 0.947, indicating high reliability.
• Average inspection time dropped by 60%.
• Error detection accuracy improved by 30% compared to a standalone computer vision approach.
• The system advanced from Technology Readiness Level TRL 4 to TRL 6, marking a major step toward full industrial deployment.

Why physical AI matters for manufacturing SMEs

The Aquila project demonstrates that Physical AI is not an academic exercise, but a tangible competitive lever for small and medium-sized manufacturing enterprises (SMEs):

• Less time spent on repetitive inspection yields higher throughput using existing headcounts.
• Fewer errors translate directly to fewer returns, claims, and rework cycles.
• A digitized process generates an audit trail of data, imagery, and results, transforming daily operations into an intelligence asset over time.

Once validated on quality control for electrical panels, this framework extends naturally to any operational context involving design-versus-reality verification: predictive maintenance, assembly verification, or surface and weld inspection. Once introduced to a single line item, Physical AI scales value across the enterprise.

This goes beyond theory: Revelis actively applies Physical AI to live machinery monitoring. At Target, an IoT sensor network connected to PlugAIn monitors every production phase in an industrial bakery, from ingredient dosing to baking, identifying machine idle times to inform preventive maintenance schedules.

In another predictive maintenance project, sensors mounted on textile loom feeders stream data via MQTT to a recurrent neural network (GRU). The network predicts warp yarn breaks with 70% accuracy (AUC 0.67), providing advance warnings that allow operators to intervene before machine stoppage occurs. These implementations confirm that Physical AI delivers value equally well in direct machinery monitoring and quality control.

The Revelis vision in brief

Physical AI is not simply another cloud platform promise; it is the strategic choice to bridge artificial intelligence with the physical reality of production using edge computing, digital twins, and robust OT/IT integration. Through PlugAIn, Revelis consolidates insights gained across projects like Aquila (AI-REGIO), Target, Ei-TWIN, and InCoP to make Physical AI a practical, measurable pathway for small and mid-sized manufacturers. It is our hardware-adjacent expression of the core principle guiding Revelis: making AI actionable.

Mini-Glossary for Non-Technical Readers

Ready to Evaluate Physical AI for Your Facility?

Revelis guides manufacturing enterprises through every step of implementing Physical AI on the factory floor, from edge computing and digital twins to full OT/IT integration.

Learn more at www.revelis.eu · Contact us at info@revelis.eu

Authors: Massimiliano Ruffolo and Achille Abritta

L'articolo Physical AI and digital twins: AI enters the factory proviene da Revelis.

What is Moneying™

An anomaly indicator, no matter how well formulated, does not exist in isolation: it must run on real data, produce prioritized alerts, and return the necessary context to the analyst to make a decision. Moneying™ is the environment where this happens. The platform ties together customer risk profiling, transaction monitoring, and relationship analysis, leveraging the Rialto™ framework for large-scale data processing.

Moneying™ leverages both core approaches of Artificial Intelligence: deductive and inductive models. Deductive models are structured using regulatory rules produced by investigative authorities and remain explicitly expressed: readable, versionable, and individually verifiable.

Inductive models learn from data and step in where rules fall short: in risk grading, recognizing relational patterns, and reducing the noise that stifles compliance teams under false positives. Deduction and induction do not replace each other; they complement each other.

As a result, the foundational requirement holding the entire architecture together is that every output produced by the platform is accompanied by the underlying evidence that determined it, in a format that is readable by the analyst and usable in audit/inspection settings.

AML control automation: a constant challenge

Anti-money laundering resembles cybersecurity: a perpetual game between attackers and defenders. On one side, bad actors seek new vulnerabilities to exploit; on the other, defenders reinforce systems to close them. The Bank of Italy and investigative authorities oversee this front by regularly updating their guidelines, describing emerging schemes to intercept in natural language; financial intermediaries are tasked with translating those indications into active controls over transaction flows.

This is where the mechanism slows down. Between the regulatory description of an anomalous behavior and the executable rule capable of recognizing it lies a manual translation process: slow, subject to interpretation, and required to be repeated at every update. The result is a safeguard relying on static, cumbersome architectures to modify, leaving defenders chasing behind, trying to seal the breach after it has already been crossed.

To narrow that lag and achieve true AML control automation, it is essential to make the dynamic phase of “translation”—namely, the implementation of inductive and deductive processes for detecting anomalies in financial transactions. Generative AI, and specifically Large Language Models (LLMs), intervenes precisely at this stage: given a sufficiently precise description of the behavior to intercept, the model produces the sequence of chained operations that recognize it in data, and a downstream system deploys it into production. Thanks to this AML control automation, the new control becomes available as a guardrail in a matter of days, rather than months.

Moneying-GenAML™

The capabilities described above are implemented in the GenAML module of Moneying™, a Moneying plugin developed under the Moneying-Plus project. It is a new plugin engineered to shorten the gap between the publication of a new recommendation by authorities and its deployment into production, revolutionizing AML control automation. The starting point is the document as-is: a Bank of Italy circular, an anomaly template, or an investigative report. The framework extracts its control logic: which data to observe, with which filters, on what thresholds, and over what time frame, and translates it into an executable indicator for the institution’s transaction flows.

In practice, GenAML is a plugin that operationalizes AML control automation starting directly from natural language descriptions.

The output is neither plain text nor a code snippet to be integrated manually: it is a series of concatenated operations called a workflow inspectable step-by-step, which the analyst can review, refine, and validate before production deployment. The model does not replace the compliance officer; rather, it provides an operational draft for effective AML control automation on which officers exercise their human judgment, shifting human labor from transcription to verification.

Furthermore, the cycle does not end with the initial generation. When guidance is updated, or when false positives indicate a threshold is poorly tuned, the same framework acts on the existing indicator and produces a new version. It is this incremental revision capability that makes enforcement dynamic: adapting AML control automation becomes a matter of minutes, not weeks.

Large Language Models and Model Context Protocol

Entrusting a language model with free-form code writing in a regulated domain is a difficult proposition to justify: the risk of hallucination is concrete, and the outcome is difficult to audit. GenAML’s approach to AML control automation is different. The LLM is not asked to generate arbitrary software, but rather to plan: given the regulatory text and relevant information on the underlying data structure, it composes a sequence of operations selected from a closed, pre-validated catalog of primitives data acquisition, SQL filter, transformation, export and defines how they chain together. Naturally, these primitives are customizable.

This is where the Model Context Protocol (MCP) comes into play the standard used to expose available tools and their function signatures to the model. The protocol acts as a contract: it defines what the model can invoke and with which parameters; anything outside that scope is simply non-expressible. Far from being a constraint, this limitation is the guarantee: every plan generated for AML control automation is syntactically valid by construction, fully traceable in its individual components, and reproducible.

This gives rise to an architecture where model creativity is confined strictly where needed (understanding regulatory intent and breaking it down into steps), while execution remains in the hands of deterministic and certified components. A balance that renders such AML control automation fully compliant with audit requirements typical of the financial sector.

AML Control Automation: How the Plugin Works

Figure 1 illustrates the pipeline used to achieve AML control automation through workflow generation. The user uploads reference documents (PDF or text files, typically official circulars and data schemas) and describes the control to be implemented in natural language. From there, the process is automated. The model, instructed as a planner, outputs the pipeline in JSON format by composing available tools. The execution plan is displayed step-by-step to the user, allowing every decision to be verified before execution.

The final step is conversion: the pipeline for AML control automation is translated into a native workflow that can be downloaded and imported directly into the Moneying™ platform, which relies on Rialto™ as its execution engine. The outcome is a native asset within the existing data management environment, rather than a parallel system requiring separate maintenance.

Alongside generation, the system provides an editing mode: starting from an existing workflow, it can be evolved using a new natural language instruction an essential feature for dynamic AML control automation that, in daily anti-money laundering practice, occurs even more frequently than creation from scratch.

The effectiveness of GenAML in AML control automation

As with any system automating a previously manual step, the effectiveness of GenAML in AML control automation depends on key baseline conditions worth highlighting. The first concerns data rather than algorithms: the quality of generated indicators mirrors the quality of the underlying sources. Knowing in advance which data streams are queryable, how frequently they update, and the reliability of customer profile fields is the single most critical factor determining project timelines.

Furthermore, human oversight remains central to the platform: AML control automation focused on generation does not eliminate human governance; it repositions it, shifting analyst workload from writing to verification. The return on investment is highest when the compliance team is involved early in defining the presentation format of indicators and accompanying evidence. It is the legibility and transparency of the output that makes the workflow not only executable, but fully governed which, in a regulated domain, is the most critical property.

Conclusions

The gap between the regulatory description of an anomalous behavior and the executable rule that detects it has always existed; what has changed is the speed at which money laundering techniques evolve, and consequently, the cost of compliance lag. As long as translation remains a manual task, defenders will continue to play catch-up, hindering true AML control automation.

GenAML does not promise to replace human oversight; on the contrary, it presupposes it. It redefines boundaries: assigning planning for AML control automation to the language model (constrained to a validated catalog of operations), deterministic execution to the infrastructure, and final judgment on recommendations to the analyst. GenAML completes this chain by outputting native workflows for Rialto™, while Moneying™ provides the operational environment where indicators run on real-time data alongside statistical models and risk profiling.

The primary value is not merely saving development effort significant as that is. It is the capability that this form of AML control automation provides to adapt to regulatory updates within days rather than months.

Would you like more information about Moneying?
Contact us at: info@revelis.eu

Authors

Simone Mungari – Enterprise Solution Engineer
Francesco Cupello – Enterprise Solution Engineer

L'articolo AML Control Automation: How Anti-Money Laundering is Changing with the New Moneying™ proviene da Revelis.

Generative Artificial Intelligence has now firmly entered business processes, but the real leap in value for companies is not asking for answers: it is delegating tasks. This is the promise of Agentic Artificial Intelligence, where autonomous software systems plan and execute tasks from start to finish by coordinating tools, data, and models. For an organization, however, autonomy also means responsibility: reliability, security, traceability, and cost control become indispensable requirements. 

At Revelis, we support companies through this transition with a guiding principle: making AI actionable in a reliable, repeatable, and governable way. Below, we describe our vision for building agentic solutions including the use of on-premise Small Language Models the architecture that makes it possible, and a directly applicable use case example.

From Generative AI to Agentic Artificial Intelligence

With generative AI, the user formulates a request and gets an answer: text, code, a summary. With Agentic Artificial Intelligence, the user assigns a goal: the system breaks down the problem, decides which tools to use, executes the necessary steps, and returns an operational result. It is helpful to think of it as a virtual collaborator to whom you can delegate repetitive tasks that carry a time cost but do not require high-value decision-making.

This paradigm shift unlocks high-value use cases from back-office automation to software development assistance, from document management to the preparation of complex bids but it introduces a technical challenge that becomes decisive in regulated contexts: the unpredictability of language models.

The critical friction point of enterprise adoption: unpredictability

By their very nature, language models can give different answers to the same question. In a creative chat, this is an asset; in a business process, it is a risk. On a practical level, the impact is twofold:

• Compliance and control: Difficulty in faithfully reproducing an action taken in the past, logs that are not always complete, and obstacles in meeting regulatory and audit requirements (GDPR, ISO, SOC 2).
• Operational risks: Varying behavior for the exact same request, decisions that are not always traceable, and unpredictable costs.

Revelis’ vision stems from this premise: the autonomy of agents must be channeled into an architecture that separates decision-making (what to do) from execution (how to do it), making the latter reliable and repeatable.

Revelis’ Vision: Reliable Agent Orchestration

With Agentic Artificial Intelligence, Revelis’ agentic framework is structured into layers with clear-cut responsibilities. The core concept is simple: AI proposes, but it does not improvise. The sequence of actions that produces concrete results is defined by pre-established, logged, and verifiable procedures. In this way, for the same request, the system always executes the same steps in the same order a necessary prerequisite for a company to trust an agent.

The architecture layers, in simple terms

• Guardrail (security): A filter that blocks irrelevant or non-compliant requests at the source, before they consume time or resources.
• Process Catalog (process memory): The list of tasks the system knows how to perform: for each one, it defines what data is needed and the correct procedure to follow.
• Semantic Router (routing): Understands the request expressed in natural language and routes it to the right task.
• Data Extraction: Extracts necessary information in an orderly fashion; if anything is missing, it asks for clarification instead of “inventing” answers.
• Execution Engine: Executes the procedure step-by-step, with the ability to resume where it left off and reconstruct every action for auditing purposes.
• Security Chain (fail-safe mechanism): When the system is not confident enough, it escalates through subsequent levels up to human intervention, never acting blindly.
• Memory/cache: Reuses decisions already made to reduce time and costs, without ever putting data at risk or repeating sensitive operations.

Two adoption scenarios, one architecture

The strength of this approach lies in its adaptability: the same architecture fits two very different starting points, without forcing the company to choose between reliability and pragmatism.

A. Already structured contexts: leveraging existing assets

Organizations that have already adopted copilot suites and cloud models possess a valuable technological heritage, as well as established governance and security processes worth capitalizing on. In these cases, Revelis reuses the existing infrastructure and adds the orchestration layer on top: agents become reliable and traceable directly on the infrastructure the company already owns, enabling a fast rollout without requiring new hardware investments.

B. On-premise LLMs: Small Language Models (SLMs) for data sovereignty and cost control

SMEs and organizations that prefer to keep data within their own perimeter for privacy, compliance, or vendor independence can adopt Small Language Models run on-premise: compact models that, thanks to optimization techniques, operate on limited hardware. Integrated into the architecture, they power reliable and repeatable execution of tasks, with predictable costs and full data control. This is where Revelis’ model evaluation work becomes a tangible advantage.

On-premise Small Language Models (SLMs): models evaluated by Revelis

Not all compact models are created equal. Revelis expanded its testing to eleven coding-oriented Small Language Models, measuring the correctness of the generated code against a sample set of test problems. The minimum benchmark set by Revelis was to solve at least 30% of the sample problems: four out of eleven models passed this threshold.

The highest score was achieved by Qwen3-Coder-Next (60% of problems solved), which is currently being tested within a coding agent using a context window of up to 200K tokens. This is followed by Gemma4 26B-A4B (48%), top of the list for upcoming agent testing. Qwen3-Coder-30B-A3B-Instruct (40%) has already been integrated into a coding agent, but in practice proved to be slow, heavy, and penalized by a necessarily restricted context window (32K tokens).

A concrete use case: automating the tender process with Agentic Artificial Intelligence

To understand where agents generate a real competitive advantage, it is worth looking at a process that many companies know well: participating in public tenders. It is the perfect example of a complex, high-value process involving multiple stakeholders, where most of the work today is performed manually.

Why it is a difficult (and expensive) process without Agentic Artificial Intelligence

A structured organization may participate in hundreds or thousands of tenders every year. Each tender brings an impressive volume of documentation: technical specifications, tender rules, and annexes that often exceed 200 pages, sometimes featuring thousands of individual “lots” to evaluate one by one. All of this must be read, interpreted, and turned into a bid within tight deadlines and under rules that vary from entity to entity.

The work touches multiple departments: the tenders office, sales, technical experts, finance, management control, all the way to executive sign-off. Today, coordination happens largely via email and spreadsheets, triggering a domino effect: the same specification document is read independently by multiple people, multiplying the hours spent. Above all, the margin for error is minimal: a single misread requirement or a non-compliant bid means immediate disqualification and a lost business opportunity.

How Agentic Artificial Intelligence intervenes

Revelis addresses this process using Agentic Artificial Intelligence. Thanks to a multi-agent system, an orchestrator coordinates multiple specialized agents, each responsible for a specific phase. The output of one agent feeds into the next, culminating in the draft proposal. No agent “decides” on its own: each one prepares, verifies, and proposes, leaving the decisions that matter to human experts.

• Scouting & Triage: Monitors portals, identifies relevant tenders, and assigns a feasibility/suitability score, allowing attention to focus where participation is truly worthwhile.
• Call Analysis (Tender Document Analysis): Reads the technical specifications and tender rules, extracting lots, requirements, criteria, deadlines, and grounds for exclusion—always citing the source document and page number so every piece of information is verifiable.
• Checklist & Envelopes: Generates the list of required documents for the administrative, technical, and financial envelopes, flags what is already prepared, and assigns tasks to the right team members.
• Pricing Support: Combines the tender formula, cost structures, and competitor estimations to propose a reasoned price alongside various scenario simulations. It proposes and explains, but the final decision remains human.
• Quality Simulation: Steps into the shoes of the evaluation committee: it applies the scoring rubric, estimates the proposal’s score, and highlights how to bridge any gaps to maximize points.
• Drafts & Memory: Prepares initial proposal drafts using company templates—eliminating the need to start from scratch—and builds a historical market memory (prices, outcomes, similar tenders) valuable for future bids.

The competitive advantage for companies using Agentic AI

Employing Agentic Artificial Intelligence means that the combined effect of these agents isn’t just about “doing things faster” it’s about competing better. By drastically reducing preparation time, companies can participate in more tenders including those currently discarded due to lack of time and focus on the ones with the highest chance of winning. Reducing compliance errors protects against disqualifications. Structured preparation helps avoid leaving money on the table: a more accurate bid raises the average winning price. And historical memory transforms even lost tenders into reusable market intelligence.

Governance and controlled evolution with Agentic Artificial Intelligence

Deploying Agentic Artificial Intelligence means utilizing an enterprise agentic system that is not static: it must grow without losing control. In the Revelis framework, every new capability starts as an activity in the process catalog and follows a governed path: drafting, overlap check, “shadow” testing (running in parallel without execution, to measure reliability on real-world cases), approval, and production deployment. Every human intervention thus becomes a learning opportunity, enriching the system in a fully traceable manner.

The result is an AI that observes, logs, and improves: every step is documented for the benefit of control, security, and continuous optimization.

The Revelis vision, in summary

Introducing Generative AI and Agentic Artificial Intelligence into a company does not mean giving up control. It means designing an architecture where reliability stems from the separation between decision and execution, security precedes all processing, and execution follows defined procedures rather than improvised choices.

Whether starting from an already structured copilot ecosystem or an on-premise setup based on Small Language Models, and whether the goal is automating back-office processes or winning more tenders, the objective remains the same: reliable, secure, and governable agents capable of generating measurable value.

It is our interpretation of the principle that has always guided Revelis: making AI actionable.

Mini-glossary for non-experts

• Generative AI — Systems that produce content (text, images, code) in response to a prompt.
• Agentic AI — Systems that go beyond simply answering questions to execute delegated tasks by coordinating multiple steps.
• Agent — A specialized software “virtual collaborator” assigned to a specific task; multiple agents can cooperate with each other.
• Orchestrator — The coordinator that assigns tasks to individual agents and synthesizes their results.
• LLM / Small-LLM — The language model that “reasons” over text; Small-LLMs are compact versions capable of running on limited hardware.
• On-premise — Technology installed and managed directly within the company’s premises rather than in the cloud, keeping data in-house.
• Human-in-the-loop — An approach where humans validate and approve what the AI proposes, ensuring full control is maintained.

Want to evaluate Agentic AI in your company? Revelis supports enterprises and SMEs in designing agentic solutions from model analysis to orchestration architecture, in the cloud or on-premise. Write to us at info@revelis.eu

Authors

Achille Abritta – Enterprise Solution Engineer at Revelis

Francesco Cupello – Enterprise Solution Engineer at Revelis

Luigi Granata – Enterprise Solution Engineer at Revelis

L'articolo Agentic Artificial Intelligence, SLMs, and Possible Applications proviene da Revelis.

• specific response deadlines
  
• full traceability of every complaint
  
• complete transparency towards customers
  

However, effective complaint management goes well beyond regulatory compliance. Each report highlights something that has not worked as expected. A well-designed process can therefore also become a powerful tool for improving internal operations.

Banks receive hundreds of messages every day through multiple channels emails, online forms, chats, and dedicated platforms. These range from simple information requests and suggestions to detailed reports and formal complaints. Handling this volume quickly and accurately, particularly for complaints, can provide a significant competitive advantage. A streamlined end-to-end process reduces the risk of sanctions, boosts operational efficiency, and improves customer satisfaction and loyalty, ultimately strengthening the bank’s reputation.

Traditional complaint management: risks, delays and operational burden

Carrying out this work without technological support, particularly without AI-powered Text Intelligence, is increasingly unsustainable. The process demands considerable time and manual effort, including:

• reading communications and any attachments
  
• determining the correct category and the appropriate response
  
• transferring messages to the dedicated management system
  

These tasks represent only a fraction of what could be automated, yet they are still performed manually in many institutions. The result is wasted time, misallocated human resources, and inefficient complaint management.

For customers, this often leads to:

• slow or missing responses
  
• inconsistent classifications
  
• excessive workloads for staff, usually due to repetitive tasks
  

Such problems inevitably influence the customer’s perception of the bank. That is why solutions capable of intelligently classifying and routing incoming communications are becoming increasingly important.

Moneying: Revelis’ solution for complaint management (and more)

Text Intelligence systems combine natural language processing with machine learning to analyse text, such as the body of an email and classify it, even across multi-level category hierarchies.

The goal is not to replace human work but to enhance it, streamlining operations and significantly reducing the time required.

With this in mind, Revelis developed Moneying, a platform that applies Text Intelligence to optimise complaint handling and, more broadly, to classify large volumes of communications.

Moneying uses advanced NLP and machine learning models to automatically interpret the meaning of a message, identify the correct category, and determine its priority.

Training these models requires previously classified messages. The larger the dataset, the better the results. However, when sufficient data is not yet available, automation can begin using:

• zero-shot learning
  
• few-shot learning
  

These pre-trained models understand a wide range of general-purpose text. As the bank’s dataset grows, they can be refined or replaced with a new model trained exclusively on the bank’s own messages.

Alongside these sophisticated techniques, the platform also supports customised rules to assign specific categories for instance, rules based on keywords in attachments, email subjects or bodies, or sender domains without engaging the machine learning models.

If a customer complains about a blocked credit card, Moneying can instantly distinguish this from an enquiry about mortgage terms or a report about a mobile-app malfunction. This speeds up processing and routes the message to the right team, reducing errors. Because the system learns continuously from historical data, its accuracy improves over time.

The result is more efficient complaint management, with:

• consistent and timely replies
  
• greater traceability
  
• fewer repeat complaints
  
• reduced risk of sanctions
  

For banks and organisations managing large volumes of email, Moneying transforms complaint handling from a manual, time-consuming task into an automated, reliable and measurable process benefiting both staff and customers.

Additional features include:

• customisable classification rules
  
• full lifecycle management of each case
  
• real-time monitoring of processing status
  
• tracking of pending cases
  
• electronic document archiving
  
• automated reporting for Internal Control Functions
  

Beyond complaints: wider capabilities

Moneying is a comprehensive RegTech platform designed to handle a broad range of needs across the financial sector, from complaints to privacy requests and documentation workflows. Its centralised approach optimises the full lifecycle of each case, from registration to resolution and archiving.

Further capabilities include:

• automatic retrieval of customer data from the institution’s central records, with support for registering cases relating to non-customers
  
• full traceability of all departments involved Legal, Compliance, Branches—through collaborative workflows and an auditable activity log
  

automatic generation of standardised documents using pre-defined templates populated with case data, ensuring consistency and speeding up responses

Monitoring and regulatory oversight

Moneying supports complaint management with continuous monitoring of critical issues and KPIs, enabling process optimisation. It also detects repeated and “invisible” complaints and prioritises requests from authorities such as the Bank of Italy and the Financial Ombudsman.

A secure corporate portal provides controlled access and allows integration with other systems. Custom dashboards show processing status in real time, and proactive alerts help ensure compliance with regulatory deadlines and internal SLAs.

The future of text intelligence in banking

Text Intelligence is becoming a strategic lever for smarter, more proactive, data-driven complaint management. As language models evolve and integrate with generative AI, banks will be able to:

• deliver personalised, context-aware responses instead of relying on fixed templates
  
• analyse communication trends to monitor sentiment and identify reputational risks
  
• anticipate operational issues before they escalate
  
• integrate text analysis with predictive models to prevent dissatisfaction
  
• personalise services by identifying customer needs and suggesting targeted solutions
  

Investing in Text Intelligence today means building a more responsive and intelligent banking model one where complaints are not merely regulatory obligations but opportunities for improvement, innovation, and stronger long-term customer relationships.

Author: Domenico Rodilosso

L'articolo Managing bank complaints with text intelligence proviene da Revelis.

Current challenges in wildfire management

Despite the growing use of advanced technologies such as satellite and drone-based remote sensing, predictive models, and artificial intelligence, wildfire management still faces several limitations:

• Lack of a comprehensive overview: The use of isolated tools without effective data integration from various sources (satellites, drones, sensors, cameras, weather models) makes it difficult to obtain a complete and real-time picture of the situation.
  
• Data incompatibility and interpretation difficulties: Data formats, resolutions, and time scales vary, making comparisons and interpretation complex and resulting in fragmented and unreliable information.
  
• Information overload and analytical challenges: The massive volume of raw data can overwhelm human operators, making it hard to identify meaningful patterns and extract actionable insights.
  
• Lack of accurate forecasting: The inability to effectively integrate weather and environmental data hampers fire risk prediction and its potential evolution.
  
• Delayed response: The absence of an integrated and automated system can lead to critical delays in intervention.
  
• Lower efficiency and higher costs: A non-integrated approach requires more human resources and results in higher operational costs.
  

PRIMA: the integrated platform for a safer future

To address these challenges, PRIMA (real-time platform for wildfire ingestion, monitoring, and analysis) was developed—an innovative project funded under the Tech4You Program (PNRR). PRIMA aims to build an integrated platform for real-time wildfire monitoring and analysis, providing comprehensive and timely decision support to enable optimal wildfire management.

The system is built on three interconnected software modules:

• Forecast and Propagation Module: This module provides advanced real-time simulation of fire spread, integrating live hydro-meteorological data and predictive models. It relies on a sophisticated 2D reaction-diffusion mathematical model, accounting for factors such as air density, specific heat, temperature, wind speed, topography, turbulent diffusivity, and solid fuel combustion. Simulations incorporate topographic variability, which influences wind behavior and thus fire propagation.
  
• Satellite Monitoring Module: Enables early detection of thermal anomalies via satellite data (e.g., Meteosat), with immediate alerts and continuous fire evolution monitoring. Satellite data offer a broad, continuous view of the territory, identifying active hotspots and estimating burned area size.
  

Post-Fire Analysis and Monitoring Module: Focuses on the rapid processing of multispectral images (satellite, drones, LiDAR) to assess fire severity, monitor natural regeneration, and analyze the structure of affected areas.

How PRIMA transforms wildfire management

PRIMA is implemented as a vertical application of PlugAIn, Revelis’ IoT Intelligence platform, originally developed for manufacturing, smart cities, and multi-utility sectors. PlugAIn is based on a data-driven and event-driven architecture, integrating data from various sources, such as:

• Satellites: Multispectral and thermal images for vegetation analysis and detection of temperature anomalies.
  
• Drones (UAS): Provide close-range, flexible perspectives. They can capture RGB images for photogrammetry (creating point clouds, 3D models, orthomosaics, DTM), thermal images to detect hotspots even through smoke, LiDAR data to map terrain and vegetation, and wind sensors for fire spread prediction.
  
• Ground Sensors: Networks monitoring temperature, humidity, smoke, and particulate matter enable early fire detection.
  
• Cameras: Fixed or mobile, equipped with AI algorithms to automatically detect smoke or flames and generate real-time alerts.
  
• Weather Data: Includes temperature, humidity, wind, precipitation, and lightning—crucial for predictive modeling.
  
• Corine Land Cover (CLC): European land use and land cover data, useful for identifying risk areas and planning prevention strategies.
  

PlugAIn can gather information from any IoT sensor network. Platform administrators can easily manage the list of sensors being monitored. The system stores data in a big data lake, enabling insight generation and analytics via web dashboards for real-time monitoring. The platform offers a range of algorithms to support both descriptive and predictive analyses, with models made interpretable for decision-makers through AI explanation techniques.

To manage and analyze this heterogeneous data, PRIMA uses cutting-edge technologies integrated in PlugAIn, such as Apache Kafka for real-time data ingestion and Elasticsearch for fast data storage and querying. Specialized gateways (Polling and Subscription) enable integration of any data source, converting it into a unified format.

Benefits and impact of the PRIMA project

PRIMA stands out for its numerous benefits and expected impacts:

• Accuracy: The integration of multiple information sources ensures comprehensive analysis and highly accurate results.
  
• Real-time Monitoring: The combination of satellite, drone, and ground sensor data allows for continuous fire monitoring.
  
• Forecasting: Thanks to AI algorithms, it is possible to predict wildfire occurrences up to 9 days in advance.
  
• Damage Assessment: Before-and-after image analysis provides a valuable tool for assessing damage and quantifying burned areas.
  
• Support for Authorities: Notifications and alerts enable rapid and effective response planning, optimizing resource allocation.
  
• Cost Reduction: The platform avoids the need for expensive equipment unless strictly necessary.

A user-centered design for effective decisions

Recognizing that wildfire management is a high-pressure activity, PRIMA has adopted a Human-Centered Design approach based on UX and HCI principles to create intuitive and high-quality interfaces. The goal is to prevent cognitive overload, support constant information exchange, and keep operators in control at all times.

The primary users of the platform are identified as:

• Unified Operational Room (SOUP) operators
  
• Field team leaders (Capi Squadra)
  

For each, specific interfaces were designed:

• SOUP Interface: A desktop application for control room use. It allows for fire coordinate input, map visualization with road and land registry layers, and fire propagation simulation based on meteorological and topographical data.
  
• Field Team Leader Interface: A tablet app for on-site use. It receives simulations from the SOUP, displays essential information, and supports strategic planning of firefighting operations, incorporating terrain topography.
  

The interface design follows usability principles such as system status visibility, consistency and standards (e.g., Google Maps inspiration and use of standardized SI.TA.C symbols from the Italian Fire Department), error prevention, and flexibility. Prototypes are being validated through qualitative and quantitative testing with real users.

Conclusion

PRIMA will have a significant scientific, economic, social, and environmental impact. It will advance wildfire modeling and satellite data processing knowledge, while creating an innovative platform ready for national and international deployment. On a social and environmental level, it will reduce wildfire damage, improve intervention safety, and raise public awareness on risks and best practices.

PRIMA is a concrete example of how integrating advanced technologies, Big Data analytics, and Artificial Intelligence can help tackle one of the most complex environmental challenges of our time—providing real tools for more effective land management and a faster, more efficient response to critical events.

Author: Luigi Granata

L'articolo The prevention and management of wildfires with PRIMA proviene da Revelis.

In this article, we will therefore explore how our AI-based solutions are revolutionizing weather forecasting, offering companies or municipalities powerful tools to make more informed decisions, optimize resources, and mitigate the risks associated with atmospheric events.

The calabrian context

The increased frequency and intensity of extreme weather events, such as floods and landslides, have made the use of AI solutions for weather forecasting essential for monitoring and providing rainfall predictions. These phenomena not only threaten human life but also have a devastating impact on infrastructure, agriculture, and the economy. A tragic example of such an impact occurred on August 20, 2018, when heavy rain caused the collapse of a dam, sweeping away a group of hikers crossing the Raganello Gorge in the province of Cosenza, Calabria. This incident highlighted the urgency of more accurate and timely forecasting systems to prevent similar tragedies and environmental disasters.

Calabria, with its particular orography and the presence of small and vulnerable hydrographic basins, is a region particularly exposed to extreme events such as landslides and flash floods. To address these challenges, it is crucial to develop more sophisticated tools for rainfall monitoring, considering the limitations of existing technologies. Although rain gauges, weather radars, and geostationary satellites provide useful data, they present difficulties related to accuracy and reliability. For example, radars can estimate rainfall intensity in real-time, but their effectiveness is reduced by the presence of physical obstacles and atmospheric variability. Furthermore, rain gauges offer point measurements that may not reflect the spatial distribution of rainfall over large areas.

In this scenario, artificial intelligence is establishing itself as an innovative solution to enhance the accuracy of weather forecasts. Thanks to the use of sophisticated algorithms capable of analyzing large volumes of data from various sources, it is possible to obtain more precise and reliable predictions.

ADELE4RAIN: here’s how ai for weather forecasting works

The ADELE4RAIN project (A DEep LEarning-based framework for RAINfall estimation and forecasting), as an AI solution for weather forecasting, aims to create an integrated and comprehensive platform for monitoring and predicting rainfall by combining modules that leverage advanced technologies.

A case study in Calabria, with its climatic variability and vulnerability to floods, offers the opportunity to test and optimize this innovative system, thereby contributing to a more effective management of hydrogeological risk in the region.

The framework that has been decided to adopt and proposed in [Guarascio et al., 2020], is as follows:

It consists of three macro-components, illustrated in Figure 1, which shows the entire workflow, highlighting the intermediate results obtained from each component.

The Information Retrieval macro-module is dedicated to the extraction and integration of different types of information. Specifically, the main data sources used include:

• Weather Station Networks: directly measure the amount of rain through rain gauges, which record the water that falls in a defined area during a specific period of time.
• Weather Radar Systems: by using radio waves to detect water particles suspended in the atmosphere, radar allows for obtaining a dynamic and high spatial resolution view of ongoing rainfall, overcoming the limitations of the surface station network.

Rainfall data was retrieved through the AllertaCAL platform, developed by the Functional Meteorological, Hydrographic, and Oceanographic Center of the Calabria Region. AllertaCAL collects and publicly exposes data from approximately 260 weather stations scattered throughout the regional territory. The stations are equipped with sensors such as rain gauges, thermometers, hydrometers, barometers, hygrometers, etc. The collected data is made available both in aggregated and point form. The information system provides REST services that allow access to the collected data, as well as the list of weather stations and the sensors present in each of them.

Regarding radar data, the Radar-DPC platform of the Civil Protection Department (DPC) of the Presidency of the Council of Ministers was used. The Platform adopts a very flexible approach and has been designed to manage, even incrementally, geo-referenced products of different natures and semantics. The Radar-DPC platform integrates a stack of REST (Representational State Transfer) services, used both by the back-end of the applications and as APIs (Application Programming Interface) for downloading radar data. These services also allow interaction with third-party applications, facilitating access and processing of the information provided. All services are accessible from this URL.

The data acquisition component was implemented using the PlugAIn platform, developed by Revelis.

The Data Analytics module includes the following three modules:

• Data Preprocessing;
• Data Sampling;
• Model Building.

The Data Preprocessing module offers cleaning methods to handle data issues such as missing values, outliers, and noise, with different strategies for each.

The framework includes a Sampler module that addresses class imbalance through under-sampling, creating a preprocessed dataset for the REM (Rainfall Estimation Model) Builder module. This module utilizes machine learning, deep learning, and deep ensemble learning algorithms to develop models capable of identifying complex patterns in rainfall data. In particular, a probabilistic hybrid ensemble classifier (HPEC), based on Random Forest and Stacking, has been proposed, along with deep learning models such as feed-forward neural networks and a deep ensemble learning approach to optimize predictions.

The combination of the data extraction and integration module with the estimation methods ensures a smooth and well-coordinated operation of the framework, creating the conditions for subsequent evaluation and validation processes. In these phases, the results obtained are crucial for refining the solution, optimizing its performance, and preparing it for real-world application.

ADELE4RAIN: the objectives

Having explored the technical details regarding the platform’s design and the methodologies used to integrate data from heterogeneous sources, it is evident that this solution represents a significant advancement in rainfall monitoring and forecasting. The combination of advanced machine learning and deep learning algorithms, real-time data processing, and the use of innovative technologies allows for increasingly accurate and timely predictions.

However, the benefits of this platform extend far beyond the technical aspect, with a direct impact on:

• management of extreme weather events;
• protection of natural resources;
• improvement of community resilience;
• prevention of environmental disasters.

At this point, it is useful to reflect on the potential impacts and future developments that this platform could enable. By providing accurate and timely forecasts, the project aims to allow infrastructure managers and decision-makers to adopt proactive measures to:

• protect infrastructure assets;
• reduce vulnerabilities;
• improve resilience to climate-related risks.

The project aims to improve agricultural management by optimizing resource use. By providing farmers with precise information on rainfall patterns and trends, the intention is to enhance irrigation scheduling, crop management, and water resource strategies, increasing productivity and promoting sustainable agricultural practices.

The project has the objective of enhancing early warning systems, evacuation plans, and emergency responses, with the intent of reducing risks to people, property, and the economy during extreme weather events.

Conclusions

In conclusion, the creation of an advanced platform for estimating and forecasting rainfall is fundamental for addressing the challenges arising from climate change and extreme weather events for the prevention of environmental disasters. Looking to the future, the adoption of solutions like this is therefore essential for building a more resilient society, ready to face climate-related challenges, and promoting sustainable and proactive adaptation to constantly evolving weather conditions.

Author Emanuela Tarantino

L'articolo AI for weather forecasting: the ADELE4RAIN project proviene da Revelis.

Today, thanks to artificial intelligence, it is possible to implement advanced corporate solutions capable of monitoring various parameters related to employee well-being in real-time.

Revelis, as a company that offers corporate solutions created using artificial intelligence, has always provided businesses with targeted solutions to achieve various goals, including employee well-being, thanks to the monitoring made possible by AI solutions.

In this article, we will therefore discuss InCoP, a platform for Industry 5.0, used for this purpose and validated in the context of Target S.p.A., a company that produces baked goods. We will analyze the reference context to understand the needs of the recipient company and the objectives to be achieved, up to seeing the planned activities and the functioning of the AI solution implemented by Revelis.

InCoP: the corporate solution for manufacturing companies

Within the scope of the public call “Future Artificial Intelligence Research – FAIR,” Revelis was selected for its project proposal “Industry 5.0 Collaborative Platform – InCoP.”

This proposal represents a corporate solution for manufacturing companies. The project involves the creation of a platform (which utilizes the framework described in WP9.6 of the call) for Industry 5.0, validated in a company that produces baked goods.

For this solution, methodologies based on AI, Internet of Everything (IoE), and intelligent industrial automation will be used.

The project proposes several aspects of innovation:

• IoE paradigm to address people, processes, and data;
• Module for intelligent task execution, supporting workers and their safety;
• Mass customization module, for the benefit of the customer and for the reduction of resource waste.

InCoP and Target S.p.A.: the reference context

An Industry 5.0 platform will be created as a customized corporate solution for Target S.p.A., a company operating in the Catanzaro area and specializing in the production of baked goods such as bread, pizza, focaccia, and snacks. In recent years, the company has invested heavily in innovative machinery to make its production process efficient and automated.

Revelis has various experiences in Industry 4.0 solutions, both with Target S.p.A. and with other important industrial clients, and therefore possesses the right skills to enable Target S.p.A.’s transition to Industry 5.0.

The state-of-the-art techniques used in the project fall into the following areas:

• Internet of Everything (IoE), an evolution of IoT, with the aim of connecting people, processes, information, and objects in a single and interoperable ecosystem. The complexity and heterogeneity of resources in the IoE require architectures capable of managing interoperability at a technical, syntactic, semantic, and structural level. Data repositories, based on NoSQL technologies, store heterogeneous data in a common and accessible format. These architectures also support the analysis and extraction of useful information for decision-making.
• Machine Learning and Deep Learning. Through descriptive and predictive algorithms for mass customization, Target will receive decision support in the creation of products and services tailored to each customer.
• Automated Reasoning. The use of logic programs and Answer Set Programming techniques will support Target S.p.A.’s personnel in the execution of tasks, through personalized indications and suggestions to optimize the performance of their duties, reducing working times, improving operational efficiency, and decreasing production costs as well as the risk of workplace accidents.

Project objectives

The general objectives of the project, in which customized corporate solutions are applied, are articulated within a broader vision that aims to achieve several goals, including:

• Develop an innovative platform for Industry 5.0: The aim is to create a versatile and scalable technological platform capable of supporting the implementation of Industry 5.0 principles in various industrial sectors. The platform will be designed to be modular and adaptable, allowing the integration of new functionalities and technologies based on the specific needs of each application context.
• Validate the platform in a real-world context: The baked goods manufacturing company will serve as an initial testing ground to assess the effectiveness and applicability of the platform in a concrete industrial environment. However, the long-term goal is to extend validation to other sectors and production contexts, demonstrating the platform’s versatility and potential impact in different areas.
• Improve the efficiency and flexibility of the production process: Through detailed analysis of the production process and manual activities, areas for improvement will be identified, and corporate solutions based on Machine Learning (ML), Deep Learning (DL), and Reasoning techniques will be designed to optimize operations and increase the ability to adapt to market demands. The goal is to create a more intelligent production system capable of learning from data and continuously improving its performance.
• Enable mass customization of products: The platform will be equipped with a specific module to enable the large-scale production of personalized products, responding to individual customer requests and creating new business opportunities. This objective aims to revolutionize the way products are conceived and manufactured, moving from a standardized production model to one based on personalization and customer satisfaction.
• Demonstrate the potential impact of Industry 5.0: Through the definition of metrics and the evaluation of results obtained in the pilot company and other application contexts, the tangible benefits brought by the platform in terms of efficiency, quality, flexibility, sustainability, and customer satisfaction will be quantified.
• Create an innovation ecosystem: The project aims to foster collaboration between companies, technology providers, research institutions, and universities, creating an open and dynamic innovation ecosystem. This ecosystem will allow for the sharing of knowledge, experiences, and resources, accelerating the development and adoption of innovative corporate solutions for Industry 5.0.

Planned activities

The activities planned in the project are grouped into three main phases:

Feasibility Analysis. This includes the following activities:

• Analysis of the production process and manual activities;
• Verification of the applicability of the WP9.6 framework with respect to the analyzed production process;

Use Case Specification. This includes the following activities:

• Definition of expected results, metrics to evaluate them, and quantification of potential impact;
• Design of ML techniques for mass personalization, and (Stream) Reasoning techniques to support personnel in task execution;
• Platform design for the use case;

Platform Development and Validation. This includes the following activities:

• Module for intelligent task execution, based on the analysis of employee behavior;
• Module for mass customization of products, to ensure organizational efficiency.

Operation of the application solution

Regarding the Industry 5.0 theme, a Software Component for worker support will be defined. This component will receive information from the IoE (Internet of Everything) system to monitor:

• the psychophysical parameters of workers;
• environmental parameters;
• production-related parameters.

It will process all this information to decide which tasks to assign to workers, with the dual objective of optimizing production without overloading workers. The reference scheme of what is to be achieved is shown in the following figure.

Specifically, the IoE system will consist of:

• Smartwatches, worn by workers, which measure: stress level; blood pressure; heart rate; blood oxygen level; and body temperature;
• Environmental temperature and humidity sensors installed in the various departments.

This data will be stored in an Elasticsearch database. Data from a suitable algorithm for predicting worker stress will also be fed into Elasticsearch. This data will then be used by appropriate Reasoning algorithms that will calculate the optimal scheduling of resources.

To calculate the optimal scheduling, the following will be taken into account: the bakery operates 24 hours a day from Monday to Friday, and workers cover shifts divided into morning, afternoon, and night. Workers must adhere to a weekly shift rotation to ensure that: those who work at night for one week, work in the morning the following week, and then in the afternoon.

The departments (in which tasks are performed) of the bakery are as follows: ovens, production, and packaging. The following diagram shows the production structure of Target S.p.a.

Regarding the theme of mass customization, it will be taken into account that the bakery produces different types of products, and every day a production order is created that indicates to the workers which types of products to produce and in what quantities.

The historical analysis of data relating to production orders will allow for the definition of ML algorithms that will support Target S.p.a. in the creation of new products personalized for each individual customer.

Conclusions

Today, in an increasingly competitive and constantly evolving world of work, employee well-being is no longer an option: it is a strategic priority. Companies that invest in healthy and positive work environments achieve better results in terms of productivity, engagement, and staff retention. For this reason, relying on companies like Revelis, specialized in corporate solutions created using artificial intelligence that allow for the collection of real-time data and the analysis of various data points to intervene promptly with targeted actions, is an increasingly essential choice for businesses aiming for long-term success.

Author Massimiliano Ruffolo

L'articolo Corporate solutions for employee well-being: InCop proviene da Revelis.

This article will explore the potential of this system as an artificial intelligence solution for energy management, taking the Ei-TWIN project as an example and its practical applications in two distinct contexts: an industrial plant of a pharmaceutical SME and an energy community.

Why an energy management system is essential

The global energy crisis, with rising gas and electricity prices, highlights the need for more efficient resource management. Italy, like many other European countries, heavily depends on gas imports, making energy efficiency a national priority.

In this context, the Ei-TWIN energy management system project aims to develop a platform for applying a Digital Twin in the energy sector. A Digital Twin is a virtual representation of an object or system, linked to it throughout its lifecycle. This technology enables performance monitoring, consumption forecasting, and energy use optimization, helping to reduce waste and improve overall efficiency.

The Ei-TWIN Project: an innovative approach to energy management systems

The Ei-TWIN project consists of 12 development objectives, divided between industrial research and experimental development, aiming to create a software platform for applying an Energy Digital Twin in two use-case scenarios:

• An industrial plant of a pharmaceutical SME: This use case focuses on optimizing energy efficiency in a production environment by analyzing consumption and processes to identify areas for improvement.
• An energy community: In this scenario, the goal is to optimally manage energy flows between buildings, maximizing self-consumption and reducing exchanges with the electricity grid.

The project aims to overcome the limitations of traditional approaches by integrating Big Data processing and AI techniques to provide advanced functionalities such as process optimization and predictive maintenance.

Key components of the Ei-TWIN Platform

The Ei-TWIN platform is composed of several layers, each with a specific function:

• Field Layer: Responsible for collecting data from IoT sensors and building management systems (BMS), using protocols such as Modbus and BACNet.
• Ingestion Layer: Acquires data from the field layer and sends it to the Data Lake, performing an initial format validation.
• Data Lake: The core of the platform, where raw data is stored and structured for analysis. This “schema-on-read” architecture allows flexible data querying. Data is acquired in real-time using Apache Kafka and stored in Elasticsearch.
• Analysis Layer: Utilizes AI and Machine Learning techniques to analyze data, provide forecasts, optimize processes, and detect anomalies.
• Presentation Layer: Provides dashboards and reporting tools to visualize analysis results and monitor the system’s status.

The Ei-TWIN data model

The Ei-TWIN project employs a flexible and comprehensive data model that includes data from various internal and external sources:

• Internal sources: Data from machinery (on/off status, usage time), energy data (active/reactive energy, active power, voltage, current), and production data (production phases, quantities produced).
• External sources: Meteorological data (temperature, humidity, solar radiation) that influence energy consumption.

The data model is structured in two main formats:

• JSON Model: Used for time-series data, with a hierarchical structure including timestamps, telemetry, diagnostics, and metadata.
• Relational Model: Used for structured information such as production batches and process phases, organized into tables with relationships.

Technologies and tools used

The Ei-TWIN project extensively leverages the technologies and architecture of the PlugAIn platform for Big Data processing. The Data Lake is based on a reduced version of the PlugAIn platform, as shown in the following diagram.

Key technologies include:

• Kubernetes: Used to create the private cloud hosting the Ei-TWIN Data Lake.
• Apache Kafka: Used for message exchange between various components and data ingestion.
• Elasticsearch: Used for storing sensor data.
• Spring Boot:: Used for developing connectors to data sources.

Use cases: practical applications of the digital twin

The Ei-TWIN energy management system project involves implementing the platform in two specific use cases to test and validate its functionalities:

• Industrial Plant: The goal is to optimize energy efficiency in a pharmaceutical company by analyzing machine and production process consumption. The Digital Twin will help identify the most energy-intensive phases, forecast consumption, and optimize energy usage.
• Energy Community: In this use case, the platform will manage energy flows between buildings in an energy community, maximizing self-consumption and reducing exchanges with the electricity grid. The Digital Twin will monitor consumption, predict renewable energy production, and optimize distribution among users.

Advantages and benefits of the Ei-TWIN Project

The Ei-TWIN energy management system project aims to provide several advantages and benefits:

• Energy Efficiency: Reduction of energy consumption and waste through process optimization and predictive models.
• Sustainability: Lower environmental impact by reducing greenhouse gas emissions.
• Cost Optimization: Lower energy costs due to more efficient resource use.
• Predictive Maintenance: Anticipating failures and malfunctions for more effective and timely maintenance.
• Flexibility: Adapting energy management to different operational conditions and specific user needs.
• Interoperability: Integration with various information sources and management systems.

Conclusion

In the field of energy management systems, the Ei-TWIN project is a concrete example of how Digital Twin technology and Artificial Intelligence can be applied to improve energy efficiency and promote sustainability.

The platform developed in the project offers an innovative and flexible energy management solution in various contexts, helping to reduce consumption, optimize costs, and enhance business competitiveness.

The adoption of an energy management system like the one proposed by Ei-TWIN is essential to addressing present challenges and building a more sustainable future.

Authors: Massimiliano Ruffolo and Simone Vizza

L'articolo Energy management system: examples of potential in a use sase proviene da Revelis.

In its classical form, Process Mining algorithms are used to build a process model from the extracted data. This model visualizes the flow of activities, representing the sequences of events and the various paths followed within the process. The output can be a graphical map that displays all possible process variants, including exceptions or deviations from standard flows. Models in this category are referred to as imperative models and, in the case of well-defined and standardized processes, they can accurately and comprehensively describe the analyzed processes.

However, for processes lacking these characteristics, imperative models often prove unsuitable because they risk losing generality by excluding valid variants or becoming unreadable and unusable—the so-called “spaghetti models.”

In Figure 2, a representation of a real process is shown as the green area on the left. An imperative model can be represented as the blue ellipse in the center of the figure; it may capture a portion of the real process while potentially including variants and paths that are not present in reality. A declarative model, on the other hand, is based on rules and constraints that are used not to define all possible process paths but to delimit the region where the process traces can be found. This is graphically represented on the right side of the figure.

Revelis believes in research and development activities, and recently developed explainable and aware process intelligence techniques. To this end, recent advancements in artificial intelligence and declarative languages and techniques have been integrated into Process Mining to represent flexible processes.

The work carried out by Revelis aimed to implement a demonstrator prototype of these techniques by applying them to an industrial case study.

Declarative Process Mining

Declarative Process Mining represents an innovative evolution in process analysis. Unlike traditional Process Mining, which mainly relies on extracting process models from event logs, declarative Process Mining introduces a higher level of abstraction. Declare is a declarative language for process modeling that consists of a set of templates expressing the temporal properties of process execution traces.

The Declare models (or templates) can be classified into four distinct categories, each addressing different aspects of process behavior:

• Existence Models: Specify the necessity or prohibition of executing a particular activity, potentially with constraints on the number of occurrences.
• Choice Models: Focus on the concept of execution choices, modeling scenarios where there is an option regarding which activities can be executed.
• Relation Models: Establish dependencies between activities, specifying that the execution of one activity requires the execution of another, often based on specific conditions or requirements.
• Negation Models: Model mutual exclusivity or prohibitive conditions in activity execution.

For example, the Existence(A) model indicates that activity A must appear in a process log. The Exclusive Choice(A, B) model specifies that activities A and B can appear in a trace but not simultaneously. For the ChainResponse(A, B) model, every time A appears in a process instance, B is the immediately following activity.

Answer Set Programming

Answer Set Programming (ASP) is a form of non-monotonic logic programming that focuses on solving problems by representing them as sets of logical rules. Instead of providing a single result, ASP provides a set of models (or answer sets) that satisfy all the rules of the program. This makes it particularly suitable for tackling complex and uncertain problems, where multiple valid solutions may exist.

Encoding Declare Models in ASP

Within the project, an automatic translation technique for Declare models and process logs into ASP logic programs was developed. This allows the use of ASP for tasks such as Conformance Checking and Query Checking. In the first case, the goal is to verify the adherence of an event log to a set of Declare templates. In the second case, the aim is to identify all possible solutions to a query by specifying a Declare template, a process activity, and one or more variables.

Example:
ChainResponse(‘A’, X) with minimum support 0.8
Let A be a process activity and X a variable. The result of the query is given by all the activities that immediately follow activity A, making the model true in at least 80% of the cases.

Case Study

The techniques presented so far were applied using data on the management of applications submitted for obtaining Development Contracts (CdS) with an important national development agency. The process under consideration is divided into two independent macro-phases (Instruction and Technical Verification), which were analyzed separately. Using the specifically implemented prototype, the first task was to derive a Declare model for the two phases. An excerpt of the model for the Technical Verification phase is shown here.

Each line represents a rule that is valid for the log traces analyzed by the algorithm. For example, lines 0 and 6 indicate the existence of a single instance of the activity “VT merit in progress.” Using this model, properly revised, it is possible to perform Conformance Checking, which determines whether process instances violate the specified constraints. This technique was applied to a set of traces provided by the client, which was not used for creating the model. Among the analyzed instances, the following two reported constraint violations, as shown in the table below:

By checking the violated constraints in the traces, we can see that:

• In trace 0, the violations of constraints 15 and 16 are due to the absence of at least one of the activities “VT merit in progress” and “VT Merit to be assigned.”
• In trace 1, constraints 23 and 24 are violated, indicating that the activity “VT merit in progress” is not preceded by the activity “VT Merit to be assigned.”

In the Declare model shown earlier, it can be seen that the activity “Merit to be Assigned” is marked as init in row 14 of the Technical Verification flow table. To check which activities are performed when it is present in the log, the Query Checking technique can be used by querying the prototype with the Precedence template, with the first parameter as “Merit to be Assigned” and the second parameter as the variable X. The results are:

• X: VT merit completed – support: 1
• X: VT merit in progress – support: 1
• X: VT Merit to be assigned – support: 1

This indicates that the three activities mentioned above are performed only if “Merit to be Assigned” is executed.

To check for alternative paths to the one starting with “Merit to be Assigned,” an “Exclusive Choice” query can be used, with the first parameter as “Merit to be Assigned” and the second as variable X. The results are:

• X: VT merit completed – support: 0.7272727272727273
• X: VT Fast Track to be assigned – support: 0.7727272727272727
• X: VT Fast Track in progress – support: 0.7727272727272727
• X: VT fast Track completed – support: 0.7272727272727273

As can be easily seen, there exists an alternative Technical Evaluation branch for the instances, called “Fast Track.”

Conclusions

The techniques and algorithms described here enable a new approach to Process Mining, offering a more flexible analysis. In particular, it is possible to describe the properties of a process using constraints that are either predefined or automatically discovered. These constraints can be used to identify anomalies in process execution through Conformance Checking analysis. Furthermore, the characteristics of the process can be explored by posing queries using Query Checking techniques.

Author: Luigi Granata

L'articolo Process Mining: what it is and the benefits in a real case proviene da Revelis.

Revelis’s approach to operational risk controls

The Regtech solution by Revelis adopts an approach to operational risk monitoring that combines:

• A standardized methodology for data analysis.
• A technological platform integrating a series of functionalities aligned with the technological innovation trends described in the aforementioned article.

Data analysis methodology: CRISP-DM

The delivery of data analysis services and the training of Machine Learning models is carried out using the Cross-Industry Standard Process for Data Mining (CRISP-DM). This is a standard method for the iterative and interactive execution of data mining and analysis processes, structured into three phases and six activities, as illustrated in the following figure.

Phase 1

The objective of the preliminary assessment phase is to support the client in estimating, evaluating, and designing the data analysis solution. This phase includes the following activities:

Business Understanding

This step focuses on analyzing the client’s application objectives based on the following tasks:

1. Analysis of the “as-is” scenario: Reviewing the current state of the client’s systems and processes.
2. Definition of expected results: Establishing the desired outcomes as defined by the client’s decision-makers.
3. Definition of evaluation criteria: Setting the standards and metrics used to assess the results of the data analysis process.

Data understanding

This step aims to understand the data and is essential to avoid unforeseen issues during the Data Preparation phase. The tasks performed in this activity include:

1. Identification of internal and external data sources: Determining the sources of data relevant to the analysis.
2. Identification of significant entities and attributes: Recognizing the key data elements necessary for the analysis.
3. Descriptive data analysis: Conducting an initial evaluation of the available data using basic statistical functions, graphs, and aggregations.
4. Evaluation of data quality: Identifying potential issues, such as formatting errors or misrepresentation of information.

Phase 2

Phase 2 involves the Data Acquisition and Preparation activity, aimed at connecting to the information sources identified in Phase 1, as well as preparing and cleaning the acquired data. The following tasks are performed during this phase:

• Data Filtering. This task focuses on selecting data relevant to achieving the client’s objectives. It involves “filtering” data and attributes using sampling techniques and/or removing “dirty” data.
• Data Cleaning. Statistical techniques are applied in this task to: Reconstruct missing data, Handle data errors, Fix measurement inaccuracies, Address sparse data issues.
• Creation of Synthetic Data. Raw data is often insufficient to effectively carry out the intended analysis. In such cases, new “plausible” data can be generated from the existing data by employing techniques to derive attributes or generate records.
• Data Integration. Data from different sources referring to the same entities can be integrated through: Merge: Enriching an entity by increasing the number of attributes available. Union: Expanding the number of entities available for analysis.

Phase 3

Phase 3 involves two distinct macro-steps:

1. Formulation and Implementation of the Analysis Model

This step aims to define (using inductive statistical techniques) and evaluate (using standard metrics) one or more data mining models. The activities include:

• Development of Connectors: Connectors are software interfaces that enable the import of data from external systems to be processed by data mining algorithms or the export of results to external systems or platforms.
• Modeling: This involves applying algorithms to an appropriate subset of the available data, called the training set. The training set undergoes preprocessing to ensure it is ready for statistical algorithms. The output is a decision-support model, which can serve various purposes:
  • Classification
  • Regression
  • Segmentation
  • Summarization
  • Link or association analysis
  • Sequence analysis
    The algorithms are based on machine learning and deep learning techniques.
• Evaluation:
  This activity measures the model’s performance using appropriate metrics from scientific literature. This allows for an objective assessment of the quality of the induced models and verifies whether the defined business objectives are met. Performance evaluation is conducted on a separate subset of the available data, known as the test set, distinct from the training set used during model training.

2. Execution of the Analysis Solution. This step involves the Deployment activity, which is the implementation of an executable version of the induced model. This version can be integrated into the client’s application infrastructure for use.

Technological Framework: The Moneying™ Platform

Moneying™ is the Regtech solution developed by Revelis, consisting of a platform that leverages Artificial Intelligence techniques for data analysis to support operational controls, such as financial transactions and client communications.

The combination of Machine Learning and Deep Learning techniques optimizes verification activities in control processes. The use of Text Analytics techniques enables automated processing of natural language texts, facilitating document management automation for both internal documents and those received from external sources or clients.

Key Features of the Moneying™ Platform, the Moneying™ Regtech solution offers the following key features:

• Scalability:
  Moneying™ is a highly scalable platform integrating technologies for Big Data analytics. It can be deployed both on-premises and as-a-service via cloud delivery.
• Big Data Analytics:
  Moneying™ provides access to over 100 algorithms for data processing and behavior prediction. These algorithms can be combined in various ways to define flexible and highly customizable processing workflows.
• Security:
  Moneying™ includes a user authentication and profiling system that controls access to data analysis and information, ensuring high-security standards in the client’s operational contexts.
• Web Interface:
  The analysis of suspicious cases is conducted through a powerful and user-friendly web interface. Users can generate customized reports and access informational dashboards related to the activities performed.

The functional architecture of the Moneying™ platform, the Regtech solution, is illustrated in the following figure.

On the left side (Area 1), the platform represents the range of possible types of information sources that may be relevant to the analysis processes. Specifically, the platform supports:

• Data from internal and external information sources: Data acquisition occurs through dedicated connectors to these sources.
• Various types of data:
  • Structured/semi-structured data: Databases, CSV files, XML files.
  • Unstructured data: Text documents.
  • Process-related data: Logs.
  • Multimedia data: Videos, images, audio files.
  • Geospatial data.
  • Streaming data: From continuous sources or sensors.
  • Web and social media data.
• Data types related to customers under analysis:
  • Subjective data: Directly linked to the individual being analyzed. This includes demographic information, profession, and residence.
  • Objective data: Related to the individual’s operational activities concerning their relationship with the bank. Examples include financial transactions, information on suspicious activity reports, and any other type of alerts.

The listed information sources are subjected to storage and/or analysis processes within the Cloud Infrastructure (Area 2) portion of the architecture. This area is logically divided into two distinct sub-areas:

1. Big Data Storage: For data storage.
2. AI/Big Data Analytics: For data analysis.

The storage and analysis processes can be implemented entirely separately, either in the cloud or on-premises. Across both sub-areas, various services for security, monitoring, auditing, backup, and disaster recovery are integrated.

The AI/Big Data Analytics section is logically divided into two parts: Core Capabilities and AI Capabilities. The first section includes distributed computing mechanisms, access to file systems (local or distributed), in-memory data access, workload management, and access to both SQL and NoSQL data sources. The second part provides a range of functionalities aligned with the CRISP-DM methodology, supporting the phases of data acquisition, information preparation, model training, and results visualization.

Regarding models, the platform notably offers both Machine Learning and Deep Learning algorithms, as well as mechanisms to enable automated reasoning.

This approach enables the creation of AI solutions that combine knowledge derived from data (sub-symbolic approaches, such as Machine Learning and Deep Learning) with domain expert knowledge (symbolic approaches, such as automated reasoning).

This feature of the platform has two key advantages:

1. It allows for the automation of decision-making processes even in cases where available datasets are too small for training Machine/Deep Learning models, as expert logic can be automated through expert systems.
2. It addresses challenges associated with sub-symbolic approaches, such as GPT model “hallucinations,” by enabling the implementation of control systems based on automated reasoning to correct such issues.

Finally, on the right side of the figure (Area 3), the application services related to banking controls are represented. A non-exhaustive list of these services includes:

• KYC (Know Your Customer)
• Enhanced Due Diligence
• Account Opening Controls
• Risk Profile Evaluation (credit, default, investment-related risks, etc.)
• AML (Anti-Money Laundering) via Transaction Monitoring
• Monitoring (authorization and spending limits, exposure limits)
• Profiling (creditworthiness, product suitability, etc.)

Completed Use Cases

AML – Transaction Monitoring, Risk Management, Management of Relations with Investigating Authorities

The Moneying™ Regtech solution applies Machine Learning and Reasoning techniques to analyze financial transactions for AML and Risk Management purposes.

Through pattern recognition and profiling techniques, it is possible to identify fraudulent or illegal behavior and enable the assessment of customer risk profiles (KYC).

The platform offers the following functionalities:

• Risk Profile:
  Moneying™ enables customer profiling based on a “Risk Profile,” calculated using both objective and subjective customer characteristics. All customer information is integrated to provide a “customer file” containing all relevant details for both B2C and B2B customers.
• Anti Money-Laundering:
  Moneying™ provides functionalities for processing Big Data related to financial transactions to detect anomalies in the use of financial instruments by one or more individuals. The platform includes a module for generating Suspicious Activity Reports (SARs) to send to the Financial Intelligence Unit (FIU).
• Risk Management:
  The platform allows for risk evaluation analysis; users can customize data analysis processes to calculate relevant Key Risk Indicators (KRIs).
• Monitoring of B2C and B2B Customers:
  Moneying™ provides a module to manage the sanctioning of B2C or B2B customers who violate contractual agreements in the use of payment instruments.
• Management of Relations with Investigating Authorities:
  Moneying™ optimizes interactions with investigating authorities by speeding up the response process to requests from judicial authorities, law enforcement, and courts.

Automatic Analysis of Customer Communications Related to Complaints and Grievances

Currently, in the banking sector, customer support relies on the management of large volumes of emails, which requires a significant amount of time and resources from human operators. By using Artificial Intelligence and Machine Learning techniques, the process of reading and classifying emails can be automated, significantly reducing the workload of operators, eliminating repetitive tasks, and allowing them to focus on other activities, thereby improving overall efficiency.

The “AIOM” vertical of the Moneying™ platform uses Text Intelligence techniques to automatically read and classify emails sent to the Complaints and Legal Departments of a major international bank. This is done by analyzing the content (body of the email and attachments) to direct the processing of complaints and grievances and, when possible, to automatically respond to the customer.

The Benefits of MoneyingAIOM, the Regtech Solution

The use of the MoneyingAIOM platform as a Regtech solution has led to the following impacts:

• Reduction of Operator Workload: The Text Analytics model can automatically handle the reading and classification of emails, significantly reducing the workload of human operators.
• Increased Efficiency: The automatic classification of emails occurs in a fraction of the time compared to manual classification, allowing for faster responses and a smoother workflow.
• Scalability and Flexibility: The automatic email classification system can be easily adapted to handle large volumes of emails without requiring a proportional increase in human resources. Furthermore, the machine learning model can be adjusted and improved over time to meet new needs and changes in email content.
• Cost Savings: By reducing the reliance on human operators for reading and classifying emails, costs associated with managing large volumes of corporate communications have been reduced by 65%.

The implementation of such a system requires special attention to compliance with regulations, such as the General Data Protection Regulation (GDPR) and privacy protection laws. Emails contain personal or confidential information that must be safeguarded. Once read by the MoneyingAIOM platform, the email texts, attachments examined, and sender information are stored in an encrypted manner.

Enhanced Due Diligence (EDD)

Enhanced Due Diligence (EDD) is a process aimed at mitigating the risk of money laundering and terrorist financing. The manual execution of the EDD process involves the following stages:

• Information Acquisition: Operators manually connect to various internal and external systems to acquire the necessary information for analysis, such as:
  • Verification of the existence of the tax code and VAT number (information publicly available on the Revenue Agency’s website);
  • A list of completed transactions;
  • The risk profile of the subject;
  • Whether the subject is on PEP (Politically Exposed Persons) or anti-terrorism lists.
• Risk Analysis and Assessment: Operators review the acquired information to identify any risk factors and proceed with the overall risk assessment of the client, considering:
  • The intrinsic risk profile (activities performed, PEP status, involved countries);
  • The client’s behavior (suspicious transactions, consistency with provided information);
  • The risk level associated with the monitored product or service.
• Enhanced Verification Measures: For high-risk subjects, a questionnaire is prepared, requesting the subject to justify a series of transactions and potentially provide additional documentation (e.g., company records, bank statements). In this phase, it may also be possible to request information from third-party sources (e.g., databases, competent authorities).
• Updating and Monitoring: Operators send the questionnaires and monitor the response from the analyzed subject. In case of no response, the closure of the existing relationships may be proposed.

The manual execution of EDD requires:

• Qualified Personnel: Employees trained on anti-money laundering procedures and capable of assessing risk.
• Defined Internal Procedures: Detailed documentation of EDD procedures and risk assessment criteria.
• Supervision and Control: A supervisory system to ensure the proper application of procedures.

The automation of EDD operations is performed through an IT system, whose functional architecture is shown in the following figure.

In particular, through appropriate RPA procedures, it is possible to acquire all the information necessary for analysis, which consists of a series of PDF or Excel files that are stored in a dedicated folder.

Using this data, the platform will utilize the Moneying™ platform by Revelis to:

• Automatically extract information from the files, taking into account the type (PDF, CSV, etc.) and structure of the files;
• Implement the decision-making logic related to EDD, starting from a series of “Golden rules” defined by the operators and represented using appropriate formalism in Answer Set Programming.

The process executed through Moneying™ produces, for each position to be analyzed, an outcome indicating whether the subject should or should not be subjected to the EDD questionnaire, and, in some cases, the structure of the questionnaire to be administered.

Other possible applications of AI in the field of controls to safeguard operational risks.

Automating KYC with Artificial Intelligence

Artificial Intelligence (AI) offers tremendous potential to automate the Know Your Customer (KYC) process in the banking and financial sector. Below are some applications that use AI techniques to improve the efficiency and effectiveness of these procedures:

• Facial Recognition and Biometrics. Using computer vision techniques, it is possible to analyze customers’ faces and fingerprints to verify their identity in real-time, using a streaming video. This eliminates the need for paper documents and allows for remote verification.
• Natural Language Processing (NLP) Analysis. As already analyzed in the context of customer communication analysis and AVR, Text Analytics techniques can be used to analyze information provided by customers in KYC and AML forms to identify potential risks. Specifically, the use of Natural Language Processing allows for extracting data from documents such as invoices and bank transfers to identify any suspicious transactions.

Automating Account Opening Requirement Checks with AI

Artificial Intelligence (AI) can be used to automate account opening requirement checks in several ways:

1. Risk Analysis. Expert systems (symbolic approaches) can be used to apply evaluation rules to the information provided by customers in account opening request forms, identifying potential risks. Symbolic approaches can be combined with sub-symbolic approaches related to Machine Learning and Neural Networks to profile customers based on the risk of fraud, terrorism financing, and other financial risks.
2. Identity Verification. As previously described in the KYC case, AI can be used to verify customers’ identities through facial recognition, document analysis, and biometric verification, preventing fraud and identity theft.
3. Background Checks. AI can be used to check customers’ backgrounds to identify any criminal records or financial sanctions. This helps protect the bank from reputational and legal risks.

Author: Francesco Cupello

L'articolo Revelis Moneying’s Regtech solution and use cases proviene da Revelis.