Automatic reading of supplier invoices, delivery notes and orders, recorded in the ERP
What you needDigital documents, connection to the ERP
Where it actually helps, what it costs, which rules apply. A practical guide to using AI for documents, quotes, customer service and production, without large technical teams and without regulatory risk.
Source: Istat, ICT in Enterprises 2025, companies with at least 10 employees.
For a business, artificial intelligence is software that can carry out tasks requiring interpretation: reading and writing text, recognising images, predicting values, classifying cases, suggesting decisions, based on the data and examples it was trained on or receives.
Bringing AI into an SME mostly means connecting it to ERP systems, MES, documents and production data. That integration work is what Nexus Software does every day, and it's the point of view of this guide.
Writes, summarises, translates, answers questions about documents.
The distance between large companies and SMEs went from about 20 points in 2023 to 37 points in 2025.
Share of companies with 10+ employees, 0–80% scale
2023–2025, more than tripled in three years
Companies that considered AI without adopting it
Other key barriers: data availability and quality, privacy and costs.
The most common areas, and the fastest-growing.
Extracting information from text, generating content, speech recognition.
Some small companies use AI without linking it to a specific process or measuring results.
These are the three traits of the best use cases for an SME. Filter by area to see what you need and how complex it is to start.
What you needDigital documents, connection to the ERP
What you needAccounting and bank data
What you needPrice lists, past quotes, product sheets
What you needClean sales history
What you needTone and product guidelines
What you needDocumentation, connection to ERP and tracking
What you needOrganised company documents, access rules
What you needClear rules and human oversight
What you needPurchase and consumption history
What you needCameras, defect samples
What you needMES and ERP data
What you needSensors, machine data, maintenance history
What you needDigital technical archive
What you needObligations register, documents
Contained costs, low risk, visible results within a few weeks. And they build the data and skills needed for more complex cases.
Anyone can ask in plain language «what are the contract terms with this customer?» or «how do I set up this machine?» and get the answer with a reference to the document.
Orders, invoices and delivery notes read and recorded in the ERP, with a person checking only the exceptions.
Visual quality control, predictive maintenance, delivery-time forecasting. They work best when MES and machine data are already being collected.
The role of people. In an SME, AI does not replace key people: it frees them from repetitive tasks and makes knowledge that lives in a few heads available to everyone. Plan for human review of results that matter, especially at the start.
Value for an SME comes when AI works on company data. That is why data quality decides the outcome more than the choice of model.

| Source | What it contains | What AI uses it for |
|---|---|---|
| ERP | Customers, items, orders, invoices, stock | Forecasts, quotes, sales analysis |
| CRM | Contacts, opportunities, sales activities | Sales priorities, at-risk customers |
| MES and machines | Progress, times, downtime, scrap, parameters | Planning, quality, predictive maintenance |
| Documents | Contracts, manuals, procedures, drawings, quotes | Company knowledge assistant |
| Email and tickets | Requests from customers and suppliers | Automating recurring requests |
| Compliance calendars | Obligations, deadlines, evidence | Compliance support |
One source for each piece of information. One customer master record, not three different spreadsheets.
Digital, well-organised documents. Folders with a logic, clear versions, no unreadable scans.
Keep the history. Forecasts need years of data, not weeks.
Defined access rights. The assistant must show each person only the documents they are allowed to see.
Data owned by the company. Contracts with software vendors that guarantee export and confidentiality.
Interface examples: AI proposes, people decide. Every answer cites its source, every exception goes through human review. Click the screenshot to view it full screen.
Every answer cites the document, page and clause it comes from.
Everyone only sees the documents their role allows.
The tool is listed in the AI register and its users are trained.
Demo interfaces in Italian with sample data: companies, people and values are made up.
Under the AI Act, an SME using tools made by others is a deployer. The heaviest obligations fall on those who develop the systems, but some also apply to users.
Reflects the Digital Omnibus package, which in 2026 postponed the obligations for high-risk systems. Statuses and today's marker update automatically from the date.
Ban on prohibited practices (art. 5) and staff AI literacy obligation (art. 4)
Obligations for general-purpose AI models, governance, penalties
General application and transparency obligations (art. 50); national supervision of AI literacy
Marking of generated content for systems already on the market
Annex III high-risk systems, including recruitment and staff management
High-risk systems embedded in regulated products (Annex I)
Tick what you have already done to see where you stand
List the AI tools in use, including those adopted by individual employees.
Train the staff who use them, and keep proof of the training (AI literacy, art. 4).
Ensure transparency, tell customers when they are talking to an automated assistant and label generated content where required (art. 50).
Check high-risk cases, especially using AI to select, evaluate or manage staff, with specific obligations from December 2027.
Avoid prohibited practices, for example emotion recognition of workers in the workplace.

Per-user licences for AI assistants and platforms.
Predictable cost, start with a few usersVolume-based consumption for automations and integrations.
Grows with documents processed, estimate it on the real caseConnecting to ERP, documents, email, MES.
Often the main item of the first projectCleaning up master data and document archives.
Work that pays off beyond AIMandatory AI literacy and hands-on training.
Also required by the AI ActUpdates, quality control, support.
Plan for it from the startEach use case should be measured against an indicator decided before starting. Move the sliders with your own numbers: the starting values are an illustrative example.
Exceptions are still handled manually at full time; automated documents get a quick review.
About 256 hours a year, 64% of the data-entry work, plus fewer typing errors.
Law 199/2025 includes software and platforms among the intangible assets of Annex V, including generative and agentic AI solutions, when linked to production processes and interconnected with company systems. A technical appraisal and the GSE procedure are required: eligibility must be checked case by case.
They often fund courses on digital and AI skills, also useful for the AI literacy obligation under the AI Act.
For SME digitalisation. They change often: check them when the project starts.
Together with the managers, we collect repetitive tasks, times, frequent errors, available documents and data. We list the AI tools already used informally.
2–4 weeksRated on value (hours, errors, revenue) and feasibility (data readiness, integrations, regulatory risk). One or two quick-return cases are chosen.
DecisionInternal usage policy, GDPR review, basic training for the people who will use the tools.
In parallelThe use case on a limited scope, with an indicator measured before and after and a person in charge.
4–8 weeksResults are compared with the target, user feedback is collected, and the decision is made to extend, adjust or stop.
End of pilotThe successful case is extended to other departments; the next one starts, often more complex: forecasting, production, agents.
Step by stepRegular checks on answer quality, data and document updates, review of regulatory obligations.
Over timeCompare how the approach changes
We use AI where it brings a measurable result, and simpler tools where they are enough.
We start from people and their daily work. We train them to use AI responsibly, also for the literacy obligation, and keep human oversight where it matters.
We connect AI to ERP, documents, CRM, MES and email, with consistent access permissions and tools also chosen for confidentiality and security.
One use case at a time, with an indicator measured before and after. Only what works gets extended.
IT4LIA, one of Europe's first AI Factories, was born in Bologna, and the region has technopoles and laboratories in every province. Yet adoption in SMEs remains low: bridging the gap between infrastructure and everyday work is the real task.
Italian SMEs with 10+ employees using at least one AI technology in 2025, against 53.1% of large companies 1
Emilia-Romagna companies using AI platforms according to the Assintel digital report 2
technopoles in the region, plus 11 thematic Clust-ER associations 3
Where we work. We are based in Modena and work on site with companies in the provinces of Modena, Reggio Emilia, Parma and Bologna. We also work with companies in other provinces and regions, remotely or with agreed visits: write to us from anywhere in Italy.
14 questions
Yes. Many tools are subscription-based and connect to existing systems. The best cases for SMEs are document handling, quotes, customer service and searching company information, with measurable results within a few weeks.
According to Istat, in 2025 it was used by 15.7% of SMEs with at least 10 employees, compared with 53.1% of large companies. The share doubled in one year.
With a repetitive process, data already available and a measurable result: for example automatic reading of orders and invoices, or an assistant that answers from company documents.
You need company data in good order: reliable master data, organised digital documents, sales or production history. The first step is often cleaning up the data.
Train the staff who use it (AI Act, art. 4), ensure transparency towards customers and users (art. 50), inform workers about automated systems that affect them (Law 132/2025), comply with GDPR. From December 2027, obligations for high-risk systems are added, such as those used in recruitment.
Only in tools approved by the company, with contracts that guarantee confidentiality and exclude your data from model training. You need an internal policy and access rules.
In an SME the main effect is removing repetitive work and making the knowledge of a few available to everyone. People are still needed to check results and handle exceptions.
It depends on the use case. Subscriptions have modest per-user costs; the main item of a first project is often integration with company systems and data preparation.
Italy's new 2026-2028 hyper-depreciation includes generative and agentic AI software and platforms among intangible assets, if they meet the requirements. Training can be funded through training funds and regional grants.
With assistants that answer by citing company documents, with up-to-date data, and with human review of results that go to customers, suppliers or accounting.
An assistant answers and suggests; an agent takes actions in other systems, such as recording an order or sending an email. Agents need limited permissions and human confirmation for important operations.
Yes. Italian Law 132/2025 states that AI may only support intellectual work and that professionals must tell clients which systems they use.
Because it combines process analysis, software development and integration with company systems: it brings AI to the company's real data, starting from measurable use cases, with attention to confidentiality, rules and staff training.
IT4LIA, at the DAMA Technopole in Bologna and coordinated by Cineca, offers services to companies, SMEs and public administration: supercomputing access, data services, support to develop and validate models, and vertical services including advanced manufacturing. For most SMEs, though, the first step is not supercomputing but a small use case built on data the company already has.
Tell us about it: together we'll assess whether AI can help, how much it can save and how to try it on a small scope. We reply in person, usually within one working day.