AI for SMEs and B2B companies: where to start

In short: a mid-sized or B2B company should start with AI by picking one or two repetitive, measurable processes, such as customer service, order entry, documents or product content. Clean up the data behind them and train the people who will use them, then choose the tools. Every project starts with a goal, a baseline and an owner.

Key takeaways

One in five EU enterprises (19.95%) used AI in 2025, but adoption ranges from 17% of small firms to 55% of large ones (Eurostat).
The main blocker is skills, not technology: lack of expertise is the top reason companies that considered AI decided against it.
Start with internal, repetitive use cases that run on data you already have; customer-facing and sales use cases come next.
Team training is part of the project, not a separate course: the EU AI Act requires companies that use AI systems to support their staff's AI literacy.
The budget is more than the licence: count integration, data preparation, internal hours, training and maintenance too.
For Italian companies with an international footprint, SIMEST has offered a subsidised loan line for AI investment since September 2026, with a 10% grant component.

Why AI adoption matters for SMEs now

Adoption is climbing fast, but unevenly. Eurostat reports that 19.95% of EU enterprises used AI in 2025, with a wide gap by size: 17% of small firms, 30% of mid-sized ones and 55% of large companies. In Italy, Istat puts the figure at 16.4% of companies with at least 10 employees, up from 8.2% a year earlier, with SMEs at 15.7% and large firms at 53.1%.

Using AI is not the same as getting value from it. Deloitte's 2026 "State of AI in the Enterprise" report splits companies into three roughly equal groups: 34% use AI to deeply transform the business (new products and services, reinvented core processes, sometimes a new business model), 30% redesign key processes but keep their business model, and 37% use it at surface level with little change to how work gets done. In the latest update of the same study, based on 3,500 leaders, 81% say AI has sped up decisions and execution, yet only 25% are truly transforming the business.

For an online store or a B2B company, starting now does not make you late. Starting without a method is how you end up "using AI" with no difference in the numbers.

Where do European and Italian companies stand on AI?

Italy sits in the middle of the European pack, behind the other large EU economies. In a report presented to EU finance ministers in September 2026, the International Monetary Fund placed Italy mid-table both for AI-driven productivity growth and for preparedness for technological change, behind Germany, France and Spain, according to Il Sole 24 Ore. The same report estimates that 60% of jobs in Europe are exposed to AI, against a 40% global average, and that around half of the most exposed jobs are ones where AI can support workers and lift their productivity.

Eurostat and Istat data show where adoption gets stuck:

Indicator	Figure	Source
EU enterprises using AI	19.95% in 2025	Eurostat
EU adoption by company size	small 17%, medium 30.36%, large 55.03%	Eurostat
Top reasons for not using AI (EU firms that had considered it)	lack of expertise (70.89%), unclear legal consequences (52.52%), data protection concerns (48.83%)	Eurostat
Italian companies with 10+ employees using AI	16.4% in 2025 (8.2% in 2024)	Istat
Italian SMEs vs large companies	15.7% vs 53.1%	Istat
Business areas where Italian firms use AI most	marketing and sales (33.1%), administrative processes (25.7%), R&D (20.0%)	Istat
Top barriers in Italy (firms that had considered AI and dropped it)	lack of skills (58.6%), unclear legislation (47.3%), missing or poor-quality data (45.2%), data protection (43.2%), high costs (43.0%)	Istat
Italian AI users that name no business purpose for it	33.4% (15.5% in 2024), 83.3% of them with 10-49 employees	Istat

Companies that walk away mostly do not know how to use AI or lack reliable data; the technology is rarely the problem. And a third of Italian firms that say they use AI cannot tie it to any business area, which Istat reads as adoption that is still experimental and loosely organised, mostly among small firms. That is why data, rules and training each get their own section below.

What does adopting AI actually mean for a mid-sized company?

Adopting AI means placing a tool that learns from data or generates content inside an existing process, with a measurable goal and a named owner for the result. A chatbot subscription used by a few colleagues is not adoption. It is individual experimentation.

A few definitions so everyone means the same thing:

Artificial intelligence (AI) is any system that takes input data and produces predictions, content, recommendations or decisions.
Generative AI is AI that creates new text, images or audio from a written instruction (a prompt), like OpenAI's models.
Automation is running repetitive steps by fixed rules, with tools such as n8n or Make. It is often the layer AI sits on.
An AI agent is a system that receives a goal and decides by itself which steps to take and which tools to use (CRM, email, ERP), within set limits.

SME projects that work almost always combine the two: automation moves the data, AI handles the step that used to need someone to read or write. Where to begin is covered in the guide to business process automation.

Which AI use cases pay off for ecommerce and B2B companies?

The first wins come from high-volume work with clear rules and data you already hold. The table lists the most common use cases, with an ecommerce example, a B2B example and the KPI to watch.

Area	Ecommerce example	B2B example	KPI to watch	Effort
Customer service	answers on shipping, returns and sizing from an assistant connected to orders	first reply to technical questions based on datasheets and manuals	tickets resolved without a human, time to first response	medium
Catalogue content	product descriptions, translations and image alt text from Shopify or PIM data	technical sheets and catalogues in several languages	time to publish per product, errors caught in review	low
Documents and back office	reading supplier invoices and delivery notes	extracting codes and quantities from emailed and PDF orders into the ERP	hours of data entry, input errors	medium
Sales and leads	customer segmentation for email campaigns	qualifying inbound leads and drafting quotes	lead response time, conversion rate	medium
Marketing	ad variations for Meta Ads and Google Ads	LinkedIn posts and industry newsletters	cost per lead, production hours	low
Forecasting	demand forecasting and stock reordering	consumption forecasts for repeat customers	stockouts, average inventory	high

Two hypothetical scenarios. An online furniture store with 300 orders a day and two people in customer service gets mostly "where is my delivery" emails: an assistant that reads the order status and answers those, handing everything else to a person, frees time for complex returns and trade customers. A B2B parts distributor receives orders by email in every format: a workflow that pulls codes and quantities from the PDF, checks them against the price list and drafts the order in the ERP cuts data entry and errors, with a person approving before sending.

If your first use case is an assistant on your website, the selection and setup criteria are in the guide to the AI chatbot for ecommerce and B2B websites.

Which sectors and business functions use AI the most?

AI spreads fastest where work is already digital. Eurostat reports that in 2025 it was used by 62.52% of EU enterprises in information and communication and 40.43% in professional, scientific and technical services; in every other sector the share stays below 25%, from 24.82% in real estate to 10.79% in construction. In Italy, Istat puts adoption at 53% among IT and information service companies and 37.3% in telecommunications.

If you are choosing where to start, the breakdown by business function is more useful:

Marketing and sales is where AI is used most, both in Italy (33.1% of AI users, Istat) and across the EU (34.70%, Eurostat).
Administration and process organisation comes next: 25.7% in Italy, 31.05% in the EU.
Logistics is the least covered: only 6.08% of EU enterprises using AI apply it there.
The technologies Italian firms use most are extracting information from text documents (70.8%), generative AI (59.1%) and speech recognition (41.3%).

For a manufacturer, a distributor or an online store, the sector matters less than the process: orders, documents, product content and support exist everywhere, and AI applies to them by the same rules. Machine builders can start from a B2B spare parts portal, companies handling shipping and returns from logistics digitalization, and online sellers from the uses of AI in retail and ecommerce.

In what order should a company introduce AI?

Go from low risk to high impact: internal processes first, where mistakes stay in-house, then the ones that touch customers and sales decisions. This is the sequence I recommend to mid-sized and B2B companies:

Map your processes and shortlist two. List repetitive, high-volume tasks (emails, orders, product content, reports) and estimate the hours they take each month. Keep the two with the most hours and the least risk.
Get the data in order. AI only works well when catalogue, CRM and ERP agree with each other. The first job is often connecting systems and removing duplicates, not choosing a model. In Italy, Istat reports that 48.8% of SMEs use an ERP and 21.1% a CRM, against 85.9% and 56.5% of large companies: for many SMEs the first AI project starts right here.
Set usage rules and approved tools. Decide which tools are allowed, with which data (never personal or confidential data in unapproved tools) and who reviews the output.
Run a 60 to 90 day internal pilot. One use case at a time, with a baseline, a KPI and an owner. Keep human review on until the numbers are stable.
Extend to customer-facing processes. Website assistant, customer service replies, lead qualification: only after the internal pilot has produced measured results.
Connect and automate. Once single use cases work, link them with n8n or Make workflows and consider AI agents for the steps that need more judgment.

To pick the KPIs for each phase and calculate the return, see the guide on KPIs to measure AI ROI. For usage rules and output controls, see the article on scaling AI governance from pilot to production.

Off-the-shelf tools, automation or custom build: which should you choose?

Let the process decide, not the tool of the moment. In an SME it pays to move up a level only when the previous one is no longer enough.

Approach	Examples	When it fits	Limits
AI features already in the software you use	suggestions and generated text inside your CRM, ecommerce platform, office suite or ERP	getting started right away, with no integration, on individual tasks	they only work inside that software and by its rules
General-purpose assistant on a business plan	ChatGPT, Claude, Gemini or Copilot with company accounts and written usage rules	drafts, summaries, translations and document analysis by the team	not connected to your systems: work is copy and paste, and results vary from person to person
AI automation connected to your systems	n8n or Make workflows that read emails and PDFs, call a model through an API and write to the ERP or CRM	repetitive, high-volume processes with clear rules: orders, tickets, product content	someone has to design, monitor and maintain the workflows
Custom build or AI agents	assistant connected to catalogue and orders, agent that drafts quotes from ERP and price lists	when the process is core to the business and simpler use cases already work	higher cost and longer timelines, dependence on whoever builds it, heavier governance

In-house or with a partner? Off-the-shelf tools can be rolled out internally, with one point of contact and clear rules. Once AI has to talk to your ERP, store and CRM, you need someone who knows integrations and data: an internal person if you already have an IT team, or an external partner who designs the workflow and trains the people who will use it. Either way, ask for documented workflows and prompts, with accounts and data held in the company's name, so you never depend on a single person. How AI fits into ERP systems is covered in the guide to enterprise AI and agentic ERP; for drafts, content and assistants, see the guide to generative AI for business.

How much does AI cost an SME, and how do you calculate the return?

There is no useful average price: the cost depends on the use case, the quality of your data and how deeply the project has to integrate with existing systems. Cost is still a real barrier: 43.0% of Italian firms that considered AI and dropped it cite high costs (Istat). Break the budget down into line items before you ask for quotes.

Cost item	What drives it	How to keep it in check
Licences and subscriptions	number of users and plan	licences only for people who use the tool in the chosen process, usage review after 60 days
Model API usage	volumes processed (emails, documents, tickets) and the model used	lighter models for simple tasks, a monthly spending cap
Integration with your systems	how many systems to connect (store, ERP, CRM) and whether they expose APIs	one process at a time, starting with systems that have documented APIs
Data preparation	state of catalogue, master data and history	clean only the data the first use case needs
Internal hours	time spent defining the process, checking output and running the pilot	an owner with dedicated time and short check-ins on a fixed schedule
Training	number of people and roles involved	use-case training, not generic courses for everyone
Maintenance and monitoring	changes to prompts, models and processes over time	a monthly error review and one person collecting issues

Calculate the return by comparing the pilot with the baseline. The monthly benefit is the hours saved multiplied by the fully loaded hourly cost, plus the margin on extra sales attributable to the project, minus recurring costs. The payback period is the upfront investment divided by that benefit. If the benefit is only "freed-up" hours that nobody reuses, the return stays on paper: decide in advance where that time will go, for example trade customers, complex returns or new offers.

Which AI risks and AI Act rules apply to an SME?

For most SME use cases the rules are few, but you need to know them. The AI Act entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions. According to the European Commission, the vast majority of AI systems used in the EU today are minimal or no risk, and the regulation introduces no rules for them.

AI literacy: the Article 4 obligation has applied since 2 February 2025 (see the training section).
Transparency: the transparency rules apply from August 2026. People using a chatbot should know they are talking to a machine, so if you put an assistant on your website, say so clearly.
High risk: some uses are classed as high risk, including CV-sorting software in recruitment and credit scoring systems. After the AI Omnibus agreement, the rules for these areas apply from 2 December 2027, and from 2 August 2028 for AI embedded in regulated products.

Beyond the law, there are operational risks to manage from the pilot onwards:

Wrong answers given confidently. A generative model can invent figures or commercial terms, so anything that reaches a customer goes through human review until the numbers are stable.
Confidential data in unapproved tools. Price lists, quotes and customer data pasted into personal accounts leave the company's control: you need company accounts and a written rule.
Data protection. If the AI processes personal data about customers or employees, review the project with whoever handles GDPR in your company before it goes live.
Security. Every new connection between systems is also a possible way in: see the guide to ecommerce security.
Vendor lock-in. Prompts, workflows and data must stay documented and in company accounts.

If you plan to use AI in recruitment or compliance checks, read the guide to AI in HR and compliance automation first.

Will AI replace employees in an SME?

In the use cases described in this guide, AI takes repetitive work off people's plates while they stay responsible for the result. The IMF report cited above estimates that in around half of the European jobs most exposed to AI, the technology can support workers and lift their productivity. The change still needs managing: if the team fears being replaced, the tool will sit unused. What works in practice:

Involve the people who do the work in choosing the process. They know where time is lost and where AI would get things wrong.
Explain upfront what changes and what does not. Which tasks move to AI, who checks the output, and where the freed-up time will go.
Tell upskilling and reskilling apart. Upskilling updates skills for the same role, such as a customer service agent who supervises the assistant; reskilling prepares someone for a different role, such as moving from data entry to data quality control.
Share the results. Show the team the baseline and the pilot results, mistakes included.

How should you train your team on AI?

Training that works starts from the chosen use cases, not from a generic course: everyone should know what their tool does, where it gets things wrong and when to stop and check. In the EU it is also a legal duty. Article 4 of the AI Act requires providers and deployers of AI systems to support the AI literacy of the staff who use them. After the Digital Omnibus amendments no "sufficient" level is mandated any more, but the obligation stays, and staff overseeing high-risk systems must still be trained.

According to the European Commission, training should cover at least what AI is and which systems the company uses, the company's role (provider or deployer) and the risks of the systems in use, matched to each person's existing knowledge. Testing staff is not required, but handing them the instructions for use is often not enough.

The available formats solve different problems:

Format	When it helps	Limit
Short intensive bootcamp	aligning owners and managers on concepts and practical cases, launching a pilot	without a project to apply it to straight away, the effect fades within weeks
Master's or long programme	building an internal person who coordinates AI projects	high cost in time and money: worth it once several use cases are live
Use-case training	helping the people who use a specific tool every day, such as customer service or sales, use it well	covers only that process

For a company with 20 to 200 employees the most sensible mix is usually a kick-off session for owners and managers, hands-on training on the first use case and one internal point of contact who collects errors and questions. A master's programme comes later, once there is work for a dedicated role. Training and hiring specialist skills can also count as eligible costs, as the next section shows.

Is there funding for bringing AI into an SME?

For companies based in Italy, yes, but requirements and available funds change often and need checking case by case. Since 21 September 2026, SIMEST has run an "AI & Quantum Computing investments" section within its Digital or ecological transition subsidised loan, for companies with an international footprint. The conditions SIMEST lists:

a reserve of 200 million euros;
a 10% grant component, subject to requirements;
a 50% advance;
a 6-year term, extendable to 8 for companies that allocate all the funds to AI and quantum computing spending;
eligible companies: those with an international footprint, especially micro, small and medium enterprises, including non-exporters that belong to an export supply chain.

According to a summary by PNO Innovation, eligible costs include AI software and platforms, integration with ERP, CRM and MES, data warehouses, cybersecurity, AI Act compliance, training and hiring specialist skills, and applicants need export revenue of at least 3% of turnover. Before committing any spending, check the conditions and the remaining funds on the official SIMEST page. Other schemes, from SIMEST's ecommerce line to Chamber of Commerce vouchers, are covered in the guide to Italian digitalization grants.

How to apply it in your business: first steps

This week: list ten repetitive tasks and estimate the monthly hours each one takes. Pick the two with the best balance between hours and risk.
Within 15 days: measure the baseline of the first process (volumes, average time, errors) for at least two weeks, using the same method you will use afterwards.
Within 30 days: write a one-page usage policy (allowed tools, forbidden data, who reviews) and run a training session on the chosen use case for the people who will use it. Before signing any quote, check whether the project qualifies for subsidised funding.
Within 90 days: move the pilot into production with human review and compare the KPIs against the baseline.
On day 90: decide whether to extend, fix or stop, and pick the second use case.

The services I use to design these projects, from automations to AI assistants, are described in the solutions section. More guides on the topic are in the AI for business category of the blog.

Mistakes to avoid

Starting from the tool. Buying licences before choosing the process leads to paid subscriptions nobody uses much.
Skipping the baseline. Without a measured "before" you cannot tell whether the project paid off, or defend it at budget time.
Running on scattered, inconsistent data. An assistant connected to a catalogue with wrong prices or stock levels gives wrong answers, and gives them confidently.
Leaving AI use to individuals. When everyone uses their favourite tool with whatever data they like, privacy and confidentiality risks grow and the company builds nothing it can share.
Training everyone on everything. A generic company-wide course is expensive and quickly forgotten; training tied to use cases and to the people who run them works better.
Ignoring the rules. A chatbot that does not disclose it is AI, or automated CV screening adopted without checking the AI Act, exposes the company to avoidable risks.
Counting only the licence. Integration, data, internal hours and maintenance are part of the cost: leave them out of the budget and the project stalls halfway.

Frequently asked questions (FAQ)

Where should an SME start with artificial intelligence?

With one repetitive, high-volume process that already has data behind it, such as first replies in customer service or keying orders into the ERP. Measure how the process performs today, set one KPI and run a 60 to 90 day pilot with a person reviewing the output. Pick the tools afterwards, based on the process.

How much does it cost to bring AI into a small or mid-sized company?

It depends on the use case more than on the technology. Automating an internal process with a language model costs far less than a demand forecasting project built on years of messy data. Budget for licences or API usage, integration with your existing systems, data preparation, internal hours, training and maintenance, and check whether public funding covers part of it.

Which processes should you automate with AI first?

Internal, repetitive ones: pulling data out of documents, product descriptions, draft replies for customer service, recurring reports. The risk is low because mistakes are caught before they reach a customer, and the return is easy to measure in hours saved. Forecasting and autonomous agents come later.

Does the EU AI Act require companies to train their staff on AI?

Article 4 of the AI Act requires providers and deployers of AI systems to support the AI literacy of the people who use them. Since the Digital Omnibus amendments no specific level of literacy is mandated, but the obligation stays, and staff overseeing high-risk systems must be trained. There is no requirement to test employees, only to provide training that fits their role and the system in use.

Bootcamp, master's programme or in-house training: which is better for AI?

They solve different problems. A bootcamp aligns owners and managers and gets a pilot moving, a master's programme builds an internal AI lead once you run several projects, and use-case training is for the people who use the tool every day. For most SMEs the sensible order is a kick-off session, then hands-on training, and a longer programme only later.

How far along are European and Italian companies with AI?

Eurostat reports that 19.95% of EU enterprises used AI in 2025, from 17% of small firms to 55% of large ones. In Italy, Istat puts adoption at 16.4% of companies with at least 10 employees, double the previous year. An IMF report presented to EU finance ministers in September 2026 places Italy mid-table on AI preparedness, behind Germany, France and Spain.

Is there public funding for AI in SMEs?

In Italy, yes. Since 21 September 2026 SIMEST has had a section of its Digital or ecological transition subsidised loan dedicated to AI investment, with a 200 million euro reserve, a 10% grant component subject to requirements and a 50% advance, aimed at companies with an international footprint, especially SMEs. Check requirements and remaining funds on the official page before committing any spending.

Sources

If you want to find out which process should be the starting point for AI in your company, book a free assessment.