The software company that created FT Mercati Go to ftmercati.com

Guide · Adoption strategy

AI consultant: what they do and how to choose one

What an artificial intelligence consultant does, when an SME needs one and how to choose one: questions to ask, warning signs, public alternatives.

By FT Support Updated on 15-minute read

The short answer

An artificial intelligence (AI) consultant helps the people who run a company decide where AI delivers a measurable result, which process to start from, under what rules and at what cost, and leaves a plan others can carry out too. Italy has no official register: anyone can use the title, so you choose with checks and the right questions. Before spending, public services help too: Italy's Punti Impresa Digitale and the Digital Innovation Hubs.

  • An AI consultant decides with the people who run the company where to start and leaves a written plan: they do not sell software and do not decide in the company's place.
  • In Italy there is no official register: anyone can call themselves an AI consultant. You check the company register extract, references for systems in production, certifications and the contract.
  • The first obstacle businesses report is the lack of skills: choose someone who passes skills on, not someone who holds on to them.
  • Decisions, the internal point of contact, training and the obligations of whoever uses the system stay with the company: the consultant helps, but does not take its place.
  • Before paying anyone: Italy's Punti Impresa Digitale (the Chambers of Commerce's digital support desks) and the Digital Innovation Hubs offer free assessments.
  • A free audit, or a percentage promised before the processes have been seen, is a warning sign.

Note. Informational content updated as of 12 October 2026. Rules, funding calls and public services change: before deciding, check against the sources at the bottom of the page and, for contracts and liability, with your own advisers. The national rules and public services described are those of Italy; other EU member states have their own.

What an AI consultant does, and what they do not do

An artificial intelligence (AI) consultant works with the people who run the company on four things: understanding which processes can get a measurable result from AI, checking whether the data to do it is there, writing a step-by-step plan (roadmap) that starts from a pilot project, and writing into the plan how the people are prepared, with the usage rules (AI policy) and the measures the AI Act requires. A good result is a document that the company can put into practice with others too.

What they do not do: they do not sell software (if they do, they are a reseller or a software company: different roles, with different interests); they do not decide in the company’s place what AI can do on its own; they do not “install AI” in an afternoon; they do not replace the person in the company who will look after the project.

In job adverts and courses the same role appears under the names AI consultant or AI strategist; “artificial intelligence expert” is a broader expression, which includes researchers and technical specialists. In Italy, the standard UNI 11621-8:2026, which describes the professional profiles of those who work in AI, has been in force since 30 April 2026 (UNI is Italy’s national standards body), and in September 2026 UNI reported a European workshop agreement by CEN, the European Committee for Standardization (CWA 18398), with the same purpose. These are useful references for understanding what the role is about, not an official register.

Who is who: consultant, agency, software company, reseller, integrator

None of these roles is wrong. The problem starts when one presents itself as another.

RoleWhat it sellsWhen it is the right oneWhere its interest lies
ConsultantTime and method: analysis, priorities, plan, rulesWhen the first process is not clear, or you need a plan before spendingNone, if not tied to a product: ask for this in writing
Agency or consulting firmTurnkey projects, often with its own technical teamWhen the scope is clear and you need someone to build itIt proposes what it knows how to build
Software companyCustom software or its own productWhen you need to build something that does not existIt has an interest in getting things built
Reseller or partner of a providerLicences and configuration of a productWhen the product has already been chosenIt recommends the product it resells
Systems integratorConnections between the ERP (the company’s management software), data and toolsWhen the hard part is integrationIt favours the systems it knows

In the EU, among businesses that use AI, more than half buy it as ready-made software and about one in four have it developed or adapted by external providers (calculation on Eurostat 2024 data, businesses with at least ten people employed, multiple answers; for Italy the figure is not collected). The choice of product is therefore the decision on which the adviser’s interest weighs most. The AI Act adds an argument: whoever puts their own name or trademark on a high-risk AI system already on the market, whoever makes a substantial modification to it, or whoever changes the intended purpose of a system, even a general-purpose one, so that it becomes high-risk, becomes its provider, with the obligations that follow (Article 25). Most of an SME’s projects are not high-risk, but the rule says one simple thing: whoever adapts and resells is not a neutral third party.

When you need one, and when you do not

You need one:

  • when nobody in the company has the skills to set up the project. In Italy, among businesses that have considered AI without adopting it, the lack of skills is the first obstacle for 58.6% (ISTAT, Italy’s national statistics institute, December 2025); in the EU, for about seven in ten (calculation on Eurostat 2025 data);
  • when it is not clear which process to start from, or the departments put forward ten;
  • when you need a bridge between senior management, IT and suppliers, someone who speaks all three languages;
  • when the first pilot has failed and nobody can say why.

You do not need one:

  • to use a ready-made assistant in an office: usage rules and basic training are enough;
  • for a task that comes up a few times a year;
  • before the basic data has been put in order: a consultant who finds the ERP in a mess will tell you to tidy it up, and you will have paid to be told so. The second step of the guide on how to bring AI into your company is one you can take on your own;
  • if nobody in the company will have time to look after the project: the plan will stay in a drawer.

According to the Artificial Intelligence Observatory of the Politecnico di Milano, 77% of the Italian AI market is made up of custom projects and only 8% of the country’s small and medium-sized enterprises have started a project. Three quarters of the market, then, are projects built to measure for a single customer, by suppliers and consultants: the choice of partner weighs on the result more than the technology.

What must stay with the company

A consultant helps. Some things cannot be delegated.

  • The decisions. Where AI is used, with which data, what it can do on its own. The NIST AI Risk Management Framework, which is voluntary, says so in its GOVERN function: responsibility for decisions about risks lies with the organisation’s top management.
  • An internal point of contact. In Italy, the Ministry of Labour’s guidelines on AI in the world of work, adopted in December 2025 and non-binding, suggest an AI officer or a similar role, who may also be the data protection officer in companies that have one: this is the point of reference for the company’s people and for the authorities. No rule requires it, not even Italy’s Legislative Decree No. 179/2026, but an external consultant does not replace this person.
  • Training. Article 4 of the AI Act requires those who use AI systems to take measures to support the AI literacy of their staff and of anyone who uses those systems on their behalf; the European Commission specifies that the people of a service provider or a contractor must also have the skills appropriate to the task. The details are in the guide AI literacy: what the AI Act requires of companies.
  • The obligations of whoever uses the system. In the language of the AI Act, the company that uses a system is the deployer: that is the Regulation’s term. For the high-risk AI systems of Annex III, from 2 December 2027, Article 26 assigns it use in accordance with the instructions, human oversight entrusted to competent people, control of the input data, monitoring, keeping the logs and informing workers. A contract can share out the tasks, not the responsibility before the law.
  • The usage rules. The people who run the company decide them; the consultant can propose them. How to write them is in the guide Company AI policy: usage rules and an example.

Who can help for free, before you pay anyone

There are public services, with their limits. The ones described here are those available in Italy.

  • The Punti Impresa Digitale (PID), the digital support desks of Italy’s Chambers of Commerce. Two tools: an online self-assessment of digital maturity, called SELFI4.0, with an automatic report at the end of the questionnaire, and an assessment guided by a Digital Promoter from the Chamber, called ZOOM4.0, with interviews with senior management and staff and a more detailed report. They were created for Industry 4.0 digitalisation, not for AI in particular, but they give a picture to start from. Since July 2026 the PIDs have also had a national call for applications run through local application windows, with vouchers for technologies, training and consulting and specific attention to AI adoption, and local laboratories called AI-Labs; each Chamber decides the dates and the requirements for suppliers, and each window’s funds run out quickly. The details are in the guide Incentives 2026 for software and AI in companies (in Italian).
  • The Digital Innovation Hubs (DIHs) and the European Digital Innovation Hubs (EDIHs). The guidelines of Italy’s Ministry of Labour mention a free digital assessment delivered through the DIH network. The European Commission’s catalogue lists, as of the date of this guide, 57 hubs in Italy (with a few possible duplicates), 13 of them funded by the Digital Europe Programme; 35 list artificial intelligence among their technologies. The services vary from hub to hub: they need to be read one by one.
  • The lists of innovation managers. Italy’s Ministry of Enterprises and Unioncamere (the national union of the Chambers of Commerce) keep public lists of qualified innovation managers, created with the voucher for innovation consulting: two application windows, in 2019 and 2023, and, as of the date this guide was updated, no new window for that voucher. Being on a list is not compulsory in order to work, but it can be checked.

These services tell the company where it stands. They do not carry out the project in its place, and they work to their own timescales.

How to check a consultant

In Italy, to take one country’s example, there is no official register. Law No. 4/2013 sets the rules for the professions not organised into ordini or collegi, the bodies that govern the regulated professions: those who practise them must cite the law in every document and in all written dealings with the client, qualification is based on UNI technical standards and is voluntary, and certification is issued by accredited bodies. Six checks, in order.

  1. The company. A company register extract (in Italy, the visura camerale) says how long the company has existed, what activity it declares, who its directors are and who signs.
  2. The documents. The reference to Law No. 4/2013 is an obligation; a certification of conformity to the UNI 11621-8 standard, issued by an accredited body, can be checked in the databases of Accredia (Italy’s national accreditation body) and is a signal, not a requirement. The same goes for an ISO/IEC 42001 certification held by the consultant or the provider: it speaks of their method, it does not guarantee the project, and for an SME it is not a requirement. Since 23 October 2026, Italy’s Legislative Decree No. 179/2026 (Article 47) has required AI training as part of the continuing education of the country’s professional bodies (the ordini professionali) and, where the sector calls for it, of the professional associations under Law No. 4/2013: you can ask a consultant who is a member about this too.
  3. The public lists of innovation managers kept by Italy’s Ministry of Enterprises and by Unioncamere.
  4. References. Systems in production, not demonstrations: talk to the people who use them today, not to those who bought them, and ask how long they have been running and who maintains them.
  5. The AI tools the consultant uses. Italy’s AI law, Law No. 132/2025 (Article 13), requires anyone practising a liberal profession to tell the client, in clear, simple and comprehensive language, which artificial intelligence systems they use in their work; if the consultant is a company, the same information goes into the contract. Wanting to know which tools go into the work and where the company’s documents end up is not mistrust: it is a condition of the relationship.
  6. Who is liable. By law, the company that uses the system answers for its use; the contract must say what the consultant answers for: a wrong plan, a delay, confidential data that has been disclosed.

The other questions to ask an AI consultant

The seven questions to ask anyone who offers consulting, ourselves included, are on the page AI consulting for companies. These are for the second meeting.

  • Which data do you need, and what do you do if it is not there?
  • Who trains the people who will use the tools, and when?
  • What is written in the scope, and what is not?
  • What remains ours at the end: code, instructions given to the models, documents, data?
  • How do we get out, and how long does it take?
  • How much time do you ask of our people, and of which ones?
  • Have you ever told a client not to go ahead?

The last is the one that sets a consultant apart. Anyone who has never advised against a project has never looked at a company through the company’s eyes, only through those of their own offer.

The warning signs

SignWhy it is a problemWhat to ask
“Free audit”, “free assessment”, “free call” as the main offerTwo hours without data do not produce a plan: it is a quote in disguise. This is different from the free assessment by the PIDs and the DIHs, which is a public service“What do you deliver at the end, in writing?”
Percentages promised before seeing the processesWithout a scope and a baseline they mean nothing. Serious measurements refer to specific, repeated tasks, and are usually more modest than those promised“On which of our processes, measured how?”
An undeclared tie to a productThe consulting becomes a sale“Are you partners or resellers of any provider?”
No pilot: the starting point is an annual licence or a long contractThe guidelines of Italy’s Ministry of Labour provide for a trial phase with objectives and an evaluation report“What is the first step at which we can stop, having spent little?”
No number before and afterWithout a measurement you will never know whether it worked“What do we measure, and how do we measure it today?”
The plan does not say how people are preparedStaff AI literacy is an obligation of the company, and according to an OECD survey in seven countries (Italy not included) a third or fewer of the SMEs that use generative AI train their staff“Who needs to know what, when, and who explains it to them?”
A contract with no clauses on data, confidentiality, ownership of what remains, exitThese are the clauses that count when something goes wrong“Where are they written?”
The consultant decides alone what AI can doThe decisions are for the company to take, with written rules“Who approves the usage rules?”
Jargon without processes: they talk about models and not about the office that prepares quotesThey have not looked at the company“Which of our processes would you start from?”

How you pay for AI consulting

No figures here: every company is different, and a rate published before seeing the processes says little. There are four ways.

  • By steps, with an expected result. First the process map and the plan, then the pilot, then going into production; each step is paid for when you can see it. You can stop; it requires a written scope. It is the way best suited to those starting out.
  • By the day. Simple and transparent; the risk is that the days pile up with no finishing line.
  • On a retainer. Suited to ongoing support after going into production; to be avoided before you have seen a result.
  • As a share of the savings. It aligns interests, but requires an agreed measurement and a reliable starting number, which few SMEs have.

What the cost depends on, item by item, is explained in the guide What an AI project costs: what it depends on. The incentives that also cover consulting and training, such as the voucher of Italy’s Chambers of Commerce managed by the PIDs, are in the guide Incentives 2026 for software and AI in companies (in Italian).

The contract: what must be in it

  • The scope. Processes, data, expected results, and what is not included.
  • Data and confidentiality. Which data the consultant sees, where they process it and with which tools, what happens at the end: return or deletion.
  • Ownership of what remains. Documents, code, instructions to the models, configurations, processed data: the company’s.
  • Liability. What the consultant answers for, bearing in mind that the use of the system remains the company’s responsibility.
  • The people. Who needs to know what, when, and who explains it to them: the company itself, the provider or an external course.
  • Exit and handover. How the work is stopped, what is handed over, in what format, with what support for the handover. The NIST framework provides for the safe decommissioning of systems and for contingency plans for third-party systems and data deemed high-risk: the exit is designed at the start, not when it is needed.
  • Updating the plan. Rules and tools change; a review on a fixed date stops the plan from going out of date in a drawer.

In a manufacturing or metals company

A consultant who does not know the sector starts from what they know: marketing, customer service, writing texts. In a plant, time is lost elsewhere: in the office that prepares quotes, where they are built from drawings and specifications; in orders and order confirmations; in material certificates and quality documents; in price lists tied to the metal, with indexation clauses to recalculate; in the commodity prices that have to go into the ERP every day; in purchasing.

A consultant from the sector recognises these points in the first conversation, knows which data already exists (orders, bills of materials, certificates, prices) and which has to be created, and knows that the hard part is not the model: it is getting it into the systems the company uses every day. A small Italian manufacturer quoted by the OECD says so too: it uses ready-made tools and names integration with the company’s systems as the real difficulty.

How we work

We start with a first conversation, no obligation, by video call, in English or Italian, in which we say whether we see a measurable result and where we would start; if we do not see one, we say so. Then come a day with senior management for the process map and priorities, and a step-by-step plan that stays with the company and can be carried out with anyone. We use replaceable models, and we write this in the plan. We put AI into production in our own company first, with the declared status of each system: we describe it in the case studies. We do not publish rates: the quote comes after the first conversation, by steps, with what you receive written next to what you pay.

What our AI consulting for companies includes and delivers, with the answers to the seven questions, is on the service page; the steps are in the method.

Sources

  1. UNI 11621-8:2026, Non-regulated professional activities: professional profiles for ICT, Part 8, role profiles for the activities carried out in the artificial intelligence sector (Italian standard, in force since 30 April 2026; in Italian)
  2. UNI (Italy's national standards body), AI professional profiles: a CWA has been published (CEN, CWA 18398; news item of 8 September 2026, in Italian)
  3. Italy's Law No. 4 of 14 January 2013, provisions on professions not organised into professional bodies (Normattiva, in Italian)
  4. Italy's Law No. 132 of 23 September 2025, Article 13, provisions on the liberal professions (Normattiva, in Italian)
  5. Italy's Ministry of Labour and Social Policies, Ministerial Decree No. 180 of 17 December 2025: guidelines for the implementation of AI in the world of work (in Italian)
  6. Regulation (EU) 2024/1689 (AI Act), Article 25: when a distributor, a deployer or a third party becomes a provider (consolidated text, European Commission AI Act Service Desk)
  7. Regulation (EU) 2024/1689 (AI Act), Article 26: obligations of deployers of high-risk AI systems (AI Act Service Desk)
  8. Regulation (EU) 2024/1689 (AI Act), Article 4: AI literacy, and Article 3: definitions (AI Act Service Desk)
  9. Regulation (EU) 2026/1744 (the Digital Omnibus on AI), in force since 27 July 2026
  10. European Commission, AI literacy: questions and answers (updated as of 27 July 2026)
  11. NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), January 2023, GOVERN function
  12. ISTAT (Italy's national statistics institute), Imprese e ICT (businesses and ICT), year 2025, report of 15 December 2025 (PDF, in Italian)
  13. Eurostat, Artificial intelligence by size class of enterprise (isoc_eb_ai), update of 15 June 2026
  14. Artificial Intelligence Observatory, Politecnico di Milano, press release of February 2026 (in Italian)
  15. OECD, AI adoption by small and medium-sized enterprises, December 2025
  16. OECD, Generative AI and the SME workforce, November 2025
  17. Punto Impresa Digitale, assessment of businesses' digital maturity: SELFI4.0 and ZOOM4.0 (in Italian)
  18. Punto Impresa Digitale, Bando Voucher Doppia Transizione 2026, the twin transition voucher call (in Italian)
  19. European Commission, catalogue of European Digital Innovation Hubs
  20. Italy's Ministry of Enterprises and Made in Italy, voucher for innovation consulting (Voucher per consulenza in innovazione), first and second application windows (in Italian)
  21. Unioncamere (the national union of Italy's Chambers of Commerce), list of innovation managers (in Italian)
  22. Accredia (Italy's national accreditation body), databases of accredited bodies and certifications (in Italian)
  23. UNI, the standard UNI CEI ISO/IEC 42001 (management systems for artificial intelligence) published in Italian (page in Italian)
  24. Unioncamere, press release of 10 July 2026 on the PIDs' Bando Voucher Doppia Transizione and on the AI-Labs (in Italian)
  25. Italy's Legislative Decree No. 179 of 7 October 2026, powers of the national AI authorities and penalties, Gazzetta Ufficiale No. 234 of 8 October 2026 (in Italian)

Checked on 28 October 2026. Rules and deadlines change: always check the official source before you decide.

Frequently asked questions

The short answers

What does an artificial intelligence consultant do?

They work with the people who run the company to understand which processes can get a measurable result from AI, check whether the data needed is there, write a step-by-step plan with a first pilot project, and put into the plan how people are prepared, with the usage rules and the measures the AI Act requires. A good result is a plan the company can carry out with others too. They do not sell software, and they do not decide in the company's place what AI can do on its own.

Does an SME really need an AI consultant?

It does when nobody in the company has the skills to set up the project, when it is not clear which process to start from, or when there is nobody to act as a bridge between senior management, IT and suppliers. It does not need one to use a ready-made assistant in an office, for a task that comes up a few times a year, or before the basic data has been put in order. In Italy, among businesses that have considered AI without adopting it, the lack of skills is the first obstacle for almost six in ten.

AI consultant, AI strategist, artificial intelligence expert: are they the same thing?

Almost. AI consultant and AI strategist are the names under which the same role appears in job adverts and courses; "artificial intelligence expert" is a broader expression, which also covers researchers and technical specialists who do not do consulting. What counts is not the name, but what they deliver and how it can be checked.

Is there an official register or a standard for AI consultants?

In Italy, to take one country's example, there is no register: it is a profession not organised into ordini or collegi (the bodies that govern the regulated professions); it comes under Law No. 4/2013, and anyone can practise it. What does exist, since 30 April 2026, is the Italian technical standard UNI 11621-8:2026 on professional profiles in artificial intelligence: Law No. 4/2013 provides that a professional can ask an accredited certification body for a certificate of conformity to the standard, but certification is voluntary. Whether any bodies already certify against this standard, and which, can be checked in the databases of Accredia, Italy's national accreditation body: it is a signal, not a requirement.

How much does an AI consultant cost, and how do you pay?

It depends on the number of processes, the state of the data, the integrations with the company's systems and the people to be trained: a rate published before seeing the company says little, and is best treated with caution. There are four ways to pay: by steps with an expected result, by the day, on a retainer, or as a share of the savings achieved. The first is the best suited to those starting out, because it lets you stop having spent little.

Can the consultant also be the one who sells or resells the software?

Yes, but it must be declared in writing, because it changes the advice you receive: whoever resells a product tends to recommend that product. The AI Act adds that whoever puts their own name or trademark on a high-risk AI system, makes a substantial modification to it, or changes the intended purpose of a system so that it becomes high-risk, becomes its provider, with the obligations that follow. Always ask whether the consultant is a partner or reseller of any provider.

Who is liable if the AI gets it wrong: the consultant or the company?

The company that uses the system, the deployer in the language of the AI Act, answers for the obligations of whoever uses it and for the decisions taken with its help that affect customers and workers: for high-risk AI systems, Article 26 assigns it use in accordance with the instructions, human oversight, monitoring and the keeping of logs. The provider of the system answers for its own obligations and for defects in the product. The contract with the consultant can set out what the consultant answers for, for example a wrong plan or confidential data that has been disclosed, but it does not transfer the obligations laid down by law. That is why the liability clause should be read before signing.

Who can help me for free to understand where to start?

In Italy, two free public services: the Punti Impresa Digitale of the Chambers of Commerce (the Chambers' digital support desks), with an online self-assessment and a guided assessment of digital maturity, and the Digital Innovation Hubs, which the guidelines of Italy's Ministry of Labour mention for a free digital assessment. They tell the company where it stands; they do not carry out the project in its place.

Is a free AI audit a good way to start?

It depends on who offers it. The free assessment by Italy's Punti Impresa Digitale and by the Digital Innovation Hubs is a public service and gives a picture to start from. The "free audit" offered by those who sell consulting or software is usually a quote in disguise: two hours without data do not produce a plan. The question to ask is what is delivered in writing at the end.

Does the consultant have to say whether they use artificial intelligence in the work they do for the company?

In Italy, if they practise a liberal profession, yes: Law No. 132/2025 (Italy's AI law), Article 13, requires them to tell the client, in clear, simple and comprehensive language, which artificial intelligence systems they use. If the consultant is a company, the same information, together with where the company's documents are processed, should be asked for and written into the contract.

Do you have a process you would like to see working better?

Tell us about it in a first conversation, with no obligation: together we work out whether and where AI brings a measurable benefit, and which step to start with.