8. Choosing Your AI Consultant

If you have read this far, you have the minimum knowledge to decide. But deciding is not enough: the transformation must be designed, built, measured and governed for years, and you do not have the time to do it alone. The main advice of this book is to engage a qualified AI consultant as a strategic partner. This chapter tells you why the consultant must be external, how to tell a qualified one from a merely certified one, how not to confuse technical skills with strategic ones, and how to set up the relationship: objectives, powers, rules, documentation. By the end you will have a scoring grid, twelve questions and a six-point pact.

The skills I describe are those of The AI Consultant's Handbook, the text consultants qualify against: if a consultant has read that book and you have read this one, you will speak the same language from the first meeting.

Why an outsider

No internal comfort zones

The first reason is the simplest: whoever is inside the company has a comfort zone, and the AI transformation touches it. The IT manager has their systems and suppliers; the sales director has their agencies; every manager has their people and routines. None of them, in good faith, will propose redesigning their own department first. An outsider has no departments to defend, no suppliers to protect, no colleagues to spare, and can say what an insider does not: that a process lives only in one person's memory, that a long-standing agency is no longer needed, that the first agent belongs precisely in the office of the most opposed manager.

A trade that cannot be improvised

The second reason is that designing and governing AI Teams is a new trade, requiring at once experience of running a business, functional knowledge of the technologies, and a method. Across the EU, the barrier most cited by companies that considered AI and did not adopt it is lack of expertise — 70.9% of them. Engaging an outside consultant is not a fallback: it is how SMEs access skills they cannot hire.

The speed advantage

The third reason is speed. A qualified consultant has a proven method, their own tools and — increasingly — their own team of agents that produces in two days the documentation an agency delivers in three weeks. In the method I describe in The AI Consultant's Handbook, the project document is delivered within 48 hours of the audit visit, and a full AI Team is running within 45-60 days. An insider learning everything from scratch will take a year, and in the meantime a competitor will have accumulated data and processes.

Watch out. External does not mean absent. The right consultant shows up in the company, talks to people, sits in the monthly reviews and answers to you for the results of every step. Whoever works only remotely and delivers slides is not a strategic partner: they are a supplier of documents.

Qualified, not merely certified

The market is full of offers

The AI consulting market is crowded with offers and empty of quality. Thousands of IT consultants, digital agencies and trainers have added the word "AI" to their business cards. Many hold a certification — on a platform, on a model, on a course — that attests a technical skill. No technical certification attests the ability to redesign a company.

The two most common images are both wrong: the expensive technician who does theory, and the bright young person who learned to build agents on a platform and offers themselves as an "AI automation agency." The first produces correct slides without working agents; the second produces brilliant, fragile demos with no return.

Technical skills and strategic skills

The consultant you need as a partner is an expert in business and organization who knows AI at a functional level — what an agent does, what it costs, what its limits are, how it fits into the organization — and who can design, have built and govern the introduction of agents and AI Teams while measuring the result. The three key words are experience, design and measure. Whoever has never been accountable for a P&L struggles to design for someone who is.

Technical skills matter, but in another role: The AI Consultant's Handbook distinguishes the senior consultant — horizontal, the owner's counterpart, who reads the business model and writes the strategy, the brief and the project — from the executive consultant — vertical, specialized in building agents, system connections and technical environments. You engage the senior; the senior brings the executives, their own or from a network, and answers for them. If you engage an executive directly, you will get agents without a strategy: the graveyard of agents from chapter 5.

Three things that mark a qualified consultant

Beyond certifications — which are becoming verifiable: ISO/IEC 42001, the AI management-system standard, already counts more than 350 certified organizations worldwide and is the one certificate worth asking a consultant's firm about — three things mark a qualified consultant.

The method: they have a repeatable way of auditing, designing and measuring, and can explain it in ten minutes in your words. Cost transparency: they can tell you what consulting, platforms and consumption will cost, how those will change over time, and they show you the three lines separately every month, each with a cap. Responsibility: they sign a project with verifiable Conditions of Satisfaction for every step and answer for it at every review. Whoever sells a demo, does not declare consumption and takes on no objectives is not a consultant: they are a salesperson.

The scoring grid

I use this grid with CEOs to compare several candidates. Ten criteria, a score from 0 to 3 on each, a minimum threshold of 22 out of 30, and no "mandatory" criterion below 2.

Try this. Have the grid filled in by you and by one of your managers separately, after the first meeting. If the scores differ by more than five points, the consultant spoke to only one of you: not a good sign.

The twelve questions of the first meeting

I have CEOs ask these in the first meeting, with the answer a qualified consultant should give. You do not need all twelve: five are enough to know.

1. "Where do we start?" — Qualified: "Half a day in the company to understand how it is made, where it earns and where it loses. The words 'agent' and 'AI' can wait." Not qualified: starts from a demo.
2. "What will I have in hand after the first phase?" — "A diagnosis with the audit of lost hours: how many hours, how many people, how much each repetitive process costs. And a list of missing data to verify, never invented."
3. "What is the first agent?" — "The one with high return, low risk and data ready: first measurable value within thirty days. Not the website chatbot."
4. "How do we measure whether it worked?" — "With Conditions of Satisfaction written beforehand: what must be true, how much, by when, verified how, what happens otherwise."
5. "What does it cost?" — "Three separate lines: my consulting, the platforms, the consumption. Each with a monthly cap. You will see them every month on a dashboard."
6. "What happens if a step misses its objective?" — "I declare it, we renegotiate or we stop. I do not hide a result in a report."
7. "What happens to my people?" — "I involve them from the first agent as supervisors, with new roles and dedicated training. Whoever has the most to lose has the most to teach."
8. "Who builds the agents?" — "Vertical executives, mine or from my network, working on the project I write. I answer for them."
9. "How do you handle the rules — AI regulation, data protection?" — "They are gates inside the method: contracts, notices, documented training, human oversight designed before an agent is published."
10. "When can I stop?" — "At every step. Every step produces a verified result that stays yours."
11. "What stays with me and what does the AI Team decide on its own?" — "We write it in the playbook: what it decides alone, what needs me, what needs you, the absolute prohibitions."
12. "Can you show me a case with before-and-after numbers?" — "Yes, two, anonymized, with KPIs and a contact you can call."

The relationship pact

Engaging an AI consultant as a strategic partner means setting up an ongoing relationship — not a project with a start and an end — because the transformation in steps lasts years and consulting turns into a cycle of continuous improvement. The best way to set it up is a relationship pact, which your lawyer then turns into a contract. These are the six elements it must contain.

1. Objectives and CoS per step. The plan in steps (audit → first agent → first AI Team → mixed organization), with the Conditions of Satisfaction of every step, negotiated and signed by both before starting. No next step without acceptance of the previous one.

2. Scalability and the right to stop. Every step has a cost, a duration and an expected result; you can accelerate, slow down or stop at the review of every step without penalties. What has been built stays yours.

3. Involvement of the top and of the organization. Who attends the monthly reviews (you, the function heads involved, HR, finance); who is the "customer" of each AI Team; how and when the plan is communicated to people.

4. Separation of duties and powers. What the AI Team decides alone; what the consultant decides; what requires your approval (price changes, tone of voice, promotions, spending above threshold); the absolute prohibitions (brand, compliance, safety). This is the decision playbook, and it is written before the first agent.

5. Rules of relationship and cadences. Daily report from the dashboard, weekly summary, monthly review, and the escalation triggers: a complaint from an end customer, an anomaly in the data, a breach of a spending threshold, a compliance risk, a KPI off target by more than 20% for two weeks.

6. Documentation and ownership. Every agent has a change log (what, when, who, why) and every version can be restored; project documents are shared; ownership rules are clear (the consultant's methods and internal tools stay theirs; the customizations, data and company-specific skills stay yours); data-processing agreements are signed before any processing.

Data point. A qualified consultant will tell you that AI has three costs — consulting, platforms, consumption — and will show them to you separately every month, each with a cap. Cost transparency is the main reason clients stay: not because AI is cheap, but because they know what it costs and what it earns. Source: G. Taviani, The AI Consultant's Handbook, chapter "The Costs of AI," 2026.

Working together, after the signature

Three behaviors of yours make the difference between a consultant who produces results and one who produces documents. Be present: at the audit visits, the monthly reviews, the presentations to staff. Ask with CoS: every request you make to the consultant — a new agent, a change, an analysis — with what must be true, how much, by when. And accept breakdowns: when the consultant declares that a step missed its objective, that is the moment the relationship proves its value; punishing them for honesty guarantees you will never hear it again.

Where to look. The AI Workspace Club (aiworkspace.club) is a community of AI consultants for SMEs: it qualifies consultants on knowledge, experience and results, and offers companies a public register of profiles. A consultant qualified there works with the method, the CoS and the cost transparency described in this book. It is not the only place to look, but it is one where the grid in this chapter has already been applied.

What to take away

The consultant must be external, without internal comfort zones, and must be a senior with management experience who knows AI — not a technician or a trainer. You recognize them by three things — method, cost transparency, responsibility with CoS — and you evaluate them with the grid and the twelve questions. The relationship is set up with a six-point pact: objectives per step, the right to stop, involvement, powers, cadences, documentation.

Your to-do list.

1. Before the first meeting, write your own answers to questions 2, 3 and 11. If the consultant's answers are better than yours, you have found the right partner.
2. Score two candidates on the grid; do not proceed with anyone below the threshold on the mandatory rows.
3. Put the six elements of the relationship pact in the contract before signing, exit clause included.