To choose an AI consultancy, ignore the logo wall and ask ten questions that expose delivery reality: who shows up, what's left behind, and what they'd refuse to sell you. Firm-by-firm comparisons are in our honest guide to the best AI consulting firms in London and the UK; this page is the interrogation kit — the questions vendors routinely dodge, why they dodge them, and what a good answer sounds like.

Key Takeaways

  • Shortlists fail on logo bias and demo theatre. Every firm has a stunning demo and three referenceable logos; neither predicts what happens in your organisation in week six.
  • The ten questions below are dodged for a reason: honest answers reveal bench-led delivery, unmeasured outcomes, stack bias and scope traps before you sign.
  • 2026's consolidation raised the stakes: Accenture completed its acquisition of Faculty in March 2026, CGI absorbed BJSS in 2025, and Mesh-AI merged into Indicium AI. Ask any independent firm who might own them mid-engagement.
  • The single most revealing question is "what work are you wrong for?" A firm that can't name its wrong-fit client will take any budget — including one it can't serve.
  • You can build a defensible shortlist in an afternoon: define the problem type, apply the ten questions to three firms, and score the answers — including ours.

Why AI consultancy shortlists go wrong

Three patterns account for most bad hires. Logo bias: impressive client lists prove a sales function, not that your engagement gets the A-team. Demo theatre: polished demos test the firm's presentation layer, not delivery in your data, politics and workflows. Category confusion: hiring a build firm for an adoption problem (or the reverse) fails on day one, before anyone underperforms — we've mapped that distinction in advisory vs implementation vs training. The questions below are designed to puncture all three.

The 10 questions AI consultancies dodge

1. Who, by name, will do the work — and at what allocation?

Why they dodge: the partner selling is rarely the team delivering. Good answer: named individuals, their track record, and their percentage on your account. "Resources from our AI practice" means bench.

2. What will my teams do differently ninety days after you leave?

Why they dodge: most engagements end at handover, and post-handover usage is nobody's KPI. Good answer: specific behaviours, per team, with the usage metric that will verify them. This is the question our whole AI adoption training guide exists to arm.

3. Show me a comparable engagement with measured outcomes — not a logo.

Why they dodge: "40% productivity gain" claims rarely survive the question "measured how, from what baseline?" Good answer: a before/after with the metric, the baseline, and a client contact willing to take a call.

4. What's your commercial relationship with the tools you'll recommend?

Why they dodge: alliance partnerships and reseller margins shape recommendations. A Microsoft-alliance firm recommends Copilot with impressive regularity. Good answer: disclosed partnerships, plus one recent case where they recommended against their partner's stack.

5. What would you tell us not to do?

Why they dodge: "no" shrinks the statement of work. Good answer: a concrete cut — "your data isn't ready for that use case; spend the money on these two instead." A proposal that accepts your entire wishlist wasn't reviewed, it was transcribed.

6. How do you price, and what exactly triggers an overrun?

Why they dodge: vagueness here is where margins live. Good answer: fixed scope with named change-order triggers, or time-and-materials with a cap. Benchmarks for every band are in our UK AI consulting cost guide.

7. Who owns what you leave behind?

Why they dodge: reusable IP, retained prompts and frameworks, licence-encumbered artefacts. Good answer: you own the deliverables, the playbooks and the training materials, in editable form, on your systems.

8. How will you handle our governance — UK GDPR, ICO guidance, sector rules?

Why they dodge: governance is often subcontracted to a slide. Good answer: named frameworks, who signs off, and artefacts your legal and security teams can actually review. Regulated sectors should push harder still — see our financial services guide.

9. What work are you wrong for?

Why they dodge: admitting a wrong-fit client costs revenue. Good answer: a specific, disqualifying answer delivered without flinching. Ours: don't hire us to build production models or run a 5,000-seat global integration — that's Faculty-into-Accenture, QuantumBlack or IBM territory, and we say so in our own comparison guide.

10. What do you need from us for this to work?

Why they dodge: "very little" closes deals. Good answer: a demanding list — executive sponsor hours, champion time, data access, decision deadlines. Change programmes that cost the client nothing in effort deliver exactly that.

"Question nine is the whole interview compressed. Every firm optimises for something — scale, margin, elegance, speed. A consultancy that claims to fit every problem is telling you it optimises for revenue. The ones worth hiring can describe their wrong-fit client faster than their ideal one."

— Toni Dos Santos, Co-Founder, Spicy Advisory

The 2026 consolidation question

The UK market is consolidating fast: Accenture completed its acquisition of Faculty — London's best-known independent AI firm — in March 2026; CGI absorbed BJSS in 2025; and Mesh-AI merged into Indicium AI with the combined brand launching in early 2026. For buyers this means two extra diligence questions for any independent firm: "what happens to our engagement, rates and team if you're acquired?" and "which of the people I'm buying are locked in?" Acquisitions reshuffle exactly the senior people boutiques are hired for.

"Clients treat 'what do you need from us' as a politeness question. It's the hardest one on the list. When we answer honestly — an executive sponsor who shows up, champions with four hours a week, decisions inside five working days — some prospects walk away. Better before the invoice than after."

— Meera Sanghvi, Co-Founder, Spicy Advisory

Build your shortlist in an afternoon

  1. Hour one — name the problem type. Build, integrate, decide, or adopt. One sentence, agreed with your sponsor. This eliminates half the market immediately.
  2. Hour two — pick three candidates in that lane. Use a comparison (ours or anyone's), one large firm, one specialist, one boutique, and confirm they serve your size. Mid-market? Start from our UK SME and mid-market page.
  3. Hour three — send the ten questions in writing. Written answers are harder to charm through, and non-answers are visible by return of email.
  4. Hour four — score and book calls. Two points per straight answer, one per partial, zero per dodge. Interview the top two; the call is for pressure-testing, not discovery.

Fair warning if you send these to us: we've published our answers in advance — named founders who deliver personally, public pricing, measured 90-day usage as the contracted outcome, and a standing list of work we'll turn down. Hold every firm, including us, to that bar. Our client reviews are the receipts.

Frequently asked questions

What should I ask an AI consultancy before hiring?

Ten things: named delivery team, the 90-day behaviour change, measured outcomes from a comparable client, tool-vendor conflicts, what they'd cut, pricing and overrun triggers, IP ownership, governance handling, their wrong-fit client, and what they need from you. Written answers beat pitch meetings.

Are big consultancies better than specialists for AI?

Better for global integration scale, audit-grade assurance and institutional cover; worse for speed, senior attention per pound, and workflow-level adoption. The honest deciding factor is problem type and engagement size, not brand.

How many AI consultancies should I shortlist?

Three from the correct category, interrogated in writing, beats seven generalists in a beauty parade. Category first — a brilliant build firm still fails an adoption brief.

What's the biggest red flag in an AI consulting proposal?

No named people and no measured outcome. If the proposal can't say who delivers and what will be demonstrably different in ninety days, the engagement is staffed and scored after signature — on their terms, not yours.

Sources

Send us the ten questions

Seriously — email them or bring them to a call. We're a founder-led boutique (50+ companies, 1,500+ professionals trained, 4.98/5 rating) and we win engagements on exactly this test, because the answers are already public. If we're the wrong firm for your problem, we'll name who's right.

Run the Free AI Adoption Scorecard Book a Free 20-Minute Call