Every one of the AI consulting firms on your shortlist has a good pitch deck now. The demos are slick, the case studies glow, and the word "agentic" appears on slide two. None of that tells you whether they'll get something working in your business and leave you able to run it.
The short answer: choose an AI consultancy on evidence of delivery, not presentation. Ask to meet the people who will actually do the work, ask for a reference whose system is still running, ask what they'd advise you not to do, and get ownership, data handling and exit terms in writing. Then start with a small, paid piece of work before committing to a programme.
Key points:
- The stakes are real. In a Gartner survey of 782 infrastructure and operations leaders, only 28% of AI use cases fully succeeded and met ROI expectations, and 20% failed outright.
- Outside help can improve your odds when it's the right help. MIT's NANDA research found external partnerships reached deployment roughly twice as often as internal builds.
- Firm size matters less than fit, seniority on your account and production experience.
What types of AI consulting firms are there?
"AI consulting companies" covers a lot of ground. Before comparing firms, decide which kind you're comparing. A global systems integrator and a two-person automation shop are not competing for the same work, even if they both rank for the same search.
| Type of firm | Usually fits | Watch for |
|---|---|---|
| Big Four and global systems integrators | Large enterprises, multi-country programmes, heavy regulation | Partners who sell, juniors who deliver; long discovery phases |
| Boutique AI consultancies | Mid-sized firms that need senior attention and practical delivery | Capacity limits; dependence on one or two people |
| Industry specialists | Sectors with specific data, workflows or regulation (health, finance, legal) | Reusing last client's answer for your different problem |
| Platform partners (Microsoft, Google, AWS, Salesforce and so on) | When the platform decision is already made | Every answer happens to be their platform |
| Independent consultants and fractional leaders | SMEs, first projects, strategy and governance, vendor selection | Whether they can build, or who they'd bring in to do it |
| AI automation agencies | Defined builds: lead handling, document processing, assistants | Retainers, lock-in, thin testing |
If you're still deciding between advice and delivery, read AI consultant vs AI agency vs building in-house first. It'll save you interviewing the wrong kind of firm.
What questions should you ask an AI consultancy?
These are the questions I'd want answered before signing. The answers matter, but so does the way they're given. A firm with real delivery experience answers specifically and quickly. A firm without it answers with frameworks.
- Who, by name, will do the work? Meet them before you sign. If the senior person in the pitch won't be on the project, find out how many hours a month they will be.
- Show me something you built that is still running. Pilots are easy. Ask how long it has been live, what broke and who maintains it now.
- Can I speak to that client? Ask the reference how the firm behaved when something went wrong, not just whether they were happy.
- What would you advise us not to do? A good consultancy will steer you away from at least one idea. One that says yes to everything is selling.
- What happens if our data isn't ready? It usually isn't. You want a plan, a cost and an honest view of whether the project still makes sense.
- How will we measure success? Expect a business measure (hours saved, response time, error rate, revenue) with a baseline taken before work starts. My guide to measuring AI ROI covers what good looks like.
- Where will our data go? Which models and vendors, which regions, whether it's used for training, and what sub-processors are involved.
- What do the first four weeks deliver? Something tangible, even if small. Months of discovery with nothing to show tends to become a habit.
- How do you hand over? Documentation, training, credentials and a named person on your side who can run it.
- How do you use AI in your own business? Firms that run AI internally tend to understand its failure modes. Firms that only sell it often don't.
That last one is underrated. A consultancy that doesn't use AI on its own work is selling you a holiday it hasn't taken.
What are the red flags?
Most of these show up in the first two meetings if you're looking for them.
- ROI figures before discovery. If they've promised a percentage saving before seeing your processes or data, the number came from a template.
- One answer for everything. Every problem leads to the same platform, model or product they happen to resell.
- Only pilots in the portfolio. Nothing they can show you has been in production for six months or more.
- Vague on the team. They can't name who will work on your account, or they're cagey about subcontractors.
- Silence on risk. No mention of data protection, testing, human review of outputs or what happens when the model is wrong.
- Too much, too fast. Gartner's analysts found that among leaders who'd had an AI failure, many said they had expected too much, too fast. A consultancy that encourages that is setting you up for it.
For the wider pattern of what sinks these projects, see why AI projects fail.
What should the contract cover?
Most buying guides skip this part, which is a shame, because it's where buyers get hurt. A proposal describes what you'll get. The contract decides what you keep.
- Ownership. Agree in writing which code, prompts, workflows and documentation you will own. Identify pre-existing supplier components, third-party licences and the continuing usage rights you need.
- Accounts and credentials. AI platforms, automation tools and API keys are set up in your name and paid on your card, or transferred to you at handover.
- Data processing. A data processing agreement, a list of sub-processors and clarity on where data is stored. In the UK, the ICO's guidance on AI and data protection is the reference point, though the ICO notes it is under review following the Data (Use and Access) Act. Your AI governance framework should say what's allowed.
- Acceptance. What "done" means, how it's tested and what happens if it isn't met.
- Exit. A reasonable notice period, a handover obligation and no penalty for taking the work in-house or to another supplier.
Treat these as negotiating objectives, not automatic legal rights. A managed service may have a different ownership model; make sure its export and continuity terms still let you leave.
Are the top AI consulting firms the best choice?
Sometimes. If you're a multinational rolling out AI across several regulated markets, the large firms bring scale, legal cover and the ability to staff a programme quickly. Your board may also simply prefer a recognised name.
For a business with 20 to 500 people, the "best AI consulting firms" lists are less useful than they look. Most are written by firms that appear on them. What predicts a good outcome is closer to what buyers in the MIT research said they valued: a vendor they trust, deep understanding of their workflow, minimal disruption to current tools and clear data boundaries. The same report found executives leaned heavily on peer recommendations and referrals. That's worth copying. Ask people in your sector who they used and whether the thing still works.
Keep the size of the opportunity in proportion, too. The UK government's AI Adoption Research found that lack of an identified need and limited AI skills were the most commonly cited barriers to adoption. For most SMEs, the first job is choosing the right problem. You don't need a global firm for that, and you shouldn't pay for one. Market rates are in how much an AI consultant costs.
My take
I've sat on both sides of this table: as a group chief digital officer and a CTO, and as a consultant and senior lecturer in project management. The firms that disappointed rarely lacked intelligence. They lacked accountability for what happened after the presentation.
So I weight three things above everything else. Seniority on the account, because AI projects need judgement calls weekly. Production evidence, because a demo proves almost nothing about running a system on messy real data. And willingness to say no, because the most valuable advice in AI right now is often "don't automate that yet".
The pattern I see often in SMEs is choosing on price or charm after one call. Ninety minutes of reference calls will tell you more than nine pages of proposal.
How to run a shortlist in four weeks
- Week one: write a one-page brief. The problem, the process it sits in, rough volumes, the data involved, your budget range and who internally owns it. If you can't fill this in, start with an AI readiness check.
- Week one: build a long list of five or six. Get names from peers in your sector, then add a couple from search. Mix types if you're unsure which fits.
- Week two: hold 45-minute calls. Use the ten questions above. Note who answers specifically and who changes the subject.
- Week three: ask two or three for a short proposal. Ask them to include the team, the first four weeks, how success is measured and their standard contract terms.
- Week three: call references. At least one per finalist, ideally a project that's been live for six months or more.
- Week four: commission a small paid piece of work. A scoped discovery or an AI proof of concept with a clear success measure and a fixed fee.
- Decide on evidence. If the small piece went well, extend. If it didn't, you've lost weeks, not a year.
Compare proposals on the same evidence
Give each finalist the same redacted example, volumes and constraints. Ask for their assumptions, exclusions and acceptance test alongside the price. A proposal that names a smaller first phase and an explicit uncertainty can be more useful than a confident fixed total based on data the supplier has never seen.
Frequently asked questions
What should I look for in an AI consulting firm? Look for named senior people who will work on your account, systems they've built that are still running, and references you can call. Check that they have a clear approach to data protection and that they'll hand over ownership of everything built for you.
How do I vet an AI consultant before hiring? Ask for a live example of their work, speak to at least one past client, and ask what they'd advise you not to do. Then start with a small paid piece of work before committing to a larger engagement.
Should I hire a big consulting firm or a boutique for AI? Large firms suit large, multi-country or heavily regulated programmes. For most SMEs, a boutique or independent consultant gives more senior attention for the money. Judge on team seniority and production evidence rather than brand.
What questions should I ask an AI consultant about data privacy? Ask which AI models and vendors will process your data, where it's stored, whether it's used for training, and which sub-processors are involved. Ask for a data processing agreement before any real data is shared.
How long does it take to choose an AI consultancy? A sensible process takes around four weeks: brief, long list, calls, short proposals and references. Ending it with a small paid project is the fastest way to find out whether the firm can deliver.
Sources
- Gartner Says AI Projects in I&O Stall Ahead of Meaningful ROI Returns, Gartner, April 2026
- The GenAI Divide: State of AI in Business 2025, MIT NANDA (Challapally, Pease, Raskar, Chari), July 2025
- AI Adoption Research, Department for Science, Innovation and Technology, 2025
- Guidance on AI and data protection, Information Commissioner's Office
If you're shortlisting AI consulting firms and want an independent second opinion on the proposals, let's talk.
