How to choose an AI consulting firm for your company
Most proposals look alike by the second meeting. This guide sorts the firms by what they actually do, then sets out the evidence and questions that tell them apart.
Contents
The hardest part of hiring an AI consultant is not finding one. It is comparing firms whose proposals describe different work, at different prices, with evidence of uneven quality.
Why the choice is harder than it looks
The cost of choosing badly is high and well documented. In S&P Global's 2025 survey, the average company scrapped 46% of its AI proofs of concept before they reached production1. MIT's 2025 study found that generic tools stall inside companies because they do not learn from or adapt to the way work is done there2. A consultant who builds an impressive pilot that never fits the company's workflow has delivered the most common failure in the market.
The other difficulty is vocabulary. Every firm in our main guide describes its offer in similar terms: strategy, data, agents, governance. The differences sit in what the firm will actually own once the slides are done.
The four types of firm
| Type | What it usually owns | Examples in our guide |
|---|---|---|
| Global integrator | Multi-year programs across data, governance, workforce and systems | Accenture, IBM Consulting, Deloitte |
| Consulting and engineering firm | Strategy through build, sometimes ongoing operation | Slalom, Thoughtworks, Fractal |
| AI specialist | One defined build, on a project or team basis | Addepto, LeewayHertz |
| Architect-led platform | A redesign of daily operations, run on a system the client owns | OpenLazarus |
The type sets the shape of the engagement more than any single capability does. Accenture's Telstra case is an integrated partnership that delivered a five-year roadmap in two years3. Addepto, by contrast, offers a delivered solution, a joint team or full managed delivery for a defined system4. OpenLazarus starts with an architect working with leadership on how the company should run, then designs processes and AI teammates with it5.
Seven questions that separate firms
- Who will we work with, and for how long? Names and roles for the first phase, not the pitch team.
- What will you own after launch? Slalom, for instance, offers managed services for agentic workflows after go-live6. Others hand over at launch. Either can work, if it is explicit.
- What will we own? Data, code, prompts, models and documentation. Thoughtworks lists reducing vendor dependency and controlling token costs among its strategy goals7, which is the kind of commitment worth seeing in writing.
- Which named client can we call? A named client with a measured result, such as Slalom's 50% cut in call wait times at Genie8, is worth more than a logo wall.
- How will you measure the first phase? The metric, the baseline and who measures it.
- How do you govern what the AI does? Look for approval limits and logs. NIST's voluntary AI Risk Management Framework, organised around govern, map, measure and manage, is a useful common vocabulary9.
- What does the first phase cost, and what does the year cost? Most firms in our guide quote only after scoping10.
Matching firm to company
For a company with a few hundred staff and no AI team, the deciding factor is usually who will run the system once the consultant leaves. A firm that trains internal staff, operates the system or leaves it on a platform the company controls reduces that risk. OpenLazarus, for example, says it trains the client's IT team to maintain the build without it11. A company with its own software engineers has more options, and an engineering consultancy such as Thoughtworks becomes a stronger fit.
A selection process in five steps
- Write a one-page brief. The business result, the processes involved, the systems they touch and who on staff will own the outcome.
- Pick a type before a firm. Use the table above to choose one or two types that fit the brief.
- Shortlist three firms. Ask each for the same things: a first-phase scope, a named reference, an ownership clause and a price for the phase.
- Score the answers. Use the seven criteria on our methodology page, weighted for what matters to you.
- Call the references. Ask what changed, how it was measured and who maintains it today.
The outcome of a good selection is a signed first phase with a named team, a measure everyone agreed in advance and a clear answer to what the company owns at the end. Before signing, read our list of red flags in AI consulting proposals.
Questions buyers ask
What should I look for in an AI consultant?
Look for four things in writing. Work with your leadership on where AI changes the business, engineers who build and run what is recommended, terms that leave your company owning its data and the system, and named client results you can verify. Then check that the firm's usual client is roughly your size.
Is a big consulting firm better than a small AI specialist?
Not for every company. The global integrators publish the deepest client results, but their flagship cases are multi-year programs at very large organisations. A specialist or an architect-led platform usually fits a single-country company with a few hundred staff better, provided it can show named work.
How many AI consulting firms should we talk to?
Three is a practical shortlist, ideally one from each type that fits your goal. More than that tends to produce proposals that cannot be compared on the same terms.
Sources
- AI project failure rates are on the rise: report. CIO Dive, 14 Mar 2025. Accessed 8 Nov 2026.
- MIT report: 95% of generative AI pilots at companies are failing. Fortune, 18 Aug 2025. Accessed 8 Nov 2026.
- Telstra connects faster to what’s next. Accenture. Accessed 8 Nov 2026.
- AI consulting and data engineering. Addepto. Accessed 8 Nov 2026.
- Pricing. OpenLazarus. Accessed 8 Nov 2026.
- Artificial intelligence. Slalom. Accessed 8 Nov 2026.
- Enterprise AI services. Thoughtworks. Accessed 8 Nov 2026.
- Cutting call wait times in half with change management and AI. Slalom. Accessed 8 Nov 2026.
- AI Risk Management Framework. NIST. Accessed 8 Nov 2026.
- AI consulting services. LeewayHertz. Accessed 8 Nov 2026.
- AI Architects. OpenLazarus. Accessed 8 Nov 2026.
Updated 8 Nov 2026: First published.