When is a large consultancy the better choice for AI work?

When the hard part is organizational rather than technical. Multi-country rollouts, programmes touching dozens of business units, work that needs sustained change management, or engagements where procurement requires a supplier that can carry significant contractual liability. Scale and process are real capabilities, and a small firm cannot substitute for them.

When is a specialist firm the better choice?

When the hard part is technical and specific: private deployment, retrieval architecture, evaluation design, or a compliance position that has to hold up under scrutiny. Specialists are also usually better where you want the team that scoped the work to be the team that builds it.

What is the main risk with a large consultancy on AI work?

Distance between the people who sold the engagement and the people who deliver it. The named experts in the pitch are frequently not the ones on the project. Ask who specifically will do the work, how much of their time you get, and what happens to the engagement if they leave.

What is the main risk with a specialist firm?

Capacity and continuity. A small firm can be excellent and still be unable to staff a programme across six countries, and key-person dependency is real. Ask what happens if your lead is unavailable, and what you would be left holding if the relationship ended tomorrow.

What questions separate good providers from bad ones regardless of size?

What will you hand over that works without you, how do you measure whether the system is correct, who owns the models and prompts afterwards, and what would you tell us not to build. The last one is the most revealing, because a provider unwilling to talk you out of anything is selling capacity rather than judgement.

Big Four vs Specialist AI Consulting: Which Fits Your Programme

This comparison is usually made badly, in both directions. Large consultancies are dismissed as expensive and slow, specialists as risky and unscalable. Both caricatures are wrong often enough to be useless as a decision rule.

The better question is what the hard part of your programme actually is.

Where large consultancies genuinely win

Organizational scale. A programme touching thirty business units across several countries is a coordination problem before it is a technical one. Running that is a real capability, and it is one most specialist firms simply cannot supply.

Change management. Getting several thousand people to work differently is harder than building the system they will use. Large firms have done it repeatedly and have the method to show for it.

Procurement fit. Some organizations cannot contract with a small supplier at all, because of liability requirements, insurance thresholds or vendor policy. That is a constraint rather than a preference, and it settles the question.

Breadth of adjacent work. If the AI programme sits inside a wider transformation touching process, operating model and systems, a firm already holding that context has an advantage.

Where specialists genuinely win

Technical depth on a specific problem. Private deployment, retrieval architecture, evaluation design and adversarial testing are narrow specialisms. The people who do them well tend to do them repeatedly and not much else.

The team you meet is the team you get. This is the most common complaint about large-firm engagements, and it is structural rather than malicious: the senior people who scope work are a scarce resource spread across many accounts.

Willingness to say no. A specialist whose reputation rests on outcomes has a direct incentive to talk you out of work that will not succeed. A firm compensated on utilisation has a weaker one.

Speed on a bounded problem. For a defined technical question, a small team with authority to decide moves considerably faster than a governance structure designed for large programmes.

What actually goes wrong, in both cases

The failure modes are more informative than the strengths.

With large firms, the recurring problem is the gap between pitch and delivery: a strategy deliverable that reads well and produces no working system, staffed by people considerably more junior than those who presented. The tell during procurement is a proposal heavy on methodology and light on who specifically does the work.

With specialists, the recurring problems are capacity and key-person dependency. An excellent three-person firm can be unavailable when you need to scale, and a single departure can hollow out the engagement. The tell is reluctance to answer what happens if the lead is unavailable.

With both, the most expensive failure is the same: an engagement that ends with a system nobody internally can operate. That is a scoping failure rather than a firm-size failure, and it is entirely preventable by making handover an explicit deliverable rather than an assumed one.

The pattern that works for most large organizations

Not one or the other, but matched to the phase.

A specialist for the technical core: deployment architecture, retrieval design, evaluation, security review, and the compliance position. These are the decisions that are expensive to get wrong and difficult to unwind.

A large firm, or your own programme function, for organizational scale: rollout, change management, training, and coordination across business units.

The interface between them needs owning explicitly, by you rather than by either supplier, or you will pay twice for the seam.

The questions that expose the difference

Ask every provider, regardless of size:

Who specifically will do this work, and how much of their time do we get?
What will you hand over that continues to work without you?
How will you measure whether the system is correct, and who agrees the threshold?
Who owns the models, prompts and evaluation sets afterwards?
What would you tell us not to build?
What happens to this engagement if your lead leaves?

Question five is the most revealing. A provider with no answer is selling capacity. A provider with a specific answer has thought about your problem rather than their pipeline.

Where to go next

The readiness playbook covers what a first engagement should be scoped to produce, and what enterprise AI consulting costs covers the commercial side. For the build side of the same decision see build vs buy.