Most buyers evaluating AI consulting firms select on the wrong axes. Firm size, brand recognition, and hourly rate are observable, comparable, and almost completely uncorrelated with whether the engagement will produce a working system. The criteria that actually predict outcome — who owns the result, how risk is structured, whether each phase ships a usable artifact, and whether the firm will tell you when the project is infeasible — are harder to assess from a pitch deck. This piece is a decision framework for that harder assessment. We write this from the seller side of the table, which is the obvious bias to declare up front. The framework below is the one we want buyers to apply to us. If we fail it, we deserve to lose the engagement. What is AI consulting, really? AI consulting is sold as one thing and delivered as two. The advisory half frames a business goal in AI terms, sizes the data and infrastructure required, identifies realistic use cases, and prices the engineering effort. The delivery half builds the data pipelines, trains or selects the models, integrates with existing systems, and stands up evaluation and monitoring. Engagements that stop at advisory rarely produce lasting value, because the gap between “here is what you should build” and a system that actually runs in production is where most AI projects fail. The structural question is who absorbs that gap. In a staff-augmentation model, the consulting firm rents engineers who follow the buyer’s direction; the buyer carries the technical risk. In an outcome-ownership model, the firm commits to a result and absorbs the risk of the path to get there. These are different products sold under the same name, and the difference is what the rest of this article is about. The four-criterion evaluation framework The criteria below are the ones we think a procurement lead should apply when comparing AI consulting firms. Each is a question the firm should be able to answer concretely, in writing, before contracts are signed. Criterion What you are testing What a good answer looks like Outcome ownership Does the firm own the result, or just the hours? Named acceptance criteria, a defined deliverable, and a pricing structure that has the firm carrying delivery risk — not just billing time. Risk structure Are there explicit pivot points and milestone gates? A risk register inside the first 2–4 weeks, with named decision points where scope can be cut or redirected based on what discovery reveals. Intermediate value Does each phase produce a usable artifact? A written discovery output, an evaluation harness, a deployable prototype, a monitoring plan — each one independently useful if the engagement ends early. Honest assessment Will the firm tell you the project is infeasible? A documented case where the firm has recommended against proceeding, or scoped down significantly, after discovery showed the original plan would not work. The framework is self-serving only if the firm being evaluated actually delivers on all four. That is the point: it is hard to fake. How boutique AI consultants differ from Big Four firms The honest comparison is not “boutique good, Big Four bad” — both have legitimate roles. The difference is in what each is structurally optimised for. Big Four and tier-one consultancies are optimised for organisational change at scale: multi-year transformation programmes, regulatory and compliance work, executive-level alignment across business units. Their AI practices are typically strong on strategy, governance, and vendor selection, and weaker on hands-on model development. The engagement structure is usually time-and-materials with large named teams. Boutique AI consultancies are optimised for a narrower problem: shipping a specific AI capability in a defined window. The teams are smaller, the specialist depth in perception models, LLM evaluation, GPU performance, and MLOps is typically higher, and the engagement structure is more often fixed-scope or outcome-based. The trade-off is less reach into the rest of the organisation. The selection rule is straightforward. If the problem is “we need an AI strategy across twelve business units,” that is Big Four territory. If the problem is “we need this specific system in production in six months,” that is boutique territory. Buying the wrong shape of firm for the problem is one of the more expensive mistakes in this space, and brand recognition is what most often causes it. What separates capable firms from rebranded ones A large share of firms now marketing AI consulting were doing something else two years ago — generic IT services, web development, RPA. Some have built genuine capability; some have rebranded the same staff augmentation under a new banner. The evidence that distinguishes them is specific and verifiable: Named production deployments, not just “case studies” that describe a workshop or a proof-of-concept that never shipped. Ask which models are in production, on what infrastructure, serving how many requests, and at what latency. Technical depth in conversation. A capable firm can discuss evaluation methodology, model failure modes, and the specific trade-offs in their architectural choices without retreating to generic language about “leveraging AI.” Engineering practices that survive contact with production: model evaluation harnesses, drift monitoring, rollback procedures, and a documented approach to data quality. These are unglamorous and rarely featured in marketing material, which is why their presence is informative. References who will discuss what went wrong, not just what went right. Every real engagement has failure modes; a firm that cannot describe theirs has not delivered enough to have learned from them. What AI consulting costs Price bands are an imperfect signal but a useful one for screening out mismatches. Day rates for senior AI consultants in 2026 range roughly £1,200–£2,500 / day in the UK and €1,400–€3,000 / day in continental Europe, with US engagements often higher. Project-based pricing for a typical 3–6 month delivery sits in the £80k–£500k range depending on team size and infrastructure scope. These are observed-pattern ranges drawn from the engagements we see in market, not a benchmarked rate card. Pricing materially below this range usually indicates a different product: junior contractors, off-the-shelf product integration, or a firm subsidising AI work to win other revenue. None of these are illegitimate, but they are not what most buyers think they are buying when they procure AI consulting. Pricing materially above the range can be justified — frontier model work, regulated industries, or genuinely scarce specialist skills (advanced CUDA optimisation, large-scale distributed training, formal verification for safety-critical systems). It should be justified explicitly, not assumed. Contractual structures and what they protect The contract shape determines who carries which risk. Three patterns dominate, and each is appropriate in different situations. Time-and-materials suits exploratory work where the scope cannot be defined up front — early discovery, research-grade investigations, or extensions to an existing engagement. The buyer carries scope risk; the firm carries no delivery risk. Engagements scoped to your problem, with defined acceptance criteria and a fixed price, suit work where the outcome can be specified before the engagement begins. The firm carries delivery risk; the buyer carries the risk of having specified the wrong outcome. This shape works when discovery has already happened. Outcome-based contracts, where payment depends on a measurable business result, suit a narrow set of engagements where the metric is clean, attributable, and within the firm’s control. They are rarer than the marketing language suggests, because most AI outcomes depend on factors the consulting firm cannot control. The relevant question is not which structure is best in the abstract but which one matches the work. A firm that only offers one shape is constrained either by capability or by risk appetite, and both are worth knowing. Handover capability The final criterion is whether the engagement creates dependency or capacity. A deliverable that only the consulting firm can operate is a liability disguised as an asset. The handover plan should exist from day one and should include documented runbooks, model evaluation procedures the buyer’s team can execute, infrastructure that uses standard tooling rather than firm-specific abstractions, and a defined period of supported transition. We pay close attention to this in our own engagements, because the engagements that produce lasting value are the ones the client’s team can run after we leave. If the engagement instead produces a system that requires the consulting firm’s continued involvement to operate, the buyer has not bought a capability — they have bought a subscription to the consulting firm. That may be the right trade in some cases, but it should be a conscious choice, not an accident of how the engagement was structured. What to ask before you sign The single most useful screening question is one the framework above implies: can you produce a risk-structured engagement plan for this work, with named acceptance criteria, milestone gates, and an honest assessment of where this project could fail? A firm that can produce one in writing has already demonstrated most of what the framework is testing. A firm that cannot has answered the question for you. For a deeper look at how the small-business end of this market plays out — where the price bands compress and the trade-offs shift — see our companion piece on realistic expectations for AI consulting in small businesses. For a sector-specific application of the same framework, AI consulting in real estate walks through how outcome ownership and risk structure look when the data and the use cases are domain-specific. Frequently asked questions What should I look for when evaluating AI consulting firms, and what should I screen out? Look for outcome ownership, explicit risk structure with milestone gates, intermediate deliverables that are independently useful, and a willingness to tell you when the project is infeasible. Screen out firms that lead with brand and team size rather than delivery evidence, that cannot produce a written risk register inside the first month, and that only offer one contract shape regardless of the work. How do boutique AI consultants differ from Big Four consulting firms in scope, methodology, and accountability? Big Four firms are optimised for organisational change at scale — strategy, governance, multi-business-unit alignment — typically under time-and-materials with large teams. Boutique AI consultancies are optimised for shipping specific capabilities in defined windows, with deeper specialist skills in model development, GPU performance, and MLOps, more often under fixed-scope or outcome-based contracts. The right choice depends on which problem you actually have. Which evidence genuinely separates capable firms from rebranded ones? Named production deployments rather than workshop case studies, technical depth in conversation about evaluation and failure modes, the presence of unglamorous engineering practices (evaluation harnesses, drift monitoring, rollback procedures), and references who will discuss what went wrong. Marketing language about “leveraging AI” is uncorrelated with capability. How much does an AI consultant cost, and what determines the price band for a serious engagement? Senior AI consultant day rates in 2026 are roughly £1,200–£2,500 in the UK and €1,400–€3,000 in continental Europe, with US engagements often higher. Project pricing for 3–6 month deliveries typically sits in the £80k–£500k range. Materially below that usually indicates a different product (junior contractors or off-the-shelf integration); materially above should be justified by frontier work, regulation, or genuinely scarce skills. Which contractual structures protect the buyer in AI work? Time-and-materials suits exploratory work where scope cannot be pinned down. Engagements scoped to your problem, with defined acceptance criteria, suit work where the outcome can be specified after discovery. Outcome-based contracts suit the narrow set of engagements where the success metric is clean and within the firm’s control. No single structure is universally best; the buyer protection comes from matching the structure to the actual work. How do I evaluate a consulting firm’s ability to hand off to my internal team rather than create dependency? Ask for the handover plan on day one. It should include runbooks your team can follow, evaluation procedures your team can execute, infrastructure built on standard tooling rather than firm-specific abstractions, and a defined supported-transition period. A deliverable only the consulting firm can operate is a subscription, not a capability transfer. Ask your consulting partner for a risk-structured engagement plan. If they can’t produce one, that’s your answer. Image by Freepik
