What Does an AI Consulting Company Actually Do?

Nobody agrees on what “AI consulting” means

Search “AI consulting company” and you’ll get a page of results that could not be more different from each other. One is a three-person agency building chatbots. Another is Accenture with 40,000 “AI practitioners.” A third is a former data scientist freelancing on Toptal. They all call themselves AI consultants. They all do completely different things.

This isn’t an accident. The term “AI consulting” is broad enough to cover strategy, implementation, training, staffing, and pure research. And the market rewards ambiguity, because a vague scope means a flexible sales pitch. When a client says “we need help with AI,” any of these firms can nod and say “that’s exactly what we do.”

The result is that most buyers have no framework for evaluating what they’re actually purchasing. They compare proposals that look similar on paper but describe fundamentally different services. One firm will audit your operations and build a working system. Another will produce a 90-page PDF with a maturity matrix and a recommendation to “invest in data infrastructure.” Both cost six figures. Only one leaves you with something that works.

The confusion has real consequences. A McKinsey survey from late 2024 found that 42% of companies that hired external AI help were “dissatisfied with the outcome.” Not because the work was technically poor, but because it wasn’t what they thought they were buying.

The four types of AI consulting, and what each actually delivers

Once you strip away the marketing language, AI consulting firms fall into four categories. Some overlap. Some pretend to be in a category they’re not. But the distinctions matter because they determine what you get at the end of the engagement.

Type 1: Strategy-only firms. These are the traditional consultancies (McKinsey, BCG, Bain) and the mid-market firms that mimic their model. They do workshops, stakeholder interviews, maturity assessments, and roadmaps. The deliverable is a document. Sometimes a very good document. But the document doesn’t build itself. You still need to hire or contract someone to execute. The gap between strategy and implementation is where roughly half of all AI initiatives stall, an execution gap that Harvard Business Review has tracked across multiple enterprise AI adoption studies.

Type 2: Implementation-only firms. These are dev shops and ML engineering agencies. You tell them what to build; they build it. If you know exactly what you need (“we need a document classification model trained on our insurance claims data”), they can execute. But if you’re unsure where AI fits in your business, they’re the wrong starting point. They build to spec. If the spec is wrong, the system is wrong. And spec-writing for AI is not the same as spec-writing for a website.

Type 3: Product companies disguised as consultants. This is a growing category and an easy trap. A firm has built a platform (a chatbot tool, an analytics dashboard, a workflow engine) and they sell “consulting” that always ends with you licensing their product. The discovery phase is real. The recommendation is predetermined. You’ll recognise this pattern when every conversation steers toward their proprietary solution, regardless of your actual problem.

Type 4: Full-cycle firms. These firms combine strategy and engineering under the same roof. They start with an assessment of your operations, identify the highest-value opportunities, design the solution, build it, and support it in production. The deliverable isn’t a deck or a handoff - it’s a working system. This is the model we use at gamgi, and it exists because we saw the gap between Type 1 and Type 2 eating our clients’ budgets alive.

What a real engagement looks like, week by week

Since the market is opaque, it helps to know what a credible AI consulting engagement actually involves. Not every firm follows this exact sequence, but the stages are consistent across the good ones.

Week 1-2: Operational audit. Before any talk of models or algorithms, the firm maps your current processes. Where does data flow? Where do humans do repetitive work? Where do errors cluster? Where is money or time leaking? This isn’t a generic assessment. A good audit involves sitting with the people who actually do the work, not just management. The output is a ranked list of opportunities with estimated value and feasibility for each.

Week 3-4: Solution design. The firm picks the highest-ROI opportunity from the audit and designs the technical approach. This means defining the data requirements, the system architecture, the integration points with existing tools, and the success metrics. The client signs off on a clear scope: what will be built, how long it takes, what it costs, and how success will be measured.

Week 5-10: Build and iterate. The system gets built. Not in isolation - in short cycles with regular client review. A document processing pipeline, a classification engine, an automated workflow, whatever was scoped. The client’s team is involved throughout, not because it’s a nice idea, but because the system has to work within their existing operations.

Week 11-12: Deployment and handover. The system goes into production. The firm provides documentation, training for the internal team, and a support period to catch edge cases. A good firm also defines the handoff criteria: at what point the client can maintain and evolve the system independently, or when to bring the consultants back for the next phase.

• A credible firm will audit before they propose solutions
• The engagement should have defined milestones and measurable outcomes
• Your team should be involved in every stage, not just the kickoff and the final presentation
• The deliverable should be a working system, not a set of slides
• There should be a clear handoff plan, not an indefinite retainer

Red flags to watch for. The firm can’t show you a previous system they’ve built. The proposal is generic and could apply to any company. The team that sells is different from the team that builds. The pricing is based on hours rather than outcomes. Every recommendation involves their proprietary platform. If you see three or more of these, you’re probably talking to a Type 3 firm or a Type 1 firm pretending to be a Type 4.

The EU AI Act, which entered enforcement in phases from 2025, is also reshaping what responsible AI consulting looks like. Firms that understand compliance obligations (risk classification, transparency requirements, human oversight mandates) will increasingly separate themselves from the ones that just ship code. If your consultant hasn’t mentioned regulatory considerations by the second meeting, they’re not thinking about production. They’re thinking about demos.

The audit-then-build-then-handover sequence behind a Type 4 engagement is described in detail on the process page. The failure mode that produces a stalled deliverable rather than a working system is covered in from AI pilot to production, and that piece is the most useful read for any buyer trying to tell a Type 1 report from a Type 4 build before signing. A structured audit at the start is the cheapest way to confirm which type of engagement you actually need.

When the four-type taxonomy isn’t the right lens

The four-type split assumes the buyer is choosing between firms to deliver a defined outcome. Several legitimate situations sit outside that buying posture, where the taxonomy answers the wrong question.

• You actually need a strategy deck because you’re raising or pitching internally. If the deliverable is a board paper or an investor narrative, a Type 1 firm is correctly priced for what you need. Hiring a Type 4 to produce a system you can’t fund the rollout for is the worse mistake.
• You have a competent internal AI team and you’re buying a specific gap. A staff-augmentation engagement (one senior ML engineer for six months) doesn’t fit any of the four types cleanly. The right shortlist is freelance marketplaces and specialist talent firms, not consultancies.
• The problem is genuinely solved by a vertical SaaS product. If a market-leading product already covers 80% of the workflow at a tenth of a custom build’s cost, the buying decision is product evaluation plus integration, not consulting. A Type 3 firm carrying that product is actually the honest answer here, even though the article treats Type 3 as a trap.