Three labels, three different jobs. An AI consultant works out where AI creates value in your business and what to do first. An AI implementation partner rolls out the tools and integrates them. An AI development company builds the systems that do not exist off the shelf. Most businesses need all three over a year, roughly in that order, and most disappointments come from hiring one to do another’s job.
The consultant
What they do: leadership awareness, discovery across functions, prioritisation into use cases, governance, a roadmap. What they should produce: artefacts you own, every idea captured, every use case scored, a readiness-sequenced plan with a first quarter. What to watch for: a consultant who recommends build or buy in every case regardless, who cannot show a method, or whose roadmap has no first quarter you could start on Monday.
Hiring a consultant to build is the first mistake. Slides do not run tickets.
The implementation partner
What they do: select and roll out the enterprise tools, Copilot, Claude, ChatGPT Enterprise, with the licence terms, data classification, policy, phased access and enablement that make the rollout stick; integrate products into your systems. What they should produce: a governed rollout with usage and time-saved measured. What to watch for: reseller margin pulling the selection, or a rollout that is a licence and an email.
Hiring an implementation partner to decide strategy is the second mistake. They will recommend what they implement.
The development company
What they do: build tailored systems, support agents, knowledge assistants, workflow agents, document intelligence, where no product fits. What they should produce: your code, an evaluation set, a production pilot, and a handover or managed service, all priced before the build. What to watch for: a demo without an evaluation set, a quote without the production path, or a build of something the market already sells.
Hiring a development company without discovery is the third mistake. They will build the exciting idea, which is usually the data-gated one.
Why the order matters
| Stage | Who | Output |
|---|---|---|
| Where is the value? | Consultant | Scored use cases, roadmap, governance, first quarter |
| Switch on what is ready | Implementation partner | Governed rollout, measured |
| Build what does not exist | Development company | Tailored system, evaluated, in production |
The roadmap tells the implementation partner what to roll out and the development company what to build, and says plainly which use cases should be bought rather than built. Without it, both are guessing, expensively.
Why one firm doing all three works, if it is honest
A firm that only consults has no reason to say “buy this, it is cheaper”; a firm that only builds has no reason to say “wait, the data is not ready”. A firm that does both can afford to tell the truth in the roadmap because it is paid either way, provided it publishes its method, takes no reseller margin, and says in writing which use cases are which. That is what we try to be. The three jobs are described under services, and the ten questions to ask any of them are in ten questions to ask an AI consultancy.