There are four parts to the cost of most AI work. Knowing which part you’re paying for makes any quote easier to judge.

1. Working out what to do

This is the diagnosis: time inside your business to find where AI would move a number you care about, and where it wouldn’t. It should be a fixed fee for a fixed period, with a written plan at the end that you own. If someone wants to start building before this step, ask what they’re basing the build on.

Ours is called Blueprint. It takes two weeks, and the price is on our strategy page.

2. Building it

The build is usually the largest single cost. What moves it is how many workflows are involved, how many of your existing systems it has to connect to, and how much control you need around it: approvals, spend limits, a record of every action. A proper diagnosis lets this be quoted against a written plan, rather than estimated in a first meeting.

3. Getting it used

Training, an operating rhythm and clear ownership. This is the part most often left out of a quote and most often the reason a system goes unused. If it isn’t in the proposal, ask who is doing it.

4. Running it

AI systems have running costs: the model providers, the hosting, the third-party tools. These can be modest. Our own fleet of agent roles runs for about £400 a month. But they carry on after the consultants leave, so you should know them before you start, and there should be hard caps so the bill can’t surprise you.

Questions to ask before you sign

  • Is the diagnosis a fixed fee, and do I own the plan if I don’t go on to build?
  • Is the build quoted against a written plan?
  • Is adoption and training in the price, or extra?
  • What are the running costs after handover, and who pays them?
  • What transfers to me, and what stays yours?

A good answer to the last question is specific: the client-specific code, prompts, dashboards and documentation should be yours, and the proposal should say so before work starts.