AISynq
Buying AI help4 min read

How to brief an AI project so the quotes are comparable

Five things to write down before you contact anyone, and why a vague brief produces three quotes that differ threefold for reasons nobody can explain.

Three firms quoting the same vague brief will come back with figures that differ by a factor of three, and the difference is almost never margin. It is that each one assumed a different scope, a different definition of done and a different owner for integration. Writing down the job, the systems, what done means, who is accountable and the budget range turns three incomparable numbers into three answers to the same question.

You send the same paragraph to three firms. One comes back at eighteen thousand, one at forty-five, one at a hundred and ten. All three are quoting honestly.

The spread is not margin and it is not greed. Each firm read the paragraph, filled the gaps with its own assumptions, and priced what it imagined. You cannot compare the numbers because they are not answers to the same question.

Five things fix most of it, and writing them takes an afternoon.

1. The job, in the words of the person who does it

Not the solution. The job.

"We want an AI agent for support" is a solution and it invites three different builds. "When a customer emails about a delivery problem, Maya opens the order in the admin tool, checks the carrier site, writes a reply, and logs it in the CRM, about forty times a day" is a job, and every firm reading it is pricing the same thing.

Go and watch somebody do it before writing this down. The written process and the actual process are different in almost every company, and the gap is where the estimate goes wrong.

2. The systems it touches, and who controls access

List them. For each one say whether it has an API, whether anyone internally knows it, and how long access usually takes.

This is the single biggest driver of a build quote and the thing most briefs omit. A job touching two systems with clean interfaces is a different project from the same job touching a nineteen-year-old ERP that one contractor understands. A firm that does not know which it is will either pad the number or discover it in week three.

Say so if access takes six weeks. Everyone's plan is better for knowing.

3. What done means

Write the sentence that ends the project.

There is a version where done is a working demonstration. There is one where done is in production, used daily by the team, monitored, with somebody on your side who knows how to change it. These are not the same project and the gap between them is often half the cost.

Most disputes later are about this sentence. Writing it now is the cheapest insurance in the whole process.

4. The number it is supposed to move, and what it is today

Even a rough figure. "This takes about forty hours a week across three people" is enough.

Two things happen when you include it. The firm can tell you whether the project is worth doing, which some of them will. And you get a baseline, which is the only way anyone finds out afterwards whether it worked. A baseline taken after the work starts is not a baseline.

If you genuinely do not know the number, say that too. It is a legitimate answer and it tells a good firm that the first piece of work is measurement.

5. The budget range

The advice everywhere is to withhold this so you do not get quoted up to it. In this market that advice costs you more than it saves.

Without a range, a firm guesses, and the guess is anchored on what it usually sells rather than on what you need. You get a hundred-thousand-pound proposal for a thirty-thousand-pound problem, and two weeks are gone.

A range also lets a firm tell you the honest thing. The best answer we give some people is that their budget does not buy what they are describing, and that the sensible version is smaller. Nobody can say that to a brief with no number in it.

What this does to the quotes

Three firms reading a brief like that will still differ, and now the differences mean something. One has excluded integration. One has assumed your team maintains it. One has carried a line for the second approach after the first fails.

Those are readable differences. You can ask about them. That is the whole point of the exercise.

One limit worth stating

A brief this specific has a failure mode worth naming. Written too tightly it forecloses the better idea, because the firm prices your solution rather than telling you the job should be done differently. The way through is to be specific about the job, the systems and the definition of done, and deliberately loose about the approach. If a brief specifies the architecture, you have hired yourself and are paying somebody else to type.

What to do with it

Send the same document to everyone on the same day, and ask the same questions of each. Eleven of them are here, and the four that matter most if you only have half an hour.

If you want the bands before you write anything, what AI consulting actually costs has ours published, which is unusual in this market. And if the brief turns out to be harder to write than expected, that is itself the finding: book the call and we will go through the job with you, free, before anybody quotes anything.

If you have a budget, a deadline, and no clear answer on which AI project deserves either, that is the conversation to have.

Book the call

Written by

Radwan Altaf

Radwan runs AISynq. Before that he delivered software inside enterprise programmes at DHL, AT&T, DirecTV and Accenture, which is where the habit of measuring a result against its baseline came from. More about the firm.

Work with us

Have a budget, a deadline, and no clear answer on which AI project deserves either?

Book the call

30 minutes. No deck.

Get the next one

New writing in Buying AI help as it goes up, roughly twice a month. One article per email and nothing else in it.