The short version: an AI brief is a one to two page document that says what problem you're solving, what data and tools you already have, what "done" looks like, and what you're not willing to change. Skip any one of those four things and you'll end up paying for a project that solves the wrong problem, or paying twice because the first version had to be scrapped. I've watched both happen, more than once, to businesses that had the budget and the motivation but not the paperwork.
Why nobody writes this down (and why that's costing you)
Most business owners come to AI the same way they came to their first website or their first Facebook ad campaign: someone recommends a tool, or a competitor mentions they're "using AI now," and there's a rush to get moving. Nobody sits down and writes a brief for a piece of software before they buy it. But AI isn't software in the normal sense. It's software that behaves differently depending on what you feed it, who's using it, and what you were trying to achieve, which is exactly why the brief matters more here than it ever did for a CRM or an email tool.
I did an AI audit for a client last year, a family-run recruitment agency in the Midlands with about 30 staff, and the founder told me point blank: "We spent £4,200 on a chatbot last spring and it's sitting there unused." When I asked what the brief for that project had said, there wasn't one. There was a Slack message to the agency that said "can you build us something like the one on that other recruiter's site." That's not a brief. That's a vibe. And vibes don't tell a developer what your candidates ask, what your CRM can and can't talk to, or who was supposed to maintain the thing after launch.
What a brief needs to contain
Forget templates that run to twelve pages with sections nobody fills in. A brief that works fits on two sides of A4 and answers six questions in plain English.
- The problem, in one sentence. Not "we want to use AI in our business" but "our customer service team spends 11 hours a week answering the same five questions about delivery times."
- The current process, honestly described. Who does this today, how long it takes, what tools touch it now (spreadsheet, CRM, email inbox, WhatsApp, whatever the messy reality is).
- What data you have. Is it in one system or scattered across four? Is it clean or does half of it live in someone's head? This single question kills more AI projects than any other, because most businesses find out too late that their data isn't in a shape any tool can use.
- What "done" looks like, with a number attached. Not "save time" but "cut response time from 4 hours to 20 minutes" or "reduce the quote-drafting task from 45 minutes to 10."
- What you won't change. Maybe you're not moving off your existing accounting software. Maybe the team of five customer service staff isn't shrinking to three. Say so upfront, because a consultant who doesn't know your constraints will design around the wrong ones.
- Budget and timeline, roughly. Not a precise figure, but a range. "Between £2,000 and £8,000, working weeks not months" tells an agency exactly which options to bring you and which to leave off the table.
That's it. Six sections. I've seen briefs this short get better results than the twelve-page versions, because the short ones force you to be specific rather than comprehensive.
A real example: the brief that worked
The same recruitment agency came back to me six months later, this time with a proper brief before we'd even had a call. It read something like this: "Our consultants spend around 6 hours a week each writing candidate summary emails from CV data. We have candidate data in Bullhorn and nowhere else. We want a first draft of each summary email generated automatically, reviewed and edited by the consultant before sending, cutting the writing time to under 90 seconds per candidate. We're not replacing Bullhorn. Budget is under £3,000, and we want something running within four weeks."
That brief took the founder about 40 minutes to write, most of it spent timing how long the task took using a stopwatch on her phone. It meant the whole scoping conversation with me took 25 minutes instead of the usual two calls and a follow-up email chasing basic details. We built a workflow using their existing data export and a prompt template, tested it against 20 real CVs, and it went live in three weeks at roughly £2,400. The chatbot that failed had cost more and delivered nothing. The email tool that worked cost less and paid for itself inside the first month, because six consultants each saving five hours a week is 30 hours of billable recruiter time back, every week.
Where most businesses get the brief wrong
The pattern I see most often, and it's uncomfortable to say because it puts some of the responsibility back on the business rather than the AI provider, is that people write briefs describing the tool they think they want rather than the problem they have. "We want a chatbot" is a solution, not a problem. "We want ChatGPT integrated into our website" is a solution, not a problem. A good consultant will push back and ask what's underneath that request, but plenty of agencies will happily quote you for the chatbot anyway, because it's an easier sale than a proper diagnosis, and because pushing back on a client's stated wants doesn't always feel like good client service, even when it is. If your brief names a specific tool before it names a specific problem, that's usually the first thing to fix.
The second most common mistake sits inside the data question. Business owners assume their systems talk to each other because they've never tried to make them. I've had briefs come to me listing "our CRM, our accounting software, and our booking system" as if the AI would simply know how to pull from all three. It won't, not without integration work that often costs more than the AI project itself. This is one of the common mistakes small businesses make with AI, and it's almost always because the brief skipped the unglamorous step of checking what's connected.
Step by step: writing your brief this week
- Pick one task, not a department. Not "improve marketing," but "writing the weekly customer newsletter" or "drafting responses to inbound quote requests." Broad briefs get broad, expensive, unfinished answers.
- Time it. Watch someone do the task for a week and note the hours, honestly. If nobody in the business can tell you how long a task takes, that's worth knowing before you spend a penny on AI for it.
- List every system that touches the task. Spreadsheets included. Especially spreadsheets, because they're where the real workflow usually lives even when the official system is something grander.
- Draft the "done" sentence. One measurable outcome. If you can't measure it, you can't tell if the project worked, and neither can whoever you hire.
- Write your no-go list. Systems you won't rip out, jobs you won't cut, data you won't hand over. Three lines, maximum.
- Set a budget range and a deadline. Then send the brief to two or three people, not one, so you can compare how differently they respond to the same document.
If you want a sanity check before you write any of this down, running through an AI readiness checklist first will tell you whether your basics (data, access, team buy-in) are sorted enough for a brief to be worth writing at all. There's no point briefing a project your systems can't yet support.
Using AI to help write the brief itself
There's a decent shortcut here that most people miss: use ChatGPT or Claude to draft the first version of your own brief. Describe the task in messy, spoken language, paste in a rough estimate of the hours it costs you, and ask the tool to turn that into the six-section structure above. It won't get the specifics right, because it doesn't know your business, but it will ask you the questions you hadn't thought to answer yourself, which is often the real value. I cover this approach in more detail in how to use ChatGPT for business development, and the same logic applies whether you're briefing an internal project or an outside consultant. If you'd rather have the same conversation with someone who does this for a living, that's exactly what an AI implementation coach is for: someone to sit across the table and ask the awkward questions before you've spent a penny.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
The uncomfortable bit nobody says out loud
Here's the part that agencies won't tell you and business owners don't want to hear: a vague brief is often more profitable for the person you're hiring. Open-ended scope means open-ended billing. "Build us something with AI for customer service" can turn into three months of workshops, discovery calls, and change requests, each one billed separately, none of it tied to the number you needed to hit. A tight brief with a specific number attached protects you from that, but it also makes the project easier to say no to for anyone who was hoping to sell you more than you need. Write the brief tightly anyway. If a consultant balks at a clear, specific ask, that tells you something useful about them before you've signed anything.
Content and design projects have the same trap, worth flagging if your brief covers anything customer-facing rather than internal process. The tools reviewed in AI writer and AI design tools for marketers can draft copy or graphics fast, but a brief that just says "make our marketing more AI" gives whoever's doing the work nothing to aim at. Same rule applies: name the piece of content, the audience, and what result you're expecting from it.
What happens after the brief is written
A finished brief isn't the end of the job, it's the start of a shorter, cheaper conversation. Send it to whoever you're considering, whether that's an in-house team member, a freelancer, or a consultant, and watch how they respond. Someone worth hiring will ask two or three sharp follow-up questions rather than sending back a generic proposal within the hour. If the response ignores your "done" sentence entirely and pivots straight to a bigger package than you asked for, that's your answer before you've paid anything.
Keep the brief itself once the project's live. Six months later, when you're wondering whether the thing you built is still doing its job, that original document with its specific numbers is the fastest way to check. Did the response time drop from 4 hours to 20 minutes? Did the newsletter task fall from 3 hours to 40 minutes? A brief without a number can't answer that question. A brief with one can, in about thirty seconds.
Frequently asked questions
How long should an AI brief be?
One to two pages. If it's running longer than that, you're probably describing the tool rather than the problem, or trying to cover more than one task at once. Split it into separate briefs instead.
Do I need a technical background to write one?
No. The strongest briefs I've seen come from people who described their work in plain, specific language, including the boring parts, rather than people who used technical terms they'd picked up from a sales page.
Should I write the brief before or after speaking to a consultant?
Before, always. A brief written after the conversation tends to reflect what the consultant wants to sell rather than what your business needs, even with good intentions on both sides.
What's the single most important thing to include?
A measurable "done" sentence with a number attached. Everything else in the brief supports that one line, and without it there's no way to tell later whether the project worked.
Related reading: AI Phone Agents for Small Business: What Happens When You Let AI Answer Your Calls and I Cut Client Onboarding From Three Weeks to Three Days Using AI. Here's What Broke Along the Way.
