How to Brief an AI Agency So the Quote Means Something

You ask three AI agencies to quote on "automating our customer enquiries". One comes back with a figure that looks low. One is double that. The third wants a call first. You cannot tell whether the gap is skill, scope or guesswork.

Most of the time it is the brief. A vague request forces each agency to fill the gaps with its own assumptions, and each one assumes something different. The low quote may leave out the part you care about. The high one may include a custom build you did not need.

A good brief takes about an hour to write and is the cheapest way to make quotes comparable. This guide covers what to put in it, using the questions we ask on the Concierge as a frame.

Quick answer: a brief needs five things. The job and what makes it slow today. The tools you already run. What done looks like, and which of cost, speed or quality matters most. A budget split into setup and monthly running costs. And what data the work touches and who checks the output. With those, an agency can quote the same job as its competitors, and you can compare the numbers.

1. The job, in your words

Describe the problem, not the solution. "Customers email us the same five questions all day" is a better brief than "we need a chatbot". The first lets an agency propose a chatbot, a help page, an inbox rule or a person. The second locks in an answer you may not need.

Include what is slow or costly today, and any numbers you already know: how many enquiries a week, how long a quote takes, who does it. You do not need to measure anything new. Rough counts are enough, and say they are rough.

The Concierge asks for this in a few sentences, and you can pick up to three areas of help, such as answering customers or lead follow-up. Keep your own brief to one job. A brief that covers six jobs gets a quote for a programme, not a project.

This matters more than it looks. When a project misses, the cause is often that the people building it and the people paying for it understood the goal differently. A brief that states the problem plainly is the first defence.

2. The tools you already run

This is the part most briefs leave out, and it changes the price more than anything else. An agency that knows you run a CRM, an automation tool and a support inbox can build on those. One that does not will either ask, or quote to build its own and charge for it.

List the tools, including the ones you are unhappy with. Four examples, from our reviews:

• A CRM such as Attio. The catch in our review: the free plan stops at a few seats, and automations and enrichment draw on monthly credits. If the job writes into your CRM, the agency needs to know which one and how many people use it.
• An automation tool such as Zapier or Make. If you already pay for one, say so. The catch with Zapier is that every successful action step counts toward your task allowance. With Make, each action uses credits and the free plan runs on a slow schedule. A new workflow built on top of yours uses your allowance, and the agency should tell you roughly how much. Our comparison of the two and our guide to Zapier, Make and n8n cover the difference.
• A support tool such as Fin. The catch in our review: it is billed per outcome, a conversation can count as resolved when the customer simply leaves, and seats, minimums and add-ons come on top. If the agency is going to connect to it, ask them to estimate the monthly cost at your current volume.
• Your office suite and accounting package. The Concierge lists common ones, from Google Workspace and Microsoft 365 to Xero, MYOB and Shopify. Name yours.

Where you do not know whether a tool can connect, say that too. "We use X and have not checked" is a useful line in a brief.

3. What done looks like, and what matters most

"Working" means different things to different people. Write down what you would see if it worked: the routine questions answered without anyone touching them, quotes out the same day, invoices matched without a spreadsheet.

Then pick one priority. The Concierge offers six: revenue, cost reduction, speed to launch, quality, compliance and customer experience. You can care about all six, but an agency has to make trade-offs, and it helps to know which way you lean. A team told "speed to launch" will start narrow and ship. A team told "quality" will test longer.

Say how you want to work as well. The Concierge asks whether this is a one-off project, ongoing support, a pilot first, or advice only. A pilot first is often the right answer for a first project, since it limits what you can lose and shows you what the agency is like to work with before you commit to more.

4. Budget, in two parts

Give a setup budget and a monthly budget separately. They are different costs and agencies price them differently.

Setup is the one-off cost to get the thing built and running. The monthly figure is what it costs to keep running: support, monitoring, and the subscriptions for the tools underneath. The Concierge asks for both because a project with a modest setup cost can still carry a monthly bill that surprises you. Our guide to what AI agencies charge explains how the models differ, and our piece on hidden costs covers what usually gets missed.

If you do not know your budget, say "not sure yet" and ask the agency to quote two options at different sizes. That tells you more than a single figure. Giving a budget does not mean the agency will spend it. A range lets them say whether the job fits.

Add your timeline. "Within a month" and "flexible" produce different plans and different prices.

5. The data, and who checks the work

Say what information the project will touch. Customer names and emails, payment details, health or financial records, staff data. The Concierge asks whether the work will handle personal, financial or health information,.

Then say who checks the output. This is the question that separates a project that runs from one that quietly goes wrong. Who reads the drafts the AI writes? What happens when it is unsure? Who can switch it off?

The Australian Government's Guidance for AI Adoption, which replaced the Voluntary AI Safety Standard in October 2025, sets out six practices for organisations using AI. We read a summary of it from the AI Governance Library rather than the government's own page, which blocked our access. According to that summary, the practices include naming accountable people, testing and monitoring, and maintaining human control, and the guidance says deployers should ask suppliers for test methods and results. That is a fair list to put to an agency: who is accountable on their side, how will they test it, and what do you receive at the end to show it works?

The rules on personal information differ by situation. For anything touching it, check the Office of the Australian Information Commissioner and get advice that fits your business.

What a good brief produces

When the brief covers these five things, the quotes you get back can be read side by side. You will see which agencies assumed they would build everything new, which are building on what you have, and which are quoting a one-off versus an ongoing arrangement. If a quote ignores something in your brief, you have a precise question to ask.

It also tells you about the agency. A good one will push back on parts of your brief. It may tell you the job is smaller than you thought, or that a tool you already pay for can do it. Our guide to vetting an agency covers more of what to look for, and the five readiness checks are worth finishing first.

Where to start

1. Write the brief. One job, your tools, what done looks like, a split budget, and the data involved. An hour is enough.
2. Use the Concierge if you want help matching. Send the brief and a person on our team chooses a reviewed agency that fits the task, budget and industry. It is free for businesses, and you decide whether to talk, ask for a quote or walk away.
3. Or browse the agencies yourself. The agencies page has the agencies on Find AI Now, and the same brief works for each one.