Your clients are asking about AI. How a small agency answers without overpromising.

It usually comes at the end of a call about something else. “Hey, while I’ve got you. What should we be doing about AI?” And there’s a half-second where you decide whether to be the person with an answer or the person who suddenly has another meeting.

Every small agency I know has had that half-second this year, and most owners feel underqualified to fill it. They run web shops, brand studios, dev teams. They did not sign up to be AI consultants, and the ones pitching their clients seem to have gotten a head start and a nicer deck. That half-second is one of the best openings your agency has had in years. You’re better placed to fill it than the people currently in your client’s inbox, and you can answer without promising something you’ll regret in week three.

The short version: answer like someone who knows their business, because you do. Your client doesn’t need an AI expert. They need somebody who already understands how their orders come in, where their team loses a day every week, and which of their processes is held together by one person and a spreadsheet, and who can say which of those is worth pointing an AI tool at first. You’ve had that access for years. Most of the people selling them AI have had it for one discovery call. So skip the tool list. Start from what you know about how their work flows, pick one small thing, and offer to run it as a Job with a beginning, an end, and something to measure.

Why this question lands on you

Your client’s inbox is full of AI promises from vendors who’ve never seen their operation. Their board or their spouse keeps asking what the plan is. A competitor posted something vague and triumphant on LinkedIn. And you’re the outside firm they already trust with their brand, their website, their campaigns, or their code.

That seat is a privileged one. You’ve seen how their approvals work in practice. You’ve watched a simple request bounce between three people for two weeks. You know which of their systems nobody wants to touch, and you know it because you were the one who had to touch it. That’s operating knowledge, and it’s the exact thing missing from every generic AI pitch they’re getting.

So the question isn’t “do you know AI.” It’s “you know us, can you help us think about this.” A small agency can answer that well, and truthfully, without pretending to be something it isn’t.

What they’re asking underneath

Under “what should we do about AI” there are usually three smaller questions.

Where would it help us? Not in theory. In our shop, with our people, this quarter.

What would it cost us in effort and disruption to find out? They’ve been burned by tools that took six months to roll out and got used by two people, one of whom has since left.

Who’s going to hold our hand? Because they don’t have a team for this, and they know it.

Answer those three plainly and you’ve done more than most consultants who show up with a deck. None of them is “which model is best.” Your client doesn’t care. They care whether Tuesday gets easier.

Answer from what you already know

Start with a short list you can write from memory: the four or five places in their operation where you’ve personally watched effort pile up. The weekly report someone assembles by hand. The intake form that gets re-typed into another system. The proposal that starts from a blank page every time. The customer question that gets answered the same way forty times a month by a person who has memorized the answer and would like to stop.

Then split the list, out loud. Some of those are good AI candidates: repetitive, text-heavy, low stakes if the first draft is wrong because a human checks it anyway. Some aren’t: anything with legal exposure, anything where a wrong answer reaches a customer unreviewed, anything nobody on their side has time to supervise.

“These two are worth trying. These two I’d leave alone for now, and here’s why.” That sentence builds more trust than any demo, because it’s the first time someone selling them AI has told them what not to do.

If your own shop has a written stance on how you use these tools, bring it. Clients find it reassuring that you’ve thought about it for yourselves first. If you don’t have one yet, our guide to writing your agency’s AI policy is the short version.

Pick one small first project

Resist the roadmap. A twelve-month AI transformation plan is how you overpromise without meaning to. Nobody can estimate twelve months of unfamiliar work, and the client will remember the plan long after everyone’s forgotten why it slipped.

Pick one process from the “worth trying” list instead. Scope it as a Job with four Deliverables: set it up, run it alongside the current way for a few weeks, measure what changed, and decide together whether to keep it. Give it a start date and an end date. Name who on their side has to show up, because someone does, and if nobody can, you’ve learned something about their readiness before it cost either of you anything.

Estimating this kind of work is hard for the obvious reason that you haven’t done it before, so don’t pretend. Give a range, name the unknowns, build in a checkpoint. The approach we use for estimating effort on unfamiliar work applies here without a single change.

Deliver it like any other Job

The mistake small agencies make with client AI work is treating it as a favor, a side conversation, or “consulting” that never gets planned. Then it drifts. The client asks for one more thing, the setup takes longer because their data is messier than anyone admitted (it is always messier than anyone admitted), and three months later there’s no result and a pile of effort nobody logged.

Run it the way you run your best client work. Plan the Deliverables. Track the effort against each one, yours and theirs. When the scope moves, write it down and re-plan instead of absorbing it. Report progress the way you’d report on a website build: here’s what’s done, here’s what it took, here’s what’s next.

This matters twice. Once because it protects your shop. And once because it’s the lesson your client needs most: AI work is still work, it has effort and a timeline, and it goes off course when nobody’s watching. Showing them that on a small Job is worth more than any tool you could hand them.

The Bottom Line

You’re not selling AI. You’re offering to be the outside firm that helps a client move forward carefully, using what you already know about how their business runs. That’s a bigger role than “the agency that does our website,” and most of your competitors can’t play it, because they don’t have your seat.

Start small. Be plain about what you don’t know. Plan it like a Job and track it like one. Let the result speak. The clients who ask you this question are handing you the relationship. It would be rude not to take it.

FAQ

What if I don’t feel qualified to advise on AI at all?

You’re qualified to advise on their operation, which is the hard part. Pair that with a modest first project and a clear “I’m learning this alongside you.” Clients trust that more than false confidence, and they’ve had plenty of the latter lately.

Should I charge for the first project or do it free to keep the client?

Charge for it. Free work gets rushed or expands on its own, and it teaches the client that AI work has no effort in it, which is the opposite of the truth. Keep the scope small enough that the number is easy to say yes to.

What if the first project doesn’t work?

Then you’ve saved them from a much larger version of the same failure, and you found out together, on purpose, for a small amount of effort. Say that plainly. Then pick the next candidate from the list.