It’s Q4 — How’s Your AI Plan Coming?

Most agencies spent 2026 experimenting with AI. Far fewer know what those experiments were for. Heading into 2027, that’s the gap worth closing.

A couple of weeks ago, I was talking with a friend who works at a web hosting company. He mentioned that they’ve had several clients come to them recently asking if they can fix the vibe-coded sites they built earlier this year.

Six months from now, I suspect his company will be getting calls to fix the Claude workflows those same clients are building today.

Six months after that? Who knows.

None of that is a knock on experimenting. Experimentation is how anyone learns what AI can actually do. But experimentation without a destination is just expensive tinkering — and the bill eventually comes due, usually to someone else.

That’s the situation most agencies are in right now. Lots of activity. Lots of tools. Very little that adds up to a clear AI strategy — and without a strategy, there’s no real way to budget for AI either. The budget is supposed to be what the strategy looks like once someone has to pay for it. Most agencies have neither.

So as we head into the last quarter of the year, I’ll ask: How’s your AI plan coming?

Your clients are experimenting too

If it’s any comfort, your clients are in the same spot — just with bigger checkbooks.

Gartner’s 2026 CMO Spend Survey found that marketing leaders are putting 15.3% of their budgets toward AI. Among the organizations Gartner considers AI-ready, that number climbs to 21.3%.

Here’s the part worth sitting with. Only 30% of those CMOs say their AI capabilities are mature. And 70% admit their internal marketing processes aren’t ready to put AI to work effectively.

The survey skews toward large companies, so your mid-size clients won’t map to it exactly. But the direction is hard to miss. Clients are spending ahead of their readiness. They’re buying the tools before they’ve figured out the workflows, the governance or the judgment calls. They are, in other words, building their own vibe-coded sites.

That’s either the biggest opening agencies have had in a decade or a serious exposure, and which one depends almost entirely on whether you’ve done that work yourself. A client whose processes aren’t ready doesn’t need another vendor with a demo. They need a partner who has already been through the mess and can show them the other side.

Six things to finish before the year ends

None of these require a consultant, a new platform or a board vote. They require a few hours and someone willing to look hard at what’s really happening.

1. Inventory what you’re already using. Pull every AI charge off the card statements. Seats, overlapping subscriptions, the account somebody in production expensed in March and never mentioned. It’s the clearest picture you’ll get of what your agency is really experimenting with — and most leaders are surprised by how much of it nobody uses.

2. Give every pilot a verdict. Scale it, fix it or kill it. “Ongoing” is not a verdict. A pilot that’s been running for nine months without a decision isn’t an experiment anymore. It’s a habit.

3. Measure three workflows. Pick one from delivery, one from new business and one from operations. Time them the way you do them today, then time them with AI in the loop. If you can’t show the difference, you can’t build a strategy around it.

4. Write the one-page policy. Which tools are approved. What client data can go where. What you disclose to clients and when. It doesn’t need to be perfect, but it needs to exist before a client with audit rights asks to see it.

5. Ask your top five clients about their 2027 AI plans. Before they bring it up with you. Find out what they’re trying to accomplish, what they’re planning to build themselves, and where they expect you to fit. Some of what you hear will be uncomfortable. Better to hear it in October than in a review meeting next spring.

6. Put a name on it. Not a committee, not a working group. One person, with real hours on their calendar, accountable for what happens next. AI that belongs to everyone ends up belonging to no one.

The strategy questions 2027 will ask

The to-do list gets you to January. These questions decide what kind of agency shows up on the other side of it.

What do you actually want AI to change? There are two ways AI can change your agency: doing the same things differently, or doing different things. Faster, cheaper production is the first. New services, new kinds of value, new reasons for clients to hire you is the second. Both are legitimate. But you have to pick on purpose, because they lead to very different investments.

Are you still being paid for hours? If AI makes your team 30% faster at a deliverable and you bill by the hour, you just gave yourself a 30% pay cut for getting better. The agencies that start pricing around outcomes in 2027 get to set the terms. The ones who wait will have their clients set the terms for them.

Where does your agency’s knowledge actually live? Your best thinking is spread across decks, Slack threads, old proposals and the heads of a few senior people. AI is only as good as what it can read, and right now most agencies are handing their tools a junk drawer. Getting that knowledge organized — with rules about what gets read for which job — is the least glamorous work on this list and one of the most valuable.

What does a junior person do now? The tasks that used to train your next generation of strategists are exactly the tasks AI is absorbing first. If you don’t redesign how people learn the craft, you’ll save money in 2027 and pay for it in 2030.

Where must a human stay in the loop? Not everywhere. Not all the time. But precisely where the decision carries weight — the strategic call, the client relationship, the moment of creative judgment. Decide where those lines are on purpose, and write them down, before your tools decide for you.

Then — and only then — the budget

Answer those questions and the budget mostly writes itself. Skip them and no spreadsheet will save you. You’ll end up with what most agencies have now: a list of subscriptions with an AI label on top.

Software is the easiest part of AI to count, which is why it’s usually the only part that gets counted. It’s also usually the smallest part of what it costs to make AI work. A real AI budget for 2027 accounts for at least five things.

Licenses and seats. The predictable part. Consolidate it based on what your inventory turned up.

Usage. More and more AI tools charge by consumption — tokens, credits, agent runs. That turns a one-time purchase into a monthly variable. Treat it like a utility. Set a cap, check it monthly, and know which workflows are driving it.

People’s time. Training, experimentation, the hours your AI owner spends doing the job you just gave them. This is the line most agencies leave out, and it’s the one that determines whether everything else pays off.

Knowledge and data work. Cleaning, organizing and maintaining what your AI reads. Unglamorous, and nearly impossible to skip.

Governance. Policy, legal review and whatever client-facing documentation your contracts now require.

Tie every dollar back to something in your strategy — a workflow you measured, a service you’re building, a question you answered. If a line can’t point to one, it’s not part of the plan. And revisit the whole thing quarterly. The models, the pricing and your team’s capabilities will all look different by April.

Experiments need a destination

Q4 isn’t the time to start an AI plan. It’s the time to find out whether you had one.

Keep experimenting. That part matters more than ever. But decide what you’re experimenting toward, so that a year from now you have something to build on instead of something to fix.

Somewhere, a hosting company is already getting ready for the next round of repair calls. Make sure your agency isn’t the one making them.