Everybody in the Agency Is Using AI. Almost Nobody Has Agreed What That Means.

I have sat on all three sides of this table. Agency side at Insight Publicis, where I left as Head of Digital Business. Client side at Jumia, buying and running advertising. Platform side at Meta and now at Aleph, selling across Sub Saharan Africa. From every one of those seats, the same thing is happening right now. Everybody in the agency is using AI. Almost nobody has agreed what that means. ## An agency holds other people's secrets This is the part that makes advertising different from most sectors, and it is the part I rarely hear discussed. A bank that mishandles AI exposes its own data. An agency that mishandles AI exposes its clients. Unreleased campaigns. Pricing and commercial terms. Product launches under embargo. Research about a client's customers that the client paid for. Competitive analysis naming brands that are not in the room. When a planner pastes a client brief into a free chatbot to get a faster start, that brief belongs to somebody else. When a strategist drops a competitive review into a tool to summarise it, they have handed another company's market position to a system nobody in the building controls. I call it BROAI. Bring Your Own AI, after the Bring Your Own Device problem that worried corporate IT a decade ago. In agencies it is close to universal, because agency people are permanently short of time and generative tools are genuinely good at the first draft. The exposure is not theoretical. Most client contracts contain confidentiality clauses that were written before any of this existed, and they do not care that the intention was efficiency. ## Where AI actually pays in an agency The visible use is creative generation, and it is the least interesting. Worth understanding, easy to overrate. **The internal work nobody sees.** Status reports. Campaign reconciliation. Post campaign analysis assembled from four platforms. Timesheet summaries. Credentials decks rebuilt for the ninth time this quarter. This is where agency hours disappear, and none of it is client facing, which makes it both the safest and the most profitable place to start. **Media planning and optimisation.** Not replacing the planner. Giving the planner faster answers about what a mix is likely to deliver, so more of the week goes on judgement and less on pulling numbers. **Creative testing before spend.** Predicting which variants are likely to underperform is far cheaper than discovering it live. **New business qualification.** Most agencies pitch for work they should have declined. The cost of a lost pitch is real money and morale. Being more honest earlier about which briefs fit is worth more than winning another one. ## The commercial point most agencies miss An agency's costs are broadly fixed. Rent, salaries, subscriptions. They barely move whether you bill a lot or a little in a given month. So every hour AI returns to a senior person is an hour that either goes to a client, to a pitch, or to nothing. That choice, not the tool, decides whether any of this shows up in the numbers. I have watched agencies automate their reporting and then quietly absorb the time saved into more reporting. Decide in advance where the returned hours go. Otherwise you have bought software and changed nothing. ## What good looks like One approved tool set. A written rule about what client material may go into which system, communicated to everybody rather than filed. A named owner for each use case, carrying a number. SWITCH, the six phase framework I use with clients, opens with Scan and Score for exactly this reason. Most agencies overestimate how organised their data and processes already are. A question for your next management meeting. If a client asked you today, in writing, which AI tools have processed their confidential material this year, could you answer? ## Frequently asked questions **What is the biggest AI risk for an advertising agency?** Client confidential material going into tools the agency does not control. The agency carries the contractual exposure, and the client usually finds out at the worst possible moment. **Will AI replace media planners and strategists?** It removes the assembly work around their jobs. The judgement about what a brand should do, and the relationship that lets you say it, remain human. **Where should an agency start?** Internal reporting and campaign reconciliation. No client sees it, the hours saved are real, and a mistake costs an afternoon rather than a relationship. **Can AI be used on client work at all?** Yes, within tools the agency controls and with the client's knowledge. Many clients now ask the question directly during pitches, so having a clear answer is becoming a commercial advantage. **Does AI reduce agency headcount?** It changes the shape of the team more than the size of it. Less assembly, more thinking, and a higher expectation on the people who remain. *Stanislaus Martins has worked agency side at Insight Publicis, client side at Jumia, and platform side at Meta and Aleph, and has trained teams at Publicis, Dentsu, WPP, ADMARP and ADVAN. Formats are on the [speaking and training page](/speaking), or start a conversation on the [work with me page](/work-with-me).*