Every agency has a list of tasks someone thinks AI could do. The list is usually long, enthusiastic and in no particular order. The result is a handful of half-finished experiments and a team that's quietly sceptical about the next one.
The fix isn't more tools. It's choosing better. Before anyone builds anything, it's worth knowing where your team's hours actually go, and which of those hours AI can give back safely.
Look for repetition with rules
The work that pays back fastest has three things in common: it happens every week, it follows rules someone could write down, and it's tedious enough that people put it off. Checking drafts against brand rules, building recurring reports, researching prospects and moving leads between tools all fit.
- It repeats: weekly or more often, across clients.
- It follows rules: a good employee could write the checklist.
- It's measurable: you can count the hours it takes today.
- It has a clear owner: someone who can approve the result.
Score on hours, risk and effort
For each candidate, three questions. How many hours a month does it take today? What's the risk if the AI gets it wrong, and can a person catch that before a client sees it? And how much effort is it to build in the tools you already use?
High hours, low risk and low effort go to the top. High risk isn't a reason to skip something, but it means the human checkpoint matters more and the build takes longer.
Know what to leave alone
Some work isn't worth automating. One-off projects, tasks that change every time, and work where the judgement is the whole job usually cost more to automate than they save. Strategy, relationships and genuinely creative work are where your people should be spending the hours AI gives back.
The best first automation is rarely the most impressive one. It's the boring, weekly task that everyone is relieved to hand over.
Start with one process, measure it against your own numbers, and only move to the next when the first has paid for itself. It's slower than a big launch, and it's the approach that actually sticks.