Where to Start with AI in Your Marketing Team

    The honest answer is not "buy a tool". It is the answer nobody selling software wants you to hear, because the tool is the easy thing to sell and the hard thing to make pay back. A tool is a solution looking for a problem you have not named yet. The teams that get a real return start the other way round: they find one painful, repeated process, prove a machine can take the grind off it, and only then bring in whatever software the job actually needs. This page is the team-level version of where to begin, a deeper companion to our pillar on AI for marketing agencies. It covers how to choose your first win, a thirty-day sequence to get there, and the false starts that burn the budget before anything ships.

    The first-win criteria

    A good first win has three things at once: it is painful, it is frequent, and it is rule-based. Miss any one of the three and the project tends to stall, so it is worth being strict about all three before you commit a single hour to building.

    Painful means the work genuinely hurts. People dread it, it eats real time, it pushes the good people off the work clients actually pay for. If nobody complains about a task, fixing it wins you nothing anyone will notice, and a saving nobody feels never gets funded a second time. Frequent means it happens again and again, weekly at least, ideally daily. A painful job that runs twice a year is not where you start, because the effort to automate it never earns back against how rarely it runs. The reporting grind, the prospecting research, the status updates: those repeat, which is what makes them worth the build. Rule-based means the work follows a pattern a machine can learn, the same steps in the same order with judgement only at the edges. If every run is a fresh creative call, a model will fumble it. If the spine of the job is "pull this, shape it like that, send it there", that is the shape AI is good at.

    Turn it into a quick-scan you can run in a meeting. Take a process and score it one to five on each of the three: how painful, how frequent, how rule-based. A process that scores four or five on all three is a strong first win. A process that scores high on pain but low on frequency is a tempting trap, because it feels urgent but never repeats enough to pay back. Anything that scores low on rule-based belongs to a person for now, not a machine. Run the scan across five or six candidates and the right place to start usually stops being a debate. If you want a more structured version of the same check, the AI Necessity Test walks a single use case through it in about eight minutes.

    A 30-day starting sequence

    You do not need a transformation programme to start. You need four focused weeks and the discipline to finish one thing. The sequence below is deliberately small, and each week earns the next. Notice that no week says "buy software". The tool choice comes after the process is mapped and the win is picked, never before, because a tool bought against a process you have not measured is how the spend gets stranded.

    1. Week one: map one process end to end. Pick a single process the team grumbles about and write down every step it actually takes, not the tidy version in the handbook. Who touches it, which tool they open, where they wait, where they copy-paste, where a number gets re-keyed. One process, one page. The point is to see the real shape of the work before you change any of it, because you cannot fix what you have not drawn. A lot of teams skip this and start buying, which is exactly how the hours stay hidden.
    2. Week two: measure the baseline honestly. Put numbers on the map. How long does the process take, how often does it run, and how many people does it pass through in a week? Time three or four real runs rather than guessing. A rough true number beats a precise invented one. By the end of the week you should be able to say a sentence like 'this eats six hours a week across three people', because that sentence is what tells you whether the work is worth automating at all.
    3. Week three: pick the win against the criteria. Now score the candidates. The first win wants to be painful, frequent, and rule-based all at once, the three boxes from the criteria above. Rank the processes you mapped against those three, pick the one that scores highest, and ignore the rest for now. One win, not a programme. The discipline here is saying no to the four shinier ideas so the team can actually finish the one that pays back.
    4. Week four: scope the pilot small and measurable. Write down what the pilot will do, what it will leave alone, and the one number that proves it worked. Keep the scope embarrassingly small: one process, one team, one measurable before-and-after. Decide who owns it and what 'done' looks like, then stop. A tight scope you can finish in a quarter beats a grand plan that never ships, and a measured result is what earns the budget for the next build.

    By the end of the month you have a mapped process, an honest baseline, one chosen win, and a scope tight enough to ship in a quarter. That is a far better starting position than a tool subscription and a vague hope. It is also the order our whole method runs in: find the waste, measure it, fix one thing, prove it, then move on. The thirty days are the on-ramp, not the whole journey, but they are the part that decides whether the journey pays back.

    What not to do first

    Three false starts swallow more agency budgets than any technical failure, and all three share the same root: they start with a tool instead of a process. The first is bolting a chatbot onto the website because it feels like "doing AI". It looks modern and it almost never touches the work that is actually draining the team, so it ships, it impresses nobody, and the real grind carries on untouched. The second is building an AI content factory, pointing a model at the blog and the social channels and letting it churn. You get volume, you lose the voice clients pay for, and a lot of that output quietly goes unread. The third, and the most expensive, is tool shopping: buying three or four platforms on the strength of a demo, before anyone has mapped a process or named a problem. That is how the stack fills with software nobody opens, which is its own slow, costly mess. We took that one apart in the marketing tool graveyard and how to stop adding to it.

    The thread running through all three is starting with the answer before you have asked the question. A chatbot, a content engine, a new platform: each is a solution in search of a problem. Flip it. Start with the painful, frequent, rule-based process, and let the problem tell you what to build. That single change in order is most of the difference between AI that pays back and AI that becomes next year's line item nobody can explain. MIT's NANDA study found that up to 95% of generative AI pilots return no measurable financial value [MIT NANDA, 2025], and the false starts above are where a lot of that money goes.

    Where to start

    You do not have to commit budget to get a real read on this. Two next steps, in order of cost.

    References

    • MIT NANDA. "The GenAI Divide: State of AI in Business 2025." MIT Media Lab Project NANDA, 2025. Source of the figure that up to 95% of generative AI pilots return no measurable financial value.