Where AI Fits in a Marketing Team's Week

Sort the week by whether the answer is already decided or still being decided. Everything else about where AI belongs in a marketing team follows from that one test.

The question comes up in some form every time a team starts taking AI seriously: where does this actually fit, and how do we add it without making a mess of how we already work.

Most answers are lists of tools. Tools are the least durable part of the answer, because the list is wrong within a year. A way of sorting the work does not change when the tools do.

Start with the week, not the tool

The useful unit is a normal week of a marketing team's work. Not a project, not a campaign. A week, because a week is where the same tasks reappear on a predictable schedule, and repetition is the thing that makes automation worth building at all.

Write down what the team actually did last week. Not what the plan said. What consumed the hours. Most teams have never done this and are surprised by the result, which is usually that a large share of the week went to work nobody would describe as marketing.

That list is the map. Everything below is about sorting it.

The four kinds of work in a marketing week

1. Production. Turning a decided thing into its finished form. The post is planned, the offer is set, the listing exists, and now somebody writes the caption, cuts the video, formats the email. High volume, high repetition, and the quality bar is consistency rather than brilliance.

2. Coordination. Moving work between people. Status updates, handoffs, chasing an approval, reformatting something so the next person can use it, answering where a file is. This work produces nothing on its own and is invisible in every plan, and in most teams it is the largest single category.

3. Judgment. Deciding what should exist. What the team should be known for, which market to lean into, whether a message is true, whether a piece is good enough to publish under the team's name. Low volume, high consequence, and no repetition by definition.

4. Relationship. Anything where the point is that a specific human did it. The call, the note after a closing, the conversation that turns an inquiry into a client.

Almost every argument about AI in marketing is really a disagreement about which category a task belongs to.

Sort by whether the work is decided or undecided. Not by whether it is creative.

Decided work: the answer exists and is being retyped. Automate it

Undecided work: the answer is being determined. Leave it with a person

The test: is the answer decided or undecided?

Here is the rule that sorts the map, and it is not the one most teams use.

The common instinct is to sort by creativity: automate the boring work, protect the creative work. That instinct is wrong often enough to be dangerous. Plenty of creative-feeling work is pure retyping, and plenty of boring-feeling work involves a real decision.

The better test: has the answer already been determined, or is it being determined right now?

Decided work means the substance exists and someone is converting it into a format. The listing facts are known. The team's position on the market is settled. The email's argument was agreed in a meeting. Producing the artifact adds no new information. That is where AI belongs, and it belongs there regardless of how creative the output looks.

Undecided work means the answer does not exist yet and this activity is how it gets made. What should we say about this market shift. Is this claim true. Should we publish this at all. AI can inform that work and should not perform it, because a system that produces a confident answer to an undecided question is not saving time, it is removing a decision the team needed to make.

By that test, production is almost entirely decided work. Coordination is mostly decided. Judgment is undecided by definition. Relationship work is decided in substance but is the one category where the point is the human, so it stays with the human regardless.

Sort by whether the work is decided or undecided. Not by whether it is creative.

Decided work: the answer exists and is being retyped. Automate it

Undecided work: the answer is being determined. Leave it with a person

Where it should not go

Worth stating explicitly, because these are the failures that make teams abandon the whole effort.

Anything where being wrong is expensive and quiet. Numbers in public copy. Claims about performance. Compliance language. The failure mode of a generated claim is that it is plausible and unverified, which is exactly the kind of error that survives review.

Deciding what the team stands for. A system asked to determine positioning returns the average of its training data, and the average is by definition what everyone else already sounds like. Teams that let a tool answer this end up with copy that could belong to any competitor.

The first version of anything genuinely new. A new offer, a new market, a new argument. There is no pattern to work from, so what comes back is the closest adjacent thing, which is subtly wrong in ways that are hard to see until it is published.

Any step with no named owner. Not a tool problem, an organizational one, but it kills more implementations than the other three combined. A workflow nobody owns stops working within a quarter and nobody notices. That failure is common enough to deserve its own set of tests before you build anything.

What a sorted week looks like

Run the exercise and the shape is fairly consistent across teams.

The production block gets systematized first, because it is the highest volume decided work and the failure is visible. The coordination block turns out to be larger than anyone estimated, and much of it disappears once the production work stops requiring handoffs. Judgment work gets more time, which is the actual return: not that the team produces more, but that the people capable of deciding things spend more of the week deciding things. Relationship work is left alone deliberately.

The point of this sorting is not to maximize how much gets automated. It is to stop the two expensive mistakes, which are automating a decision and hand-producing something already decided. Most teams are making both at the same time, in different parts of the same week.

If you only do one thing with this: take last week's actual hours, mark each line decided or undecided, and look at how much of the undecided column went to people who had no time left to think. That gap is the whole argument, and it is usually more about how the work moves than about which tools are in the stack.