AI marketing sense: a working definition
A working definition of AI marketing sense — what it means, why most teams don't have it yet, and how to start building it in 90 days.
There's a phrase you hear in every AI marketing pitch right now: AI marketing sense. The phrase is doing a lot of work, often for tools that haven't earned the word sense. So what does it actually mean — and how do you know whether you're building it, or just buying more software?
Why most marketing teams don't have it yet
The instinctive answer is skill gap — not enough prompt engineers, not enough data scientists. That's misleading. The shortfall isn't headcount. It's the absence of a feedback loop. Most teams plug an LLM into one or two workflows, watch them get noticeably faster, and stop there. There is no record of what changed and why.
Sense, by contrast, is something a team accumulates. It needs cycles, and it needs to know which cycles to remember. As Peter Drucker famously argued, what gets measured gets managed — but AI work compounds an order of magnitude faster than the work he had in mind, so what gets instrumented gets internalised.
What 'sense' actually means
The closest analogue isn't expertise — expertise is depth in a domain. Sense is closer to taste: the intuition that lets a senior operator look at three options and know which one to ship. In a marketing context, that intuition has to span positioning, audience, channel, creative, and measurement — five surfaces where AI now competes for influence.
The good marketer used to ask: what does the customer want? The marketer with sense asks: which of these choices belongs to me, and which belongs to the model?
Two skills sit underneath that, and they pull in different directions.
Judgement about delegation
Knowing which choices to keep human is half of sense. Brand voice, ICP refinement, and one-off launches all need a human in the loop because they shape the surface area that everything else runs on. Subject lines, ad variants, and audience segmentation can run on models — provided someone reviews the aggregate at the end of the week. Get the line wrong and you either over-delegate (and dilute the brand) or under-delegate (and waste the leverage).
Discipline about instrumentation
The other half is keeping the receipts. Every delegated decision needs a record of what the model saw, what it chose, and what happened next. Without that, the team can't tell whether a campaign worked because of the model or in spite of it. Most teams skip this step and end up with confident anecdotes instead of compounding intuition.
How to build it in 90 days
Sense is a habit, not a vendor. The compact version of the plan:
- Weeks 1–2: log every AI-assisted decision the team makes. One Notion table, one row per decision, three columns: input, output, outcome.
- Weeks 3–8: review the log every Friday. Tag each row kept (model was right), corrected (model was almost right), or reversed (model was wrong). The ratio is your delegation map.
- Weeks 9–12: move the kept decisions into pipelines. Move the reversed ones back to humans. Leave the corrected ones in the workflow with a checkpoint.
By the end of the quarter you'll know three things you didn't know before:
- Which decisions the model gets right unsupervised.
- Which decisions the model can handle with a five-minute review.
- Which decisions you should never have delegated in the first place.
The 'teammate, not tool' frame
Most teams operate AI like they'd operate a CMS — feature requests, prompt libraries, training docs. That framing works for software. It fails for systems that learn from feedback.
Treating AI as a teammate is more uncomfortable and more accurate. You onboard a teammate by giving them context, watching them work, correcting their early calls, and trusting them with more over time. You also document their wins and losses, because that's how the team gets better. Sense is just that documentation, applied to a teammate that improves faster than any human you've ever hired.
The teams that win the next five years won't have the most prompts. They'll have the longest, best-tagged log of decisions — and the judgement to know which ones to keep delegating. That's sense. Build the muscle, then turn up the volume.
Frequently asked questions
No. AI literacy is the ability to use the tools. Sense is the judgement to know which problems they should be solving — and the discipline to remember the answers.
No. The instrumentation half is closer to journaling than to data science. A Notion table, a Friday review, and a willingness to tag your own decisions honestly is enough for the first 90 days.
Strategy is the plan. Sense is the muscle that lets the plan survive contact with real users. A team can have an AI strategy without sense — and most do — but the strategy won't compound until the muscle exists.
Over-delegating before they've instrumented. Without a log of what the model decided, you can't tell whether scaling up is helping or hurting. Instrument first, then scale.