Trying to write the post I wanted to read
The current commentary on AI in marketing splits into two useless halves: it changes everything, or it produces generic slop and should be ignored.
Neither has been my experience. So I spent six weeks logging what I actually used it for, whether the output survived to a client, and roughly how long the task took before and after.
This is that log, with the caveat that it reflects one person's work on B2B growth problems in spring 2023, and the tools are moving fast enough that parts of it will date quickly.
Where it genuinely saved time
Turning a messy analysis into a first draft. My most common use by a distance. I have the findings, the numbers, and a rough argument in bullet form; I need it in prose. This reliably saves an hour and produces something I then rewrite substantially. The structure is usually right and the language is usually flat.
Generating variants of something that already works. Fifteen subject lines from one that performed, ten ad headline angles from a positioning statement. The hit rate is maybe one in five, which is roughly my own hit rate, delivered in two minutes instead of forty.
Interrogating my own reasoning. The use I didn't anticipate and now value most. "Here's my argument for reallocating budget from paid social to content. What's the strongest case against it?" It's a competent devil's advocate and it has caught two genuine holes in client recommendations before the client did.
Reformatting and restructuring. Turning a transcript into structured notes, converting a framework into a checklist, rewriting something long into something short. Boring, reliable, and it adds up.
Where it cost me time without me noticing
Anything requiring current or specific facts. It will produce a confident, plausible, incorrect benchmark. I asked for typical CAC payback ranges for a vertical and got numbers that looked entirely reasonable and cited nothing real. Checking them took longer than looking them up would have.
I now don't ask it factual questions at all. That rule has saved more time than any prompt technique.
First-draft strategy. Ask for a channel strategy and you get a competent summary of the average of everything ever written about channel strategy. It reads well, contains no wrong statements, and is worthless for a specific company, because the entire value of strategy is in what you choose not to do.
The failure mode is subtle: the output is good enough to edit rather than obviously bad enough to discard. So you edit it, and you end up somewhere generic that you'd never have written yourself.
Anything where the thinking is the deliverable. If I outsource the first pass on a hard analytical question, I lose the process that would have produced the insight. I've caught myself doing this on work where the struggle was the point.
The pattern underneath
Everything on the useful list has the same shape: I already know the answer and need it expressed, varied, or stress-tested.
Everything on the costly list has the opposite shape: I don't know the answer and hoped the model did.
That's the whole heuristic, and it's held for six weeks without exception. Judgement doesn't transfer. Expression does.
What I've changed in how I work
A standing rule: it never touches client-facing analysis unverified. Draft, restructure, critique — yes. Produce a number or a claim that reaches a client without me checking it — no.
I write the argument first, then ask for the prose. Reversing this produces text that sounds like me and argues like nobody.
I use it hardest at the critique stage. Least risky, highest value, and the stage where I'm most likely to be fooling myself.
What I think this means for growth teams
Not headcount reduction. The bottleneck in every growth team I've worked with is deciding what to do, not producing the assets — and this helps with the second.
What I do expect: the cost of producing decent-quality content approaches zero, which means decent-quality content stops being a differentiator. If everyone can produce a competent blog post on demand, competence in the output isn't the moat. What remains scarce is specific, evidenced, first-hand material — original data, real customer research, a genuinely contrarian argument someone will stake their name on.
Which is, awkwardly, the same conclusion good content strategy reached long before any of this. The tooling has just removed the excuse.
The honest summary
Six weeks in, my estimate is that it saves me around four hours a week and has produced zero insights I wouldn't have reached.
That's a real gain and a much smaller claim than the discourse suggests. I'd expect the number to grow as I get better at knowing what to hand over — and I'd be sceptical of anyone reporting a transformation this early.
Hilal Tasdan
B2B SaaS Growth Marketing Consultant & Fractional CMO. Partner in Growth.