Key takeaways
- Ask which jobs are repetitive, low-risk and slow. Not the open-ended question of what AI can do.
- Good fits: research, first drafts, repetitive checks, messy-data wrangling - work where the output gets reviewed.
- Keep AI away from final judgement, anything where a confident wrong answer is expensive, and relationships.
- Start with one job, build a reliable workflow around it, then do the next. That is the whole method.
Most teams approach AI the wrong way round. They ask "what can AI do" and end up with a folder of prompts nobody uses. The better question is narrower: which jobs on this team are repetitive, low-risk, and slow - and would a draft from a machine save real time?
That question sorts the work into two piles.
What AI is genuinely good at#
It earns its place on research and summarising: the shape of a topic, a SERP, a competitor's pages, a stack of call notes. You start from something instead of a blank page.
It is good at first drafts of structured things too. Briefs, outlines, meta descriptions, FAQ blocks. Not the finished work, the version a person then fixes.
It does not get bored, which makes it good at repetitive checks. Whether every page in a batch has a sensible title length, a clean H1, no duplicate meta. A human gets sloppy on page 80. A machine does not.
And it is good at wrangling messy data, the reformatting and tagging that quietly eats an afternoon.
The thread running through all of it: the input is clear, the output gets reviewed, and a wrong draft costs you a few minutes rather than a client.
What it should not touch#
Keep it away from final judgement. What to prioritise, what to kill, what to tell a client. That is the job, not the overhead.
Keep it away from anything where a confident wrong answer is expensive. Legal claims, pricing, statistics you are about to publish. AI is fluent, not accurate. Fluent but wrong is the dangerous combination.
And keep it away from the relationships. The call, the bad-news email, the conversation that needs someone who remembers the history.
Start with one job, not a strategy#
The teams that get value do not roll out "AI". They take one annoying, repeated task - say, turning call notes into a first-draft brief - and build a small, reliable workflow around it. Clear input, a review step, a known output.
Once that one works and the team trusts it, you do the next one. That is the whole method. If you want help finding the first job worth doing, that is exactly what AI consulting is for.
Frequently asked
Will AI replace my marketing team?
Where should a team start with AI?
What should AI never be used for in marketing?
