Most AI conversations in business start too big. Strategy decks, transformation programmes, a working group. Meanwhile, somebody in the office is spending ninety minutes every morning re-typing enquiry emails into the CRM, and nobody thinks of that as “an AI use case.” It’s the best one you have.
The tasks AI does brilliantly today are the ones with a pattern: read this, extract that, draft the reply, file it, summarise it, route it. Which is precisely the work your team already describes with the word “tedious.” So skip the brainstorm and run an audit instead: for one week, have everyone note tasks that are repetitive, text-shaped, and done more than five times a week. That list — not a consultant’s framework — is your AI roadmap.
The gap between an AI feature that helps and one that embarrasses you is integration quality, not model choice. The rules we build to: a human approves anything customer-facing until the accuracy record earns autonomy; the AI only sees the data it needs; every automated action is logged so you can see what happened and why; and there’s always a clean hand-off to a person when the AI is out of its depth.
Measured this way, the payback shows up in weeks — as hours returned. And the compounding is real: each automated task frees attention that makes the next automation easier to spot. Start with the repetition. The moonshots can wait until the boring stuff pays for them.