"Where should we use AI?" is the wrong first question, because it starts from the technology and works backwards to a use for it. The useful version of the question starts from the task: what's slow, repetitive, or backed up right now, regardless of what fixes it.
Where the obvious starting points usually sit
- Answering the same customer questions over and over, by email or phone, that a written answer would cover just as well.
- Quoting or pricing work that follows a pattern, but still gets typed out by hand each time.
- Getting information out of documents — invoices, forms, applications — and into a spreadsheet or system, by reading and retyping it.
- Following up on things that fall through the cracks because nobody has time to chase them: unpaid invoices, unanswered quotes, leads gone quiet.
A three-step way to find your actual starting point
- 1Ask whoever does the most repetitive part of the job what they'd remove first, if they could remove one thing. They already know. Nobody's asked.
- 2Count it — roughly how many hours a week, on the clock, by the person who actually does it. A task costing six hours a week is worth solving properly. One costing forty minutes a month probably isn't, yet.
- 3Check what you already pay for. Most CRMs, accounting packages and office suites shipped AI features in the last two years. Sometimes the fix is a settings toggle, not a project.
This is deliberately the same test we'd apply to someone else's $45,000 quote before recommending you sign it. It doesn't get skipped just because we're the ones doing the recommending.
What doesn't need to be the starting point
A custom-trained model, an "AI strategy document," or a chatbot on your homepage are rarely where the actual time is going. They're the visible, easy-to-pitch version of "using AI," which is exactly why they get proposed first — and exactly why they're worth being skeptical of as a first move.
The businesses that get real hours back tend to start narrow: one task, counted honestly, fixed properly, with a person still checking anything that matters. Wider comes after, once the first one's working.