Your first 90 days with AI: a realistic plan
If you're a $1M–$20M operator deciding to take AI seriously, here's exactly what your first 90 days should look like — week by week.
Most AI consultants will sell you a 6-month transformation plan. Don't buy it. The 6-month plan is where momentum dies. Here's a real 90-day plan that ends with one live, working system.
Weeks 1–2: measure
Don't build anything. Pick three numbers that map to where you think the leak is. Inbound response time. Quote-to-win conversion. Hours/week the founder spends in admin. Track all three for two full weeks before deciding what to build.
Most operators skip this. Then six months in they can't tell you whether the build worked.
Weeks 3–4: scope one bet
Pick the leak that's both biggest and easiest. Not the most exciting. Not the most 'AI-first'. Biggest leak × easiest fix. For most $1M–$20M operators this is missed-call recovery, quote follow-up, or note-writing.
Write a one-page brief. What gets built. What success looks like (in one number). What can be turned off in 5 minutes if it goes wrong. Done.
Weeks 5–10: build
Six weeks. One agent. End-to-end. Integrated into your real systems on real data. You should see early signal by week 8.
Weeks 11–12: ship + measure again
Live in production. Same three numbers tracked weekly. By the end of week 12 you should have proof — the number moved, or it didn’t.
If it moved: you've earned the right to do the next build. If it didn't: stop, diagnose, don't keep building blindly. Most agencies keep building. We don't.
End of day 90: one live system, one number moved, or one honest conversation about why not.
Common mistakes in the first 90 days
- Building two things at once. Don't. Pick one.
- Skipping the measurement weeks. You'll regret it at week 12 when you can't prove anything.
- Picking the most exciting build instead of the biggest leak. Boring is bigger.
- Outsourcing without owning the brief. If you can't say in one sentence what success looks like, don't sign the SOW.