Why most businesses pick the wrong first project
Owners usually start with whatever they last saw demoed: a chatbot, an AI voice, an automation that connects five apps. Those can work, but they're rarely the best first step. A first AI project should pay back quickly, be easy to check and build confidence, so the second one happens. That points to boring, repetitive work you already do every week.
Step 1: list your weekly tasks
Spend 20 minutes listing what you and your team do in a normal week that isn't the core work itself. Be specific: "write quotes for kitchen jobs," not "admin." Common ones:
- Answering the same customer questions by phone, text or email
- Writing estimates, proposals and follow-ups
- Scheduling, rescheduling and reminders
- Chasing unpaid invoices
- Writing job notes, reports or handover summaries
- Replying to reviews and posting updates
- Training new staff on how things are done
Step 2: score each task
Give each task three scores from 1 to 5:
| Score | Hours per week | Repeatability | Low risk |
|---|---|---|---|
| 1 | Under 1 hour | Different every time | A mistake is costly or public |
| 3 | 2 to 4 hours | Similar most times | A mistake is awkward but fixable |
| 5 | 5+ hours | Same pattern every time | A person checks it before anyone sees it |
Multiply the three scores. A task scoring 5 × 5 × 4 = 100 beats a task scoring 2 × 3 × 2 = 12 every time, however exciting the second one sounds. Hours matter most, so if two tasks tie, pick the one that takes longer.
Step 3: pick one and run a two-week test
- Pick the top scorer. Just one.
- Use a tool you already have if you can: your field-service, accounting, CRM or email software may already include AI features. If not, a general assistant such as ChatGPT, Claude or Gemini is enough to start.
- Write one good prompt or template and save it, so you aren't starting from scratch each time.
- Time it. Note roughly how long the task took before and during the test.
- Keep a person checking every output for the two weeks.
- Decide: keep it, adjust it or drop it, then go to the next task on your list.
Tasks that usually score well
- Writing: quotes, proposals, follow-ups, review replies, job descriptions. See our guides on estimates, proposals and review replies.
- Summarizing: turning calls, site visits or meetings into notes and task lists.
- Standardizing: turning how you do a job into a written checklist (see AI for SOPs).
Tasks that usually score badly as a first project: anything customer-facing with no human check, anything involving pricing decisions, and anything that needs several systems connected before it works.
Before you start: one rule about data
Decide what your team may and may not paste into AI tools before the first test, not after. Our one-page AI policy covers it in ten minutes.
This scoring sheet is the do-it-yourself version of what your business's AI plan does for you. In a 15 to 30 minute voice interview we learn how your business runs. Then we score your tasks by impact and effort, match each one to specific tools with monthly cost and setup time, and put the result in a written plan within 24 hours.