Solo Guide No. 001 · Strategy · October 2026
How do I choose my first AI project? Start where it costs you most.
Start with the work that wastes the most hours or loses the most money every week, not the most exciting idea. List your repeated tasks, score each from 1 to 5 on hours lost, money at stake, how often it happens and how easy it is to automate, add the scores, then pick the highest total you can launch in two to three weeks with a person still approving the output.
Why most first AI projects disappoint. Usually for boring reasons.
Most small businesses start with a tool ("we should use ChatGPT for something") instead of a problem. The project gets built, nobody owns it, and six months later nobody can say what it saved. A good first project is boring on purpose: a task you already do every week, with a cost you can count before and after.
The five steps. About a week of your time.
- 1
List the repeated work
Spend one week writing down every task that happens at least weekly: answering new leads, sending quotes, chasing invoices, updating a spreadsheet, scheduling, writing follow-ups. Note who does it and roughly how long it takes.
- 2
Put a number on each task
- Hours a week it takes, and the hourly cost of the person doing it.
- Money at stake: leads that go cold, invoices paid late, mistakes that cost a refund.
- Frequency: daily work beats monthly work.
- 3
Score how ready it is
A task is ready when the steps are written down (or could be in an hour), the information it needs already lives in a tool you use, and a mistake is cheap to catch.
- 4
Rank with a simple sheet
Score each task from 1 to 5 on hours lost, money at stake, frequency and ease, then add the four numbers. The highest total is your first project. Use the sheet below.
- 5
Measure before you build
Write down today's number (for example: average reply time 4 hours, 30% of leads booked). Build the smallest version, keep a person approving every message at first, and compare after 30 days.
The scoring sheet. Fill it in with your own tasks.
Type your own weekly tasks over the examples and score each from 1 (low) to 5 (high). The total, out of 20, ranks them. The top row is your first project.
| Task | Hours lost | Money at stake | Frequency | Ease | Total | Remove |
|---|---|---|---|---|---|---|
| 17 / 20Start here | ||||||
| 14 / 20Start here | ||||||
| 13 / 20Start here | ||||||
| 10 / 20Start here |
Reading the example's top row: 3 + 5 + 5 + 4 = 17. In most service businesses the lead reply wins, because a slow answer costs the sale, not just the time. Your scores stay in this browser only.
Free download
Get this sheet as a spreadsheet. Totals built in.
A file you can open in Excel or Google Sheets, plus a one-page printable version. We'll also send the next guide when it is out.
04A worked example · Sales
Answer leads in 5 minutes, not 4 hours.
A home-services company gets 60 new leads a month, answers in about 4 hours and books 30% of them: 18 jobs at an average of $900. Lead reply scored highest on its sheet, so that is the first project.
Measured against: booked jobs, before and after 30 days. Example figures for illustration, before counting the hours the owner no longer spends on the phone.
| Measure | Before | After |
|---|---|---|
| Time to first reply | ~4 hours | Under 5 min |
| Leads booked | 30% | 36% |
| Jobs a month (60 leads) | 18 | About 22 (21.6) |
| Added revenue a month | About $3,240 (3.6 jobs × $900) |
What to avoid
- Starting with the shiniest idea.A chatbot nobody asked for rarely beats fixing the follow-up you already miss.
- No owner.Every project needs one person who checks it weekly.
- No "before" number.Without it, nobody can prove the project worked.
- Removing the human too early.Keep approval on anything that reaches a customer until the results are steady.
Do it yourself, or get help?
Do it yourself when
- The top task lives in one tool.
- Someone on your team has a few hours a week to own it.
- You can measure the result without help.
Bring in help when
- The task touches several systems: CRM, inbox, calendar.
- Nobody has time to own it.
- You want the savings measured properly.
That is the work Solo Studios does: we find the bottleneck, build the tool that clears it inside the software you already use, and keep it running.
Questions
How long should a first AI project take?
Two to three weeks for the first working version is a good target. If it needs months, pick a smaller slice.
Do I need to replace my software?
Usually not. Most first projects connect to the CRM, inbox or accounting tool you already pay for.
What should I measure?
One number that matters to the business: reply time, booked jobs, hours saved, days to get paid.
Is it safe to let AI answer customers?
Start with AI drafting and a person approving. Move to automatic only when you trust the results.
Next step
Bring your top row. We'll find the first win.
Book a discovery call with your scoring sheet. We'll check the task worth fixing first and what it should save.