Owners usually have a list of twenty things AI might help with. The mistake is starting with the most exciting one. The first project should be the one most likely to finish and show a result, because that result buys you trust, budget and staff goodwill for the second.
Run each candidate through five questions. A good first workflow answers yes to most of them.
The five questions
1. Does it happen often?
Daily or weekly work gives you data quickly and compounds savings. A task that happens twice a year is a poor first project even if it is painful, because you cannot tell whether it is working.
2. Is the volume meaningful?
Ten items a week at fifteen minutes each is not much. Two hundred items a week is. The point is not to hit a magic number but to make sure the saving is large enough to matter after you account for setup and review time.
3. Can you describe what "done correctly" looks like?
If two of your best people would disagree on the right answer, an AI will not settle it. Look for work with a checkable result: the invoice is coded to the right job, the quote includes every line item, the call is booked on the calendar.
4. What is the cost of a mistake?
Drafting an internal summary is low risk. Sending a client a wrong price is not. High-risk work is not off limits, but it needs an approval step and it makes a poor first project. Start where an error is cheap and visible.
5. Can you measure it?
You need a number that exists today or that you can capture in two weeks: minutes per item, days to respond, error rate, items waiting. No baseline, no proof. Surveys show why this matters: in Bluevine's survey, 52% of AI-using small businesses reported a return, while 24% had not yet seen one.[1]
A scoring sheet you can use today
List your candidates down the left side. Score each question 0 to 2, where 2 is a clear yes. Add them up. The top two or three are your shortlist. Do not average in how excited you are about the idea.
| Candidate | Frequency | Volume | Clear "done" | Low mistake cost | Measurable | Total |
|---|---|---|---|---|---|---|
| Example: coding vendor invoices | 2 | 2 | 2 | 1 | 2 | 9 |
| Example: writing custom proposals | 1 | 1 | 0 | 0 | 1 | 3 |
| Example: following up on missed calls | 2 | 1 | 2 | 1 | 2 | 8 |
The numbers above are an illustration of how to use the sheet, not data from a real business.
Where first workflows usually come from
- Inbound requests: calls, forms and emails that need sorting, a first reply and a booking.
- Documents in, data out: invoices, receipts, applications and intake forms that someone retypes.
- Chasing: reminders for missing documents, unsigned forms and overdue invoices.
- Recurring reports: the weekly or monthly summary someone builds by hand from three systems.
- Internal drafts: meeting notes turned into action items, first drafts of routine letters.
Keep it small on purpose
McKinsey found that the companies seeing the most value redesign the workflow rather than bolting a tool on top.[2] That is easier to do with one contained process than with a whole department. Aim for something that can be live in weeks, not quarters, with a named owner, one number and an approval step on anything risky.
Once it works, you will know your actual error rate, your real review time and how your staff respond. Those facts make the second workflow far easier to scope than the first.
Sources
- Bluevine, "2026 Small Business AI Trends Report"
- McKinsey, "The state of AI in 2026: On the road to ROI"
Figures are quoted from the sources above and were reviewed in October 2026. Where a source is a secondary summary of other research, we say so in the text. Survey results reflect each survey's own sample and definitions. Examples labeled as illustrations use made-up numbers to show the method. Nothing here is a forecast or a guarantee of results.
Want this worked out for your business?
Tell us what is wasting your time.