Most AI projects that disappoint were chosen by instinct. A one-week time audit gives you a baseline: what the work costs today, in hours, so you can tell later whether anything improved. It is unglamorous, and it is the single most useful thing to do first.
Pick fixed categories before anyone logs
Free-text logs are impossible to total. Agree six to eight categories that fit your business before the week starts, for example: client delivery, client communication, sales and proposals, admin and finance, internal coordination, and learning or improvement. Add an 'other' category, but treat heavy use of it as a sign the categories need fixing.
Ask people to note the task as well as the category. 'Admin' tells you little; 'chasing missing receipts' tells you exactly what to look at.
Log in 30-minute blocks for five working days
Thirty-minute blocks are precise enough to total and loose enough that people keep going. A shared spreadsheet with columns for person (or role), day, category, task and minutes is enough. If people drop off mid-week, reduce to one entry at the end of each day rather than abandoning the audit.
Use roles rather than names if you plan to share or analyse the data with any tool. A time log is personal data about your staff.
Total by task and look for repeat work
At the end of the week, total hours per task per person. Tasks above roughly three hours per person per week deserve a closer look. Then split each one into steps and ask of every step: could someone do this from written instructions alone? Steps that pass are rule-based; the rest need judgement.
Tasks that are both frequent and mostly rule-based are your strongest candidates. Tasks that are mostly judgement, or carry high cost when wrong, stay with people for now — AI can help prepare, not decide.
- Frequent (weekly or more)
- Mostly rule-based steps
- Low cost of an error, or an easy human check
- Information already digital and in a known place
Write the baseline down before changing anything
For your top two or three candidates, record hours per week, how often errors happen, and how long a typical turnaround takes. This is the yardstick for any pilot. Without it, 'it feels faster' is the best evidence you will ever have.
Questions
- Is one week enough?
- For a first pass, yes, as long as the week is typical. Avoid holiday weeks or month-end if those are unusual for you; if month-end matters, log that week separately.
- Do I need time-tracking software?
- No. A shared spreadsheet works. Software helps only if your team already uses it.
Written and reviewed by The AI Scale Academy team. Guidance is general and educational, not legal, tax or financial advice.