Guide
How to measure AI adoption across a portfolio
Measure AI adoption at three levels: whether people use it (weekly active users as a share of licensed staff), whether it is built into real work (workflows live per team), and what it is worth (hours saved and a change in output such as proposals sent or response time). Track the same few numbers at every portfolio company so the operating partner can compare them in one view.
The five numbers to track
| Metric | What it tells you | Where it comes from |
|---|---|---|
| Weekly active users, as a share of licensed staff | Whether people use AI at all | The AI tool's admin console |
| Workflows live per team | Whether AI is part of the work, not just tried | The rollout checklist |
| Hours saved per person per week | What the time is worth | Before-and-after time samples |
| One output measure per team | Whether the business result moved | The company's own systems (agency management system, Deltek, Bullhorn, ERP) |
| Policy and training coverage | Whether use is safe and supported | Training and signed-policy records |
Seats purchased is not on the list. It tells you what was spent, not what changed.
Set a baseline before anything changes
Before the first workflow goes live at a portco, record a week of the work you plan to change: how long a quote, proposal or service reply takes, how many go out, and how fast clients hear back. Without that baseline, every later claim about time saved is a guess.
Count hours saved honestly
Self-reported time savings are a useful signal and a weak proof. In Liberty Mutual's 2026 Independent Agency Growth Study, insurance agents who use AI said it saves them an average of four hours a week. That is a reasonable figure to test, not a result to promise.
To get a number you can defend to a board:
- Time a sample of the same task before and after AI is set up.
- Multiply the difference by how often the task happens each week.
- Report a range, not a single number.
- Check it against an output measure. If proposals take half the time, proposal volume or turnaround should show it.
Pick one output measure per team
| Team | Example output measure |
|---|---|
| Sales | Quotes or proposals sent per week; time from request to quote |
| Service | Time to first reply; requests handled per person |
| Account management | Renewals reviewed before the deadline; cross-sell conversations started |
| Operations | Call notes logged in the CRM the same day |
Industry context helps choose the measure. In architecture and engineering, 71% of firms say they are submitting more proposals than last year, while win rates fell from 56% in 2024 to 50% in 2026, according to Unanet's AEC Inspire Report, so both proposal turnaround and win rate matter.
The portfolio view
One row per portco, the same columns everywhere, updated monthly:
| Portco | Status | Weekly active users | Workflows live | Hours saved per week | Output measure | Policy signed |
|---|---|---|---|---|---|---|
| Company A | Pilot | — | — | — | — | — |
| Company B | Pilot | — | — | — | — | — |
| Company C | Next | — | — | — | — | — |
The status column (pilot, live, next) does most of the work in board conversations. The rest shows whether live really means used.
Warning signs
- High seats, low weekly use. Licenses went out before setup and training did.
- Use concentrated in one or two people. The workflow lives in someone's habits, not in the team's process.
- Hours saved with no change in output. Either the savings are overstated or the time is going somewhere else. Find out which.
- No policy coverage. Use is growing in accounts the company doesn't control. Our AI policy template covers the basics on one page.
For how the measurement fits into the wider rollout, see how to roll out AI across portfolio companies.
Questions
What is a good AI adoption rate for a portfolio company?
There is no reliable public benchmark for portco teams yet. Set targets from your own pilot: measure weekly active users and hours saved at the two pilot companies, then use those numbers as the bar for the rollout.
How do you measure hours saved by AI?
Time a sample of the same task before and after AI is set up, multiply the difference by how often the task happens, and report a range. Treat self-reported estimates as a starting point and check them against output numbers from the company's own systems.
How often should operating partners review AI adoption?
Weekly at the pilot companies, while workflows are still changing, and monthly across the portfolio once companies are live.