← All articles

AI Leadership

Ask Your Team What AI Helped Them Finish, Not Just How Fast It Felt

A five-question staff survey and two worked responses that distinguish reported usefulness from measured value and identify the next evidence to collect.

By , Co-Founder & CEOUpdated 4 min read
An open survey notebook sits among examples of completed work and a response tray.

An employee says AI makes them twice as productive. Another says it saves no time but helps them finish a report they used to avoid. A survey that asks only for a speed estimate will miss the second benefit and may overstate the first.

Ask what staff completed, what changed, and what another person had to do afterward. Use the responses to choose a follow-up, not to calculate company ROI from people's impressions.

Ask five questions about a recent task

Give employees a clear period, such as the previous two working weeks, and ask for one task. They should avoid including customer names, confidential documents, or private account details in a general survey.

  • What task did you use AI for, and what finished output did you produce?
  • How would you have handled that task without the tool, including whether you would have done it at all?
  • What did you check or change before accepting the result?
  • Who else reviewed, corrected, or used the output, and what did they need to do?
  • What should we examine next to decide whether this use deserves support?

Leave room for “I do not know.” Forcing a minute estimate encourages precision employees may not be able to provide. Ask for a useful example before asking them to quantify it.

Read two answers differently

The following responses are hypothetical demonstrations of how a manager can interpret the survey.

ResponseWhat it supportsNext action
“I drafted six renewal emails faster. My manager rewrote four because the terms were wrong. I am not sure how much time they spent.”AI helped initial drafting; total effort and correctness remain unknown.Review representative emails with the manager and measure correction effort before widening use.
“I produced a supplier comparison I would not have had time to prepare. Procurement found it useful but checked the prices itself.”The employee reports a new useful output; speed savings are not established.Ask procurement which decision the comparison supported and which parts need dependable sources.

For the first response, counting the drafter's time alone would ignore the manager. For the second, a zero time-saving answer could hide an output the business values. The manager needs a different follow-up for each.

Keep perceptions separate from observed work

METR's May 2026 technical-worker survey distinguishes reported speed from reported value. The authors describe a convenience sample and caution about selection and respondents' difficulty estimating counterfactual effects. Those estimates do not establish typical returns for a small business.

Use that distinction in your own survey. An employee can tell you what they experienced and which work felt easier. A manager still needs accepted outputs, reviewer effort, or evidence about the supported decision to assess the business value.

Do not turn “I would have taken longer” into a cash-saving claim. If you want to estimate staff capacity, define the task and collect relevant work evidence. If you want to establish savings, identify an actual cost the business avoids. Our AI ROI guide covers that distinction.

Return a useful response to the employee

Tell the employee what you will do with the example. For the renewal emails, the next step might be fixing access to approved terms before testing drafting again. For the supplier comparison, it might be supporting the research format while keeping current price verification with procurement.

Keep this survey separate from a tool inventory. You are asking about useful work and the next evidence needed, not counting accounts or enforcing a policy. If a response raises an account or data concern, route that concern to the relevant owner through the company's process.

Avoid rewarding the biggest claim

Tell staff that a careful account of a failed or mixed result is as useful as an enthusiastic answer. If leaders celebrate only large reported gains, employees may feel pressure to describe routine assistance as transformation. Review the examples for concrete finished work and unresolved questions rather than ranking people by claimed speed. In the hypothetical survey, the renewal drafter’s correction problem deserves attention even if another employee reports a larger benefit. Use the information to improve the task and support decision, not score employee enthusiasm.

At the next review, ask whether the agreed follow-up changed the decision to support the use. For leadership that connects employee experience with practical evidence and investment choices, explore fractional CAIO support.

Luke Thompson

About the author

Luke Thompson

Co-Founder & CEO, The Ops Guide

Luke Thompson is Co-Founder and CEO of The Ops Guide. He writes about practical uses of AI, workflow automation, custom software, and SEO for growing businesses.

About The Ops Guide

The Ops Guide helps businesses improve how work gets done through custom software, workflow automation, AI systems, and SEO services.

The Ops Guide Brief

Get AI news you can use.

The developments worth your attention, what they change at work, and what to do next. Choose a daily briefing or a weekly selection.

Free to subscribe. Confirm by email. Unsubscribe any time.