Practical guide
AI fluency for people analytics: protect survey meaning
Review a fictional learning survey for denominator errors, nonresponse and small-group disclosure before writing a team-level summary.
People analytics managers can use AI to summarize development evidence while protecting the meaning and limits of a dataset. This fictional survey exercise distinguishes invitations, responses and favorable answers, then checks whether a proposed breakdown reveals too much. It does not rank employees or infer individual capability from survey answers.
Define what the survey actually measures
A self-reported response should not be relabeled as an observed work result.
A fictional learning survey asks whether respondents feel able to find the approved AI guidance. Twenty people were invited, ten responded and eight selected yes. The question concerns reported access confidence, not demonstrated verification skill or task performance. Keep the wording attached to every summary.
Ask AI to calculate both eight of ten respondents and eight of twenty invited people. The first is eighty percent of respondents; the second is forty percent of invitees who supplied a yes answer. Neither means eighty percent of the entire team has demonstrated AI fluency.
Keep nonresponse visible
People who did not answer have not supplied a negative answer.
Ten invitees did not respond. The dataset does not explain why. They may have missed the invitation, lacked time or had other reasons, but those possibilities are not findings. Do not fill their answers with a model prediction or treat silence as evidence of low confidence.
Write a summary that states the response count beside the favorable count. If the owner asks for a team-wide conclusion, identify the evidence gap instead of stretching the result. AI can help draft a clear limitation; inspect whether the shortened version drops the response denominator.
Inspect the proposed breakdown before sharing it
A table can reveal information through a tiny subgroup even without names.
The exercise’s proposed table separates a subgroup containing one respondent. Showing that row would reveal that respondent’s answer to anyone who knows the group’s membership. This is a logical property of the fictional table, not a universal legal threshold for reporting.
Remove the identifying breakdown from the public-facing exercise summary and ask the responsible data owner what reporting rules apply in real use. Consider whether totals or other rows let readers reconstruct the removed value. Do not promise anonymity merely because a spreadsheet omits a name column.
Make the next question proportionate to the result
A limited survey finding can support a limited follow-up without becoming a personnel judgment.
The eight yes answers suggest some respondents report knowing where guidance is. The two other responses and nonresponse leave different questions open. A proportionate next step might check whether the guidance link is easy to locate for the intended audience, using a separate consented or synthetic usability exercise.
Do not automatically assign remedial training, infer resistance or compare named employees. Keep any follow-up purpose and data access bounded. If the real question is applied capability, explain that a different evidence method is needed rather than renaming this survey as an assessment.
Deliver a transparent aggregate note
The final summary should retain the question, denominators and sharing limits.
Include the exact fictional question, twenty invitations, ten responses and eight yes answers. Explain the small-group reporting issue and show the aggregate wording you would share for this exercise. Keep the unshared breakdown out of the generated narrative as well as the visible table.
Ask a reviewer to reconstruct the calculation and identify what cannot be concluded. Introduce a second survey with a different question and check that the worker does not treat the results as a trend. This task concerns interpretation and disclosure of collected survey evidence, not forecasting individual performance.
Sources and scope
NIST: AI risk management. Skills England: workplace AI foundations.
These references provide background, not validation or endorsement of this exercise. The case details, calculations and suggested review questions are original instructional material. Use them to discuss observable work, not to infer customer outcomes, professional credentials or performance in every setting. Before adapting the exercise, confirm the relevant facts, approved tools, data permissions and decision owners. If you change the case, revisit the expected answers and checks as well. These examples describe practice tasks, not a promise that a particular product includes the fictional features.
Sources: [1] [2]
Sources
- 1.AI RMF Core · NIST
- 2.AI foundation skills for work benchmark · Skills England