Corporate AI7 min read

How Can HR Use AI for Performance Reviews?

On this page

HR and managers can use AI to prepare performance reviews: turning objectives into clear goals, organising the evidence a manager has collected, checking feedback for vague or unfair wording, and planning the conversation. AI should never rate, rank or judge a person. The manager decides from evidence, and only anonymised notes go into a public AI tool.

This guide is for HR teams and line managers in Kenya who run appraisals, probation reviews or mid-year check-ins. It gives seven tasks, each with a prompt you can copy and what to check. For a wider starting plan, read an HR manager's first 30 days with AI.

Why reviews need more care than most HR writing

A performance review affects pay, promotion, training and sometimes a person's job. Two Kenyan laws make that concrete.

The Employment Act, 2007. Section 5 bars direct or indirect discrimination in "recruitment, training, promotion, terms and conditions of employment, termination of employment", and puts the burden on the employer to prove discrimination did not take place. Before terminating employment for poor performance, section 41 requires the employer to explain the reason in a language the employee understands, allow another employee or a shop floor union representative of their choice to be present, and hear and consider what they say. (Text as published by the State Department for Labour, checked November 2026.)

The Data Protection Act, 2019. Review notes about a named person are personal data. Section 26 gives employees the right to access their personal data and to have false or misleading data corrected. Regulation 9 of the Data Protection (General) Regulations, 2021 says an access request must be met within seven days, free of charge, and the ODPC's 2026 guidance notes on data protection policies give the same 7-day timeline. So write every review as if the employee will read it, because they can ask to.

Section 35 also gives people the right not to be subject to a decision based solely on automated processing that significantly affects them. In our view, a bonus or a performance rating set by software with no real human judgement is likely to be that kind of decision.

Seven tasks where AI helps, with prompts

Each prompt uses the four Cs we teach at FuKazee: Clarity, Context, Constraints and Customisation. Replace the square brackets with invented or anonymised details. Remove names, job titles that identify one person, branch names and dates that point to someone.

1. Turn vague objectives into clear goals

I am a line manager at a [sector] organisation in Kenya. Rewrite these objectives for a [role type] so each one says what will be done, how we will measure it, and by when. Keep each under 30 words. Where a measure is missing, write [MEASURE NEEDED] instead of inventing one. OBJECTIVES: [paste]

Check: the employee can influence every measure. A goal that depends on someone else's work is not fair.

2. Build a simple review template

Create a one-page performance review template for [role type]. Sections: agreed objectives, evidence for each, what went well, what to improve, support needed from the manager, development goals for next period, and the employee's comments. Plain English for staff who are not HR specialists.

Check: the template matches your organisation's policy and rating scale, if you use one.

3. Organise the evidence across the whole period

Managers tend to remember the last few weeks best. AI can lay out the whole year so nothing is forgotten.

Below are a manager's anonymised notes about one employee's work from [month] to [month]. Sort them by month and by objective. For each objective, list the evidence for and against, quoting the notes. List any month or objective with no evidence. Do not give a rating, a judgement or a recommendation. Do not add anything the notes do not say. NOTES: [paste]

Check: compare the summary with the original notes. Gaps tell you where to look for more evidence before you write anything.

4. Check feedback for vague or unfair wording

Review the draft feedback below. Flag: comments about personality rather than behaviour; vague words such as "attitude" or "not a team player" with no example; anything that touches age, sex, pregnancy, family, health, religion or ethnic background; and any claim with no evidence. For each, explain the problem in one sentence and suggest behaviour-based wording. Where an example is needed, write [EXAMPLE NEEDED]. Do not change the manager's overall view. FEEDBACK: [paste anonymised draft]

Check: you add the real examples. If you cannot find one, take the comment out.

5. Write self-assessment questions

Write six questions an employee can answer before their review to reflect on their year: achievements, challenges, support they needed, skills they want to build, and one thing the manager could do differently. Warm, plain English, answerable in 20 minutes.

6. Plan the conversation

I am preparing a performance review conversation with a member of my team. Main points: [summarise in general terms, no names]. Write a plan: how to open, the order of topics, two open questions for each topic, how to respond calmly if they disagree, and how to close with agreed next steps. The tone should be respectful and direct.

Check: the plan fits the person in front of you. A script read aloud sounds like one.

7. Find team-wide training needs

Below are anonymised "areas to improve" comments from reviews across one department, with no names or roles. Group them into themes and count how often each appears. Suggest training that would address the top three themes. COMMENTS: [paste]

Check: small teams can be re-identified from context. If a comment points to one person, remove it.

What AI must not do in performance reviews

  • Rate, rank or judge people. The manager decides the rating from evidence and owns it.
  • Decide pay, promotion, probation or termination. These need a person, and termination for poor performance needs the hearing section 41 describes.
  • Invent examples. A made-up example in a review is a false record about a real person.
  • Score staff from monitoring data. Using AI to read emails, chats or keystrokes to judge productivity is likely to be high-risk processing. Regulation 49 of the Data Protection (General) Regulations, 2021 lists automated decisions with significant effects and the "innovative use or application of new technological or organizational solutions" among high-risk processing, and section 31 of the Act requires a data protection impact assessment before high-risk processing starts. Take legal advice first.
  • See named reviews in a public tool. Never paste a review, appraisal score, warning letter, sick note or disability detail linked to a real person into a public AI tool.

If your organisation uses an AI assistant inside its own systems, such as Microsoft 365 Copilot, check with IT what it can see before anyone uses it on review documents. Approved tools still need rules. Our guide to an AI usage policy for Kenyan organisations covers what those rules should say.

This is general information, not legal advice. Your data protection officer or lawyer should confirm your own position.

How FuKazee trains HR teams on this

FuKazee runs hands-on AI training for HR teams in Kenya, built on the work they do in each season of the HR calendar, including appraisals. Practice records are invented or masked throughout, and we never ask a team to bring a real staff file into the room. We start with the four Cs, show the free version of each feature first and promote no single product. See what an HR programme covers, or read why HR managers pick FuKazee.

Questions people ask

Can AI write my staff's performance reviews for me?

It can help you organise evidence and improve your wording. It cannot know how the person worked, so the judgement and the examples must come from you. A review the employee can tell was generated will damage trust.

Can AI give performance ratings?

It should not. A rating affects pay and careers, and section 35 of the Data Protection Act restricts decisions based solely on automated processing that significantly affect people. Use AI to prepare; the manager rates.

Is it safe to paste review notes into ChatGPT?

Not with names or identifying details in a public tool. Remove names, unique job titles, branch names and anything that points to one person first, or use a tool your organisation has approved under a written agreement.

Should employees use AI for their self-assessments?

They can use it to structure their thoughts and fix wording, as long as every achievement is real and they can explain it. Agree a simple rule for the team so nobody feels they are breaking one.

Can AI help with 360-degree feedback?

It can group anonymised comments into themes and suggest clear wording for the summary. Keep raw comments out of public tools, and check that a summary cannot reveal who said what.

Sources

All sources last checked 14 November 2026.

Where FuKazee fits in

FuKazee is a Nairobi training company that has trained more than 3,000 people since 2022. We help HR teams and managers use AI to prepare fairer, clearer reviews, while people keep the judgement. Tell us how your organisation runs appraisals and who needs training. Email lead@fukazee.com

Want guided support instead?

Plan AI training around your team's real work. Answer a few quick questions about your sector, team size and goals, and we will recommend the right programme and format. Sessions are built on your workflows and include a usage policy.

Rodney MainaFounder & Director, FuKazee

Rodney has run FuKazee's training since 2022 and still teaches inside every programme line: corporate sessions, the postgraduate cohort and the KCSE examiner webinars. First Class Honours in Journalism and Mass Communication, University of Nairobi.