Bora Unlu is the co-founder and CEO of Teamflect, a performance management & employee engagement platform designed for Microsoft 365.
When the discussion of AI in performance management first became prominent, the debate centered on judgment. Could an algorithm evaluate an employee more fairly than a manager could?
A 2025 Gartner survey of nearly 3,500 employees found that 87% thought algorithms could give fairer feedback than their managers. For anyone who manages people, that is a difficult sentence to read.
The software, however, went somewhere else. Gallup’s workforce data recently found that writing and editing are the most common uses of AI at work, at 51%. The performance review category, therefore, shipped writing help rather than judgment: Draft the review, summarize the year, turn scattered notes into competency language.
The largest employers did the same. Early adopters like Citi and JPMorgan both introduced AI to help managers write evaluations, not to grade anyone.
Building Teamflect has given me a close view of how employees, managers and HR teams use AI around performance reviews. Along the way, I’ve been surprised to find that these teams aren’t asking AI to make the judgment calls (well, occasionally they are).
Employees want help expressing what they know. Managers want a second reader. HR teams want help understanding across a cycle. In most cases, the consequential decision of judgment still belongs to a human owner.
Employees Use AI As A Translator
Individual contributors usually come to AI a few days before the review deadline with a particular job in mind. They ask for help with their self-review highlights.
The requests are relatively plain. Some paste in the question they are stuck on and ask what to write. Some want every answer filled in. Some employees even ask for help when rating their own performance.
This actually makes sense, considering what most review forms ask. We are all used to sections like:
• Rate your ownership of project execution.
• Describe your progress against the action plan.
Employees know what they did, but they can get stuck finding the right vocabulary.
There are two honest ways we can read this form of usage.
The first is relatively uncomfortable. For some, the self-review is not a reflective exercise but a compliance activity that they would rather get off their desk.
The second is more generous, and equally true. The traditional format many organizations use for performance reviews can actually form a barrier between employees, their accomplishments as they see them and their managers. In short, employees use AI as a translator between what they did and what the review form requests from them.
Managers Use AI As A Second Reader
Managers draft reviews, too. But more of what they bring to AI is a review they have already written, along with a question: What is wrong with this? A meta “review of the review,” if you will.
The requests tend to be more specific. Is this too positive? Can we make both sections more professional, because this person may be going onto a performance plan? Has my frustration about the transfer leaked into the forward-looking section?
None of these managers in these scenarios are uncertain about the content. They know what they think, but are left wondering about the proper tone and exposure, as well as how the document will read to anyone who sees it later. When the stakes turn legal or emotional, people want a second set of eyes. AI offers the perfect set of digital eyes, which can be beneficial with something as sensitive and private as performance reviews.
A truly unexpected use case comes in the form of calibration. Early on, the anticipated use cases for most organizations were centered around reviewers and reviewees.
However, there is the middle layer. Skip-level managers and HR tend to ask AI review assistants what should be flagged and sent back to the reviewer. In plenty of scenarios, the concerns that can be brought to AI can be about a review someone else writes.
I believe this can be a double-edged sword. While helpful, AI usage can subtly damage a review. When asked to make a passage more professional, it smooths the edges. Framing facts and feedback in a professional manner is a core competency for managers, and offloading that skill to AI can lead to it atrophying.
Admins Use AI As An Analyst
Administrators do ask configuration questions more than anyone else. Mostly, they tend to ask questions of their own data. Summarize progress for these three people over the past week, grouped by parent goal. Compare the internal candidates for a senior opening and weigh their competency ratings. List how many review forms are still awaiting finalization, by reviewer.
Our work on post-review organizational insights changed how I think about this use case. Questions like “Who scored highest?” are often less useful than “Where are strong teams carrying hidden workload? Where is unclear ownership appearing across departments? Where is the quality of written feedback making the results difficult to trust?” The score often only tells HR where to look, but the comments can tell them what to do next.
This is a more consequential use of AI than generating another paragraph of review prose. But it still supports judgment rather than replacing it. AI can expose the organization behind the ratings; people must decide what to do about it.
Why The Decision Stays Human
If we look at these three groups of AI users in performance appraisals, we can see one thing in common: People don’t actually hand over the judgment.
Employees use AI to translate a decision they had already made about themselves. Managers use it to test one before delivering it. Administrators use it to measure decisions after the fact.
The debate around using AI for performance reviews assumed people wanted a fairer verdict from a machine, but they actually reached for help before, around and after the verdict. My conclusion from building in this category is that AI should make human judgment better informed, easier to explain and harder to exercise carelessly. It should not make judgment disappear.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?