Continuous Performance Feedback Sounds Transformative Until You See How Companies Are Running It
AI Benefit: AI-driven feedback can provide real-time insights, reducing biases seen in traditional reviews.
Data Quality: Many organizations struggle with insufficient, outdated, or inconsistent data affecting feedback quality.
Trust Issues: Transparency in AI-generated feedback is crucial; disclosure impacts employee trust and feedback effectiveness.
Implementation Gap: There's a growing disconnection between AI's potential benefits and its practical application in organizations.
Feedback Loops: Ineffective feedback loops may shorten review cycles but do not resolve the issues of annual evaluations.
Continuous performance feedback powered by AI has moved from conference wishlist to enterprise rollout faster than most people management practices in recent memory.
The pitch is consistent across vendors: real-time input, reduced recency bias, feedback that isn't shaped by how a manager happened to feel on a given afternoon.
Employees have long said they want more frequent, more specific input than annual reviews provide. AI-assisted systems are designed for exactly that, but the gap between the promise and the implementation is becoming harder to ignore.
Keep Reading—and Keep Leading Smarter
Create a free account to finish this piece and join a community of forward-thinking leaders unlocking tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
The Manager Rubber-Stamp Problem
Managers are approving AI-generated summaries without examining the models that produced them.
Andrew Whyatt-Sames, an organizational psychologist and co-founder of UptakeAI, which works with organizations deploying AI across people and performance processes, has watched this play out repeatedly.
When AI generates a performance summary, most managers treat it as a draft to approve rather than a starting point to interrogate.
One line manager, three months into a deployment at a 2,000-person business, told him: "It said 'strong communicator' and I agreed — the data's all there, isn't it?"
The underlying model was weighting meeting attendance as a proxy for communication. Nobody had told the managers that.
Part of the problem is foundational. Continuous feedback systems need continuous, high-quality data. Most organizations don't have it.
"It only has whatever context you give it," said Matt Poepsel, VP of Talent Optimization at The Predictive Index, said when we spoke at the Transform conference in March. "It would be like asking a friend for advice about your boss, and they'd say, 'I don't really know your boss — you're going to have to tell me more.'"
People data, he noted, was already inconsistent before AI got involved. Job descriptions pulled from outdated postings, skills assessments tied to requirements that have shifted, behavioral profiles that managers never built because no one trained them to ask.
We have gobs of data. The question is whether it’s the right data.
When the underlying data is weak, continuous feedback doesn't self-correct. It compounds the problem at higher frequency.
Disclosure Is a Problem Either Way
There's a trust dynamic that further complicates adoption. As we've covered previously, a 2021 study published in the Strategic Management Journal found that AI-generated feedback outperforms human manager feedback on measurable outcomes, but only when employees don't know it came from a machine. Once disclosed, the performance gains disappear.
Kamaria Scott, Founder and CEO of Enetic, has watched the disclosure problem play out directly.
"Just don't ever lie about it," she said. "You have to be transparent."
Managers who disclose AI-generated feedback risk employees dismissing it. Those who don't are sitting on a disclosure problem that tends to surface eventually.
What's at stake, Scott argues, is whether these systems are being used as a starting point or an end point.
The assessment is not the end goal. The assessment is the conversation starter.
In her view, the most defensible use of AI in performance management is as a data input that opens a discussion, not a summary that stands alone. "Something that says, 'This is a neutral place for us to start' — that's useful. But you have to come to the middle."
Employees Know What's Being Measured
The gaming problem tends to emerge around six months into a deployment, according to Whyatt-Sames.
Employees learn what the system is tracking, usually through informal channels rather than anything official, and adjust accordingly. More check-in comments, more tagged objectives, output that is legible to the system rather than useful to anyone. An L&D lead he worked with called it "playing the dashboard."
The volume problem follows a similar arc. Organizations measure adoption by activity counts, so three times the feedback volume looks like success.
When you audit the feedback itself, actionability drops sharply," Whyatt-Sames said. "It becomes shorter, more formulaic, and concentrated on whatever the system rewards.
The signal-to-noise inversion often only surfaces when someone runs a retrospective, by which point the pattern is well established.
When the Work Itself Is Shared With an Agent
Scott raised a question in March that hasn't found a clean answer: "What does performance look like if people themselves aren't doing the work, or agents are doing half the work? How are you going to evaluate my performance as a person for work I'm not even doing fully myself anymore?"
It's a question most current implementations aren't designed to address. They're still measuring individual output in a context where the boundaries of individual contribution are becoming less defined.
Whyatt-Sames' diagnosis of why these problems persist connects the individual failure modes.
Organisations treat the AI layer as the solution and skip the change architecture," he said. "The system generates the feedback; nobody asks whether the feedback changes behaviour. That is a deployment design problem, and most implementations do not have anyone whose job it is to close it.
As for the underlying stakes, Poepsel put it plainly.
"Every people decision is a high-stakes decision. We need to be thoughtful about it so that we can go as fast as we'd like."
The case for AI-assisted continuous feedback is still intact. The assumption that deploying the technology resolves the management problem it was built to solve is not.