What is changing in performance management

Performance systems are rebuilt roughly every few years, and the same five moves keep coming round. Each one arrives described as modern and each one is a response to a real defect in whatever preceded it.

This is not a prediction about 2027. It is a way of reading a proposal, because the useful question is never whether a design is current. It is which defect it fixes and what it costs to fix it that way.

What to take away

• Every performance system design trades one distortion for another. There is no neutral option, and a proposal that claims to have no downside has not been examined.
• The cost of any of these lands on managers as hours and on everyone else as uncertainty during the transition. Neither appears in the case for the change.
• Ask what the previous design was for. Most redesigns remove a mechanism that was solving something, and nobody remembers what.

Move one: removing numerical ratings

What it answers. Ratings compress a year into one figure, and the figure becomes the whole conversation. People argue about the number instead of about the work.

What it costs. Ratings were doing a job: they forced a decision and they made comparisons visible enough to be challenged. Remove them and the comparison still happens, in private, with nothing written down. Pay decisions do not disappear when the rating does.

What to watch. Whether an informal shadow rating appears within a year. It usually does, and it is worse than the formal one because it cannot be appealed.

Move two: replacing the annual review with continuous check-ins

What it answers. An annual conversation is too late to change anything and is dominated by whatever happened in the last two months.

What it costs. Manager hours, considerably more than the estimate. It also removes the forcing function: an annual event guarantees that a difficult conversation happens at least once, and a continuous system guarantees only that it could.

What to watch. Whether the difficult conversations still happen. Continuous systems are excellent for the routine cases and they let the hard ones be deferred indefinitely, one check-in at a time.

Move three: adding calibration sessions

What it answers. Two managers applying the same standard differently, so that an assessment describes the assessor.

What it costs. Time, and a subtler thing: the assessment moves away from the person who observed the work toward a room that did not. Calibration improves consistency and reduces accuracy, and both of those are real.

What to watch. Whether outcomes are being moved to fit a distribution. Consistency work is legitimate; adjusting individuals so the shape looks right is a different activity wearing the same name.

Move four: adding peer or upward input

What it answers. A manager sees a fraction of someone's work, particularly in distributed or cross-team settings.

What it costs. It is expensive to collect and it is only honest under conditions that are hard to guarantee. Input that can be traced back to its author will be polite, and input that cannot be traced will occasionally be unfair with no way to test it.

What to watch. Whether it is used for development or for evaluation. It is genuinely useful for the first and it degrades quickly under the second, because the moment it affects an outcome, people start managing what their peers will say.

Move five: formalizing objective setting

What it answers. People working hard on things nobody needed, and no shared view of what good looks like.

What it costs. Whatever is written becomes the definition of the job, and the unwritten parts stop being done. It also creates an incentive to set objectives that can be met, which is a rational response and destroys the information content of meeting them.

What to watch. Whether objectives are being negotiated downward at the point of setting. That is the tell, and it is the ordinary consequence of using an indicator for decisions, described as Campbell's law.

The test to apply to any of them

Four questions, answered in writing before agreeing.

What specific defect in the current system is this fixing, with an example from the last year? A proposal that cannot produce an example is fixing a reputation rather than a problem.

What was the removed mechanism doing? Every one of these five removes something. Naming it does not block the change; it tells you what has to be replaced.

Who pays for it, in hours, and when? These changes cost manager time in the first two cycles, heavily, and the estimate is always optimistic.

What would tell us in a year that this was worse? Write it down now. Afterward, any outcome will be read as evidence for whichever design the reader preferred.

What does not change

Underneath all five moves, three things are constant and no design touches them.

Someone has to say the difficult thing out loud, to the person, in time for them to do something about it. No system does this and every redesign is partly an attempt to find one that will.

An assessment can only be as good as the record it is built from. A new framework applied to no contemporaneous notes produces the same reconstructed judgments in a different template, and the obligation to keep that record sits with the manager rather than with the system, which is the argument in management foundations.

And people respond to what is measured. Any of these designs will reshape behavior around whatever it counts, which is a reason to count few things carefully rather than many things loosely. The general treatment of that is in operating processes metrics, the version specific to teams is in team management metrics, and the underlying separation of feedback from evaluation is in performance management. Guidance from the US Office of Personnel Management on the planning stage of a performance cycle treats expectation setting as the part that carries the weight, which is the opposite emphasis from most redesign proposals.

Common questions

Is any of these five better than the others?

They answer different defects, so the question is which defect you have. The one thing worth saying generally is that doing several at once makes the result uninterpretable, and organizations do exactly that regularly.

Our system is being changed and we were not asked. What is worth contributing?

The hours estimate and the removed mechanism. Both are concrete, neither reads as resistance, and they are the two things the designers are least likely to have.

How long before you can tell whether a redesign worked?

Two full cycles. One cycle is all transition, and the transition tells you about the rollout rather than about the design.

Should we just keep what we have?

Sometimes, and it is a legitimate answer that is rarely offered. A mediocre system that people understand and use consistently outperforms a better one halfway through its second migration.