The Problems With Annual Performance Reviews

The annual performance review is one of the most disliked rituals in corporate life, and most attempts to fix it have made it longer rather than better.

The important point is that its problems are structural, not executional. They do not come from managers doing it badly. They come from asking a once-a-year conversation to carry four jobs that pull against each other.

The four jobs it is asked to do

Development and compensation are the fatal pair. Nobody discusses their weaknesses openly in a conversation that sets their pay. So people manage the conversation instead of using it — and then everyone concludes the employee "wasn't receptive to feedback."

Any fix that keeps all four jobs in one annual meeting will fail for the same reason the original does.

The structural problems

The feedback arrives too late to act on. Something noticed in February discussed in December is not feedback, it is history. The behaviour has had ten months to compound.

It relies on memory. A year assessed in an hour is assessed from recall, which is dominated by the recent and the vivid. See recency bias and the four that cost more.

Annual cadence doesn't match anything real. Priorities shift quarterly; projects run for weeks. A twelve-month assessment cycle matches the finance calendar and nothing about how work actually happens.

It is enormously expensive. Manager writing time, employee self-assessment, calibration meetings, HR administration. For a 200-person company this is routinely thousands of hours. Few companies ever calculate it, and fewer ask what it bought.

The ratings often don't predict much. If the largest source of variance is which manager someone has, the number is telling you about raters as much as performers.

What companies tried instead

Abolishing ratings entirely

Several large firms removed numerical ratings in the 2010s. Mixed results, and a consistent failure pattern: compensation still had to be decided. Removing the rating did not remove the decision — it moved it somewhere less visible and less appealable.

Verdict: works only if you also redesign how pay decisions get made and communicated.

Continuous feedback

Frequent, lightweight check-ins replacing the annual set-piece. Directionally right and the most durable of the changes. The failure mode is adding continuous feedback while keeping the annual review intact, which produces more process, not less.

Verdict: works, if it replaces rather than supplements. See continuous performance management.

Separating development from compensation

Two conversations, deliberately months apart. Cheap, unglamorous, and probably the highest return-per-unit-effort change available.

Verdict: works. Do this first, whatever else you decide.

360 feedback

More raters, which genuinely reduces single-rater distortion. Degrades when it becomes a popularity survey or when raters are asked for too much, too formally, too rarely.

Verdict: works in moderation; fails at volume.

Continuous evidence collection

Rather than changing the conversation's frequency, change its input — assemble the record as work happens, so the review interprets evidence rather than recalling it.

Verdict: works, and composes with all the above. See evidence-based performance reviews.

What to do, in order

1. Split development from compensation. Different conversations, different months. Cheapest change, biggest immediate effect on candour.

2. Fix the input before the format. Contemporaneous records remove the memory problem without redesigning anything. Most companies do this backwards and redesign the form first.

3. Shorten the loop for development only. Quarterly or monthly, lightweight. Keep the compensation cycle annual if payroll requires it — the two do not have to match.

4. Only then change ratings. Ratings are the most visible part and the least of the problem. Changing them first is why so many redesigns disappointed.

5. Measure whether it worked. Rating distributions by manager and by group, time from observation to feedback, hours spent on the cycle. If you cannot say whether the new process is better, you have changed the ritual rather than the outcome.

The realistic expectation

No process makes assessing people easy or fully fair. Judgement about human beings under incomplete information is irreducibly hard.

What is achievable: feedback that arrives soon enough to act on, records that outlast memory, and pay decisions that can be explained. That is a meaningfully better system than the annual review, and it is a long way short of solved.

Anyone selling you the fully solved version is selling the same promise the annual review made in the first place.

Related

• Evidence-based performance reviews
• Recency bias in performance reviews
• Continuous performance management