Every few years a company decides its performance review process is broken and sets out to fix it. New rubric, new ratings scale, new software, occasionally a bold decision to abolish reviews entirely.

A year later satisfaction is roughly where it started.

That's because the problem was never the instrument. It's that the annual review is being asked to serve four separate purposes that actively undermine one another, and no form design can resolve a contradiction.

The four jobs

1. Deciding pay. A backward-looking judgment about relative contribution, used to allocate a finite pool.

2. Helping someone improve. A forward-looking developmental conversation about weaknesses and growth.

3. Documenting problems. A legal and procedural record, mostly relevant for the small number of situations heading toward separation.

4. Telling someone where they stand. The plain human question of whether they're doing well and whether their future here is bright.

Now notice what happens when you run these simultaneously.

Job 1 makes job 2 impossible. The moment someone knows their raise is being decided in this conversation, they are — correctly — in a negotiation. Nobody explores their genuine weaknesses with the person holding their compensation. They advocate. You'd do the same.

Job 3 makes job 4 dishonest. Because documentation matters mainly in adversarial situations, managers write carefully, and careful writing reads as hedged. The employee, trying to decode where they actually stand, gets prose engineered for a lawyer.

Job 2 gets crowded out by all of them. Development is the least urgent and most easily deferred, so it becomes a paragraph at the end that nobody acts on.

The result is a conversation that decides pay adequately and does nothing else well.

Pull them apart

The fix isn't a better review. It's separating the conversations so each can be honest.

Pay gets its own conversation, and it should be short. Here's the decision, here's the reasoning, here's what would change it. Ten minutes. It's a decision being communicated, not a discussion. Trying to soften it by wrapping it in developmental language just makes people distrust the developmental language.

Development happens continuously and separately, and it is explicitly not tied to compensation. Say that out loud, repeatedly, because people won't believe it at first. The value of this conversation is entirely dependent on whether the person feels safe being honest about what they're bad at, and that safety takes months of consistent behavior to establish and one contradicting incident to destroy.

Documentation is a management practice, not an annual event. If someone is genuinely struggling, that's a live conversation happening now with a clear plan — not something you're carefully wording in December to protect the company later. A performance problem first documented at annual review time is a management failure that predates the paperwork.

Where someone stands should never be a surprise. If an employee learns their standing at an annual review, the review isn't the problem. Nobody has been telling them for eleven months.

The rating scale question

Most review redesigns spend their energy here — five points or three, forced distribution or not, numbers or labels.

It matters far less than people think, with one exception: whatever the scale, people will decode it into "am I safe, am I doing fine, or am I on a track to something better." Three states. If your five-point scale collapses into three in practice, you may as well use three and be honest about it.

Forced distribution is worth a specific warning. It solves a real problem — rating inflation makes reviews meaningless — by creating a worse one: it puts managers in competition on behalf of their people, which poisons exactly the cross-functional cooperation that growing companies need most. I'd rather fix inflation by having leaders calibrate together in a room, out loud, than by imposing a curve.

What I'd actually run

For a company in the range I work with most — past the informal stage, not yet large enough for real HR machinery:

  • Weekly or biweekly one-on-ones that are genuinely about the work and where feedback happens in the moment, small and specific.
  • A twice-yearly conversation about direction: where is this person going, what do they need, what's the honest gap. No ratings, no pay, no forms.
  • An annual compensation decision, communicated clearly and briefly, with the reasoning stated.
  • Immediate, direct conversation whenever there's a real performance problem, with a plan and a timeline, on the day it becomes clear.

That's more total conversation and dramatically less process. Most of the administrative weight in review systems exists to compensate for managers not having these conversations naturally — and it doesn't actually compensate. It just produces a record of the failure.

The uncomfortable root cause

Every over-engineered performance process I've encountered was built to work around managers who won't say hard things directly.

That's the real problem, and no form solves it. If your managers can't tell someone clearly that they're falling short, adding a rubric doesn't create the conversation — it creates a document where the conversation should have been, and everyone can tell the difference.

Fix that first, and you'll find you need much less process than you thought.