Appraisal

Most appraisal systems are asked to do three things in one conversation.

• Allocate. Produce a rating that drives the increment, the bonus and the promotion decision.
• Develop. Identify what the person should work on and how.
• Document. Create a record of performance that can be relied on later.

The first defeats the second. An employee who knows that this conversation sets their increment has every reason to present their year as successfully as possible, and none to volunteer what they struggled with. The manager, knowing the rating carries money, has an incentive to soften an honest assessment or to justify a predetermined one.

The common remedy is to separate the conversations: development discussed regularly through the year, allocation decided in its own cycle. That is right, and it is undermined the moment the development conversation is minuted on the same form.

The third job is the one nobody designs for and everybody needs. When performance is disputed at exit, the appraisal record is the evidence, and a file full of ratings inflated to avoid difficult conversations is worse than no file.

• Expectations set in advance, in writing, and specific enough that two people would agree whether they were met.
• Evidence recorded through the period rather than recalled at the end. Recency bias is the strongest distortion in appraisal, and contemporaneous notes are the only real defence against it.
• Calibration across managers before ratings are finalised, on real cases, so a rating means the same thing in two teams.
• The employee's own account taken and recorded, including where it differs.
• A reason given for the rating, referring to the evidence rather than to a general impression.
• A route to challenge it that does not run through the person who gave it.

A rating that fails these is not usable for the third job, and it is often not usable for the first either, because an increment defended by an assessment nobody can explain generates exactly the grievance the process was meant to avoid.

Many organisations require ratings to fit a distribution: a fixed share in the top band, a fixed share at the bottom. The stated reason is to prevent grade inflation. The real reason is usually paybill control.

Both are legitimate aims and neither is served well by disguising them as an assessment. The costs are specific.

• In a small team, the distribution forces a bottom-band rating regardless of actual performance, which is arbitrary at the individual level and known to be.
• It sets colleagues against each other, since one person's higher rating requires another's lower one.
• It corrupts the record. A rating assigned to meet a quota is not evidence of anything, which matters when it is later relied on.
• It teaches managers to game allocation across cycles rather than to assess.

If the constraint is budget, the honest structure is to assess honestly and then allocate a fixed pot against the assessments. That keeps the record clean, makes the constraint visible as a constraint, and stops the assessment carrying the weight of a financial decision it was not designed for.

The annual cycle is a convention rather than a requirement, and its weaknesses are well known: too infrequent to guide anyone, too distant from the work to be accurate, and heavily weighted towards the last quarter.

The alternatives are variations on more frequent and lighter.

The last row is the one organisations have oscillated on. Removing ratings solves the distortion and removes the record, and organisations that have gone that way and then faced a disputed termination have generally reintroduced some form of documentation.

Whatever the cycle, the thing that determines whether it works is whether managers have the conversations between the reviews. A process cannot manufacture feedback that the organisation does not otherwise give.

Appraisal records become material when employment ends badly, and this is the use nobody writes them for.

• A termination for poor performance, defended by a file of satisfactory ratings, is a difficult position. Where standing orders apply, the disciplinary and performance procedures they set out govern, and an appraisal record inconsistent with the reason given for termination undermines it.
• A performance improvement plan should be a documented process with stated expectations, support offered, a defined period and a recorded outcome, not a formality performed shortly before a decision already made.
• Where a grievance alleges that a rating was unfair or discriminatory, the evidence is the rating record and the calibration data. Aggregate rating distributions by group are worth reviewing for the same reason.

The practical instruction is short: write the rating and its reasons as though someone outside the organisation might read them, because occasionally someone does.