Absolute Ratings

The bottom row is the crux. Absolute ratings answer the question a manager and an employee actually care about. Relative ranking answers the question the compensation budget poses. Organisations that need both, which is most of them, keep drifting between the two and settling on a compromise that does neither well.

The stable arrangement is to rate absolutely and allocate separately: assess each person against the standard, then distribute a fixed pot against those assessments. The rating stays honest and the budget stays controlled, at the cost of admitting openly that a high rating does not guarantee a large increase.

An absolute rating is only as good as the standard it measures against, and most scales fail because the levels are labels rather than descriptions.

• Describe behaviour at each level, not adjectives. Exceeds expectations means nothing; delivers work that others rely on without rework, and improves how the team does it, can be assessed.
• Keep the scale short. Five levels are argued over; four or three are usually enough and reduce the pretence of precision.
• Avoid a middle that is treated as failure. Where the central rating denotes solid performance and the organisation treats it as a disappointment, managers stop using it and the scale collapses upward.
• Separate what from how where the organisation cares about both, and be explicit about the weight each carries.
• Anchor the levels with examples from the actual work, refreshed periodically.

The test is whether two managers reading the descriptions would place the same employee at the same level. Where they would not, the scale is a container for opinion.

The characteristic failure of absolute rating is that standards move without anyone deciding to move them.

• Managers avoid difficult conversations, so ratings creep upward year on year.
• A manager under pressure to retain someone rates them highly to support a counter-offer.
• Different functions develop different implicit bars, so a rating in engineering does not mean what it means in operations.
• Ratings become linked to increments so tightly that the rating is chosen from the desired increase and worked backwards.

Calibration is the standard remedy and it has to be genuine: managers comparing evidence, not negotiating numbers. The diagnostic for whether drift has taken hold is to look at the distribution over three years. If the proportion in the top band has risen steadily and organisational performance has not, the scale is inflating.

The other remedy is to decouple rating from increment enough that the rating is not worth gaming. Where the two are welded together, no amount of calibration will hold the standard.

They fit where the work is not zero-sum and where the standard is describable.

• Small teams, where relative ranking is arbitrary and known to be.
• Roles with clear output standards, where meeting the standard is the point rather than beating a colleague.
• Collaborative work, where relative ranking actively discourages the behaviour the organisation wants.
• Organisations that need a defensible record, since a rating against a described standard explains itself in a way that a rank does not.

They fit poorly where the budget genuinely requires differentiation and the organisation is unwilling to make that constraint explicit. In that situation absolute ratings will be quietly overridden by allocation, and the visible result is a rating system nobody trusts.

The honest version, in that case, is to say plainly that the assessment establishes performance and a separate, budget-constrained decision establishes pay. Employees accept that far more readily than a rating that was adjusted after the fact for reasons nobody will name.