Every few years a large organisation publicly announces it is "getting rid of performance ratings" and the rest of the industry spends 18 months debating whether to follow.
The debate usually generates more opinion than evidence. The data is less tidy than either camp wants to admit: ratings produce real harm in some contexts and real clarity in others, depending on how they are designed and used.
The right answer for most organisations in 2026 is neither keep nor kill. It is evolve, with deliberate design choices on what the rating is, what it is linked to, and who sees it.
What the data says
Three findings hold across the research and our client data.
Ratings plus compensation plus forced distribution consistently produces the worst outcomes. When ratings are high-stakes, tied to comp, and forced into a curve, managers spend more time gaming the system than using it. Employee trust drops. Behavioural differentiation reduces over time, not increases. This is the worst of the worst.
Pure "no ratings" models produce a different failure. Without any differentiating signal, ambiguity grows. Managers hedge in feedback, compensation decisions get made in back rooms, and high performers eventually leave because they cannot see how they are seen. This is less bad than the first failure mode but not a good outcome.
The middle path, lightweight ratings, uncoupled from comp, used for development: consistently produces the best behavioural and commercial outcomes. It is also the hardest to implement because it requires discipline.
The seven design choices
Whatever direction you go, seven decisions matter.
1. How many levels? Three to four is the sweet spot. Five or more creates noise without signal. Two is too coarse.
2. What does the rating mean? "Against target for role" is cleaner than "against peers". Comparative ratings generate more political friction than they are worth.
3. Who is rated? Everyone, or a risk pool? Most organisations over-rate, rating every employee every quarter when 85% of them are sitting comfortably at target.
4. How often? Quarterly calibration produces better signal than annual. See replacing the annual review with quarterly calibration.
5. Who sees it? The employee, their manager, the skip-level, HR. That is it. Broader sharing creates politics without improving decisions.
6. What is it coupled to? This is the key question. Compensation coupling is where most harm comes from.
7. Where is calibration done? In isolation by one manager, or peer-calibrated? Peer calibration is the single biggest driver of quality.
The three realistic paths
Based on those design choices, three paths make sense in 2026.
Path A: Lightweight quarterly rating, uncoupled from comp, used for development. Three-level scale ("on track / ahead / needs attention"). Quarterly calibration. No forced distribution. Comp decisions made through a separate process that references the rating history but is not mechanically tied. This is the path we recommend for most organisations.
Path B: Ratings only for two populations, high performers and underperformers. The middle 70% is simply "on track, nothing to formally capture this quarter." This reduces the cost of the system and focuses manager attention where it matters. Works best in mature performance cultures.
Path C: No ratings, rigorous calibration conversations. This works in small to mid-size organisations with strong manager trust. It fails at scale because the signal disappears. Only pursue if you have the manager bench depth to sustain the discipline.
Pick one. Commit to it for three years. The worst outcome is switching paths every 18 months.
What to stop doing
Four common mistakes to cut:
Forced distribution. The single biggest corrosive practice. Unless your organisation is small enough that a curve is statistically defensible, do not force one.
Secret ratings. If the employee does not see their rating, the rating exists for HR comfort, not performance. Transparency has costs; opacity has bigger ones.
Two ratings (performance and potential) rolled into one compound number. These measure different things. Combining them produces a number that means neither.
Annual-only rhythm. The evidence that annual rating cycles drift toward noise is now overwhelming. Quarterly at minimum.
The comp coupling question
The hardest decision in the list.
Fully decoupling rating from comp is the right answer in principle. In practice, most organisations cannot do it in one step, because rating data is where comp decisions have historically been made.
A staged approach works. Over three years:
Year 1. Keep the coupling but weaken it. Ratings remain a direct input to merit but are one of several. Calibration conversations happen first, comp conversations happen 60 days later.
Year 2. Introduce a second input into the comp decision (market data, role scope) with equal weighting. Rating is now 50% of the input.
Year 3. Make rating a diagnostic input only. Comp decisions are made through a lightweight merit framework with market, scope, and performance contribution as three inputs.
By year three, the rating serves development primarily. Comp decisions are better calibrated because they have more inputs than one number.
Where to start
Most organisations are Path A candidates but currently run something closer to annual forced-distribution ratings coupled to comp. The gap is meaningful. Plan the move over three years, not one.