Performance Calibration Meeting: How to Run It Well

In short

In a performance calibration meeting, managers compare draft ratings before they become final. They confirm a shared standard, review the spread of ratings by level and change individual ratings with a recorded reason. In a field study at a large multinational organisation, calibration committees made ratings more consistent across raters and curbed leniency, but pushed more ratings towards the middle. We recommend pre-read data, evidence before opinions, a neutral facilitator and guideline ranges, not a forced curve.

Key takeaways

A calibration meeting compares draft ratings from different raters against one shared standard; it does not replace careful appraisal or remove rater bias.
At one large organisation, calibration committees adjusted 25% of ratings, more often downwards than upwards; leniency fell, but central tendency rose.
Prepare a written standard for each rating and level, a level × rating matrix and rater statistics, so the meeting debates evidence, not impressions.
Review the top and the bottom of the matrix explicitly, vary the order of cases and ask for information that only one person holds.
Forced curves and guideline ranges both have costs. As of 2020, no firm-level field experiment had tested forced distribution; guideline ranges went with lower commitment in German panel data.
Record every change with its reason, explain final ratings through criteria rather than rankings, and in Germany check works council rights early.

What a calibration meeting is for, and what it cannot fix

A performance calibration meeting sits between draft and final ratings. Managers who rated people at the same level put their ratings side by side and ask: do we mean the same thing by the same rating? The aim is a shared standard, not a predetermined curve. A talent review is related but also covers potential.

Definition

Performance calibration meeting A facilitated meeting in which several raters compare draft performance ratings against a shared standard and change individual ratings, with a recorded reason, before they become final. Researchers call the group a calibration committee.

What it can do. Reveal that raters apply different standards, pool information no single manager holds, and make every rater show evidence.
What it cannot fix. Performance nobody observed, or rater bias at source; see rater bias in performance reviews.
What it is not. A forced ranking exercise or a bonus negotiation. Turning ratings into pay is covered in linking performance ratings to bonus.

What research on calibration committees shows

Calibration was already common in large organisations in 2011 but is thinly researched. In a poll of SHRM members at organisations with 2,500 or more employees, 54% said they ran formal calibration; 39% cited fitting ratings to a planned distribution as a reason for changes (Hastings, 2011). That does not show today’s practice. Three field studies say more, each with clear limits:

Three field studies of calibration meetings: what they found and where their limits lie.
Study	Data	Finding	Limit
Demeré, Sedatole and Woods (2019)	A large multinational organisation	25% of ratings adjusted, more often and further downwards; ratings from supervisors who rated above average were adjusted down, those from supervisors below average up. More consistency, less leniency, more central tendency.	One organisation
Grabner, Künneke and Moers (2020)	Panel data from a professional-services firm	Supervisors who inflated ratings strategically got a lower rating themselves; those with less compressed, more informative ratings were more likely to be promoted.	One firm
Speer, Tenbrink and Schwendeman (2019)	Two field studies	Format and cadence varied widely. Meetings went with a stronger perceived shared standard, accountability and information; post-calibration ratings correlated more with other performance measures.	Correlational; no ideal format

One finding changes how you run the meeting: in the study by Demeré, Sedatole and Woods, calibration curbed leniency but made central tendency worse. In one firm’s data analysed by Bol (2011), central tendency went with weaker later improvement, while leniency went with better future performance; these are correlations, not causes. A meeting that only trims generous ratings may therefore work on the wrong problem.

Preparation: the level × rating matrix and pre-reads

Many calibration problems start before the meeting: without pre-reads, the group debates impressions. In Bol’s study, high information-gathering costs went with more leniency and more central tendency; pre-reads should lower those costs (our inference). Send five things in advance:

A written standard for each rating and level. Describe what “exceeds expectations” means for a consultant and for a project lead, with behavioural examples. Frame-of-reference training brought raters’ picture of performance closer to a reference standard, which went with more accurate ratings (Gorman and Rentsch, 2009).
The level × rating matrix. One page per level: rows are the rating steps; each cell lists name, draft rating and rater. The group sees where ratings cluster.
Statistics for each rater. Mean and spread compared with colleagues who rated the same people. Flag only clear deviations.
Evidence for each person. Two or three concrete observations with project and period, especially for ratings at either end of the scale.
A rule on what stays out. Development feedback collected in confidence, such as a 360°, stays out of the meeting; why is covered in linking performance ratings to bonus.

Agenda, roles and facilitation

A calibration meeting needs clear roles. In one field study the committee also judged the supervisors’ rating behaviour (Grabner et al., 2020); in another, meetings went with higher perceived accountability (Speer et al., 2019). The advice below is our inference from these studies and from general judgement research, not a tested protocol.

Facilitator. Someone from HR or a neutral partner with no cases in the group. They hold the standard and the clock, ask for evidence and record every change with its reason.
Raters. The managers who gave the draft ratings, usually project leads and partners, in one meeting per level so everyone presents their own cases.
Decision owner. Often someone from firm leadership, who settles disagreements and signs off each level’s distribution.

Confirm the standard. Read the rating definitions before anyone names a person.
Look at the spread. The matrix and rater statistics show where standards diverge.
Discuss both ends. Review the top and the bottom of the matrix explicitly, as well as the borderline cases. Checking only the top curbs leniency and may let the middle grow.
Evidence before opinions. The rater presents first. Then ask who else worked with this person and knows something not yet said.
Vary the order. Do not open with the strongest or weakest case, and change the order between levels, so no case becomes the anchor.
Close. Read out every change and give each open case an owner and a date.

Distribution guidelines vs forced ranking

Does calibration need a target distribution? Forced distribution, often loosely called forced ranking, sets fixed shares for each rating. A recommended distribution sets guideline ranges that raters may depart from with a reason. Calibration can also run without one.

Three ways to handle the distribution of ratings (a summary of studies and practice, not a measurement).
Model	What it requires	What studies show	Risk
Forced distribution	Fixed shares for each rating	In a lab experiment, about 6–12% higher productivity, but harmful once sabotage was possible (Berger, Harbring and Sliwka, 2013); gains from firing the bottom group were only simulated (Scullen, Bergey and Aiman-Smith, 2005)	Small units squeezed into a curve; an incentive to undermine colleagues
Recommended distribution (guideline ranges)	Target ranges for each rating; departures need a reason	Lower organisational commitment, mainly among employees without managerial responsibility (German panel data; Loberg et al., 2024)	Becomes a quota if departures are never accepted
Calibration without a distribution	A shared standard and evidence	Less leniency, but more central tendency (one organisation; Demeré et al., 2019)	Ratings bunch in the middle unless both ends are reviewed

For consultancies, three points argue against a fixed curve. Units are small: whether a level of eight consultants must contain exactly one top rating is an assumption, not an observation. Project work depends on cooperation, and in the experiment forced distribution did harm once sabotage was possible. And Sliwka’s 2020 review found no firm-level field experiment on its costs and benefits.

Failure modes: politics, the loudest voice, anchoring on the first case

A calibration meeting is a group decision with biases of its own. Most findings below come from lab and sports-judging studies, not calibration meetings, but describe mechanisms that can operate in any meeting.

Anchoring on the first case. People adjust too little from a starting value (Tversky and Kahneman, 1974), and sequential performance judgements are pulled towards or pushed away from the previous one (Damisch, Mussweiler and Plessner, 2006).
Shared information wins. Groups tend to discuss what everyone already knows and what supports existing preferences; what only one person knows gets lost (Stasser and Titus, 1985).
The loudest voice. Seniority and negotiating skill can carry a case further than the evidence; evidence before opinions and a neutral facilitator counter it.
The safe middle. Moving doubtful cases to the middle avoids conflict and loses information.
Budget instead of performance. Once money is on the table, ratings can bend to the bonus pool. Keep “how good?” separate from “how much?”.

After the meeting: documentation, communication and employee rights

The meeting is finished only when changes are documented and explained. Also give raters feedback on how they rate: in the study by Demeré, Sedatole and Woods, supervisors learned the organisation’s rating expectations through the committee’s adjustments.

The manager speaks, not the committee. The person who rated delivers the final rating, even if the meeting changed it, and explains it through criteria and evidence, not the debate or a ranking.
No quota as an explanation. “The distribution left no room” devalues every rating. If guideline ranges played a part, say so.
A route for objections (Germany). § 82(2) of the Works Constitution Act (BetrVG) lets employees ask for their performance appraisal to be discussed with them and bring a works council member. Not legal advice.

In the EU, and especially in Germany, check three points early:

No decision based solely on automated processing (EU). Article 22(1) GDPR gives people the right not to be subject to a decision based solely on automated processing that produces legal effects concerning them or similarly significantly affects them; Article 22(2) sets out exceptions. A meeting in which people decide, with reasons, supports the position that no such decision is made; a software proposal routinely adopted unchanged weakens it. For credit scores, the CJEU held that automatically generating a score is itself an automated decision where a third party draws strongly on it to establish, implement or terminate a contract with that person (C-634/21); whether this extends to ratings is open, and would be an analogy, not a ruling.
Appraisal principles (Germany). Under § 94(2) BetrVG, general appraisal principles need the works council’s agreement, and the Federal Labour Court has extended this to the design of the appraisal procedure (BAG, 1 ABR 48/13). Since calibration changes ratings, its rules arguably fall under it; that is our reading, not a ruling. We recommend agreeing them in a works agreement.
Software and pay (Germany). Calibration software can be a technical device under § 87(1) no. 6 BetrVG if it is objectively suitable for monitoring (BAG, 1 ABR 7/15). If pay follows the ratings, § 87(1) no. 10 (pay structure) may apply too.

How ValueScore supports calibration

ValueScore is built by LEADBeyond for consultancies and professional-services firms. For the calibration meeting, it shows the level × rating matrix on one page, rater effects with a confidence interval, and a cross-project view of one consultant as rated by several project leads. Leniency and severity are detected statistically and corrected. We recommend that people still decide in the meeting and record their reasons.

Frequently asked questions

What is a calibration meeting in HR?

A facilitated meeting in which managers compare draft performance ratings against a shared standard before they become final. They review the spread by level, discuss cases on the evidence and change individual ratings with a recorded reason. The aim is comparability across raters, not a predetermined curve.

What happens in a calibration meeting?

Raters present draft ratings for one level on a matrix of rating steps. The group sees where ratings cluster, compares each rater’s average with colleagues who rated the same people, and discusses the top, bottom and borderline cases. At one large organisation, calibration committees adjusted 25% of ratings (Demeré et al., 2019).

Who attends a performance calibration meeting?

The managers who rated the people at that level; in a consultancy, usually project leads and partners. Add a facilitator from HR, or a neutral partner who rated nobody in the group, and a decision owner for disagreements.

How do you run a calibration session?

Agree ground rules first: read the rating definitions before anyone names a person, hear the rater’s evidence before any view, never argue from a quota or the bonus pool, and record every change with its reason. A facilitator with no ratings of their own enforces them; the standard, matrix and rater statistics go out as a pre-read.

What is the difference between calibration and forced ranking?

Calibration aligns standards: a rating changes when the evidence supports it. Forced ranking, or forced distribution, prescribes fixed shares for each rating, however performance is actually spread. Between the two sit guideline ranges that raters may depart from with a reason.

Does calibration reduce rating inflation?

It can. At one large organisation, committees curbed leniency and lowered ratings from supervisors who rated above average, but central tendency increased (Demeré, Sedatole and Woods, 2019). In a professional-services firm, supervisors who inflated ratings strategically got a lower rating themselves (Grabner et al., 2020). We recommend reviewing both ends of the scale.

What should a calibration meeting agenda include?

Six steps. Confirm the rating standard. Review the matrix and the rater statistics. Discuss the top, the bottom and the borderline cases, hearing the rater’s evidence first and varying the order of cases. Close by reading out every change and giving each open case an owner and a date.

How do you communicate calibrated ratings to employees?

The manager who rated delivers the final result, even if the meeting changed it, and explains it through criteria and evidence, not the debate or a quota. In Germany, employees can ask for their appraisal to be discussed with them and bring a works council member (§ 82(2) BetrVG). Not legal advice.

Sources

Hastings, R. (2011, December 15). Survey: Most large firms calibrate performance. SHRM., Society for Human Resource Management (2011). SHRM news item on a member poll run with APT Metrics (510 members at organisations with 2,500 or more employees). Cited for the 54% (formal calibration) and the 39% (fitting a planned distribution, given as a reason for changes). Not cited for how common calibration is today or in any particular country.
Demeré, B. W., Sedatole, K. L., & Woods, A. (2019). The role of calibration committees in subjective performance evaluation systems. Management Science, 65(4), 1562–1585., INFORMS (2019). Peer-reviewed field study using internal data from a large multinational organisation. Cited (from the abstract) for the 25% adjustment rate, more frequent and larger downward adjustments, the adjustment of ratings from supervisors above and below average, more consistency, less leniency, more central tendency, and supervisor learning. One organisation, so transfer is open.
Grabner, I., Künneke, J., & Moers, F. (2020). How calibration committees can mitigate performance evaluation bias: An analysis of implicit incentives. The Accounting Review, 95(6), 213–233., American Accounting Association (2020). Peer-reviewed study using panel data from a professional-services firm. Cited for the committee lowering the own rating of supervisors who inflated ratings strategically and for less compressed ratings going with a higher likelihood of promotion. A single firm; not generalised to all consultancies.
Speer, A. B., Tenbrink, A. P., & Schwendeman, M. G. (2019). Let’s talk it out: The effects of calibration meetings on performance ratings. Human Performance, 32(3–4), 107–128., Taylor & Francis (2019). Peer-reviewed paper with two field studies. Cited for the wide variation in meeting format and cadence, the association with higher perceived frame of reference, accountability and information, and post-calibration ratings correlating more with other performance measures. Correlational; no causal proof and no recommended format.
Bol, J. C. (2011). The determinants and performance effects of managers’ performance evaluation biases. The Accounting Review, 86(5), 1549–1575., American Accounting Association (2011). Peer-reviewed study using data from one firm. Cited for information-gathering costs and close relationships going with more central tendency and leniency, central tendency going with weaker later improvement and leniency with better future performance; correlations, not causes. That pre-reads lower the costs is our inference. Not a study of calibration meetings.
Gorman, C. A., & Rentsch, J. R. (2009). Evaluating frame-of-reference rater training effectiveness using performance schema accuracy. Journal of Applied Psychology, 94(5), 1336–1344., American Psychological Association (2009). Peer-reviewed study of frame-of-reference training. Cited (from the abstract) for trained raters holding a picture of performance closer to the reference standard, which in turn went with more accurate ratings. No finding about calibration meetings; applying it to a written standard before the meeting is our inference.
Loberg, L., Nüesch, S., & Allendorf, A. M. (2024). Unwanted recommendation: The dark side of recommendations for performance rating distribution on employees’ emotions, attitudes, and behaviors. Zeitschrift für Arbeits- und Organisationspsychologie, 68(1), 36–48., Hogrefe (open access, CC BY 4.0) (2024). Peer-reviewed study using panel data from Germany’s Federal Employment Agency. Cited (from the abstract) for a recommended distribution going with lower organisational commitment, satisfaction, turnover intention and cooperation being unaffected in the whole sample, and the effects appearing among employees without managerial responsibility. No findings for private consultancies.
Berger, J., Harbring, C., & Sliwka, D. (2013). Performance appraisals and the impact of forced distribution: An experimental investigation. Management Science, 59(1), 54–68., INFORMS (2013). Peer-reviewed real-effort lab experiment. Cited for productivity about 6% to 12% higher under forced distribution, the less clear effect with prior experience, and the harm when sabotage was possible. Not a field result and not knowledge work.
Scullen, S. E., Bergey, P. K., & Aiman-Smith, L. (2005). Forced distribution rating systems and the improvement of workforce potential: A baseline simulation. Personnel Psychology, 58(1), 1–32., Wiley (2005). Peer-reviewed simulation. Cited only for gains from firing the lowest-rated share each year being simulated, appearing mostly in the first years and depending on the share fired and on voluntary turnover. Not evidence of real performance gains.
Sliwka, D. (2020). Bonuses and performance evaluations. IZA World of Labor, 478., IZA – Institute of Labor Economics (2020). Review article. Cited only for the point that, at the time, there was lab evidence on forced distributions but no firm-level field experiment on their costs and benefits. Not cited for any recommended distribution.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131., AAAS (1974). Classic paper on judgement heuristics. Cited only for insufficient adjustment from an initial value (anchoring). No evidence from calibration meetings.
Damisch, L., Mussweiler, T., & Plessner, H. (2006). Olympic medals as fruits of comparison? Assimilation and contrast in sequential performance judgments. Journal of Experimental Psychology: Applied, 12(3), 166–178., American Psychological Association (2006). Peer-reviewed study using 2004 Olympic Games data and lab data. Cited for sequential performance judgements being influenced by the previous one, towards it or away from it depending on perceived similarity. No workplace data.
Stasser, G., & Titus, W. (1985). Pooling of unshared information in group decision making: Biased information sampling during discussion. Journal of Personality and Social Psychology, 48(6), 1467–1478., American Psychological Association (1985). Peer-reviewed lab experiment on group decisions. Cited for discussion being dominated by shared information and information that supports existing preferences, with information held by individuals getting lost. No evidence from calibration meetings.
Works Constitution Act (Betriebsverfassungsgesetz, BetrVG), as promulgated on 25 September 2001 (BGBl. I p. 2518), § 82: Employee’s right to be heard and to discuss., Federal Ministry of Justice, gesetze-im-internet.de (2001). German statutory text. Cited for the right to have one’s performance appraisal discussed and to bring a works council member (subsection 2). Not legal advice.
Works Constitution Act (Betriebsverfassungsgesetz, BetrVG), as promulgated on 25 September 2001 (BGBl. I p. 2518), § 94: Personnel questionnaires, appraisal principles., Federal Ministry of Justice, gesetze-im-internet.de (2001). German statutory text. Cited for works council agreement on general appraisal principles (subsection 2). That calibration rules may fall under it is our reading, not a ruling. Not legal advice.
Federal Labour Court (Bundesarbeitsgericht). (2015). Order of 17 March 2015, 1 ABR 48/13, ECLI:DE:BAG:2015:170315.B.1ABR48.13.0., Bundesarbeitsgericht (2015). German decision. Cited for its headnote that co-determination under § 94(2) BetrVG extends to the design of the appraisal procedure. The case concerned appraisal interviews, not calibration meetings. Not legal advice.
Works Constitution Act (Betriebsverfassungsgesetz, BetrVG), as promulgated on 25 September 2001 (BGBl. I p. 2518), § 87: Co-determination rights., Federal Ministry of Justice, gesetze-im-internet.de (2001). German statutory text. Cited for subsection 1 no. 6 (technical devices for monitoring conduct or performance) and no. 10 (pay structure in the establishment). Not legal advice.
Federal Labour Court (Bundesarbeitsgericht). (2016). Order of 13 December 2016, 1 ABR 7/15, ECLI:DE:BAG:2016:131216.B.1ABR7.15.0., Bundesarbeitsgericht (2016). German decision. Cited only for the principle (para. 22) that technical devices are intended for monitoring if they are objectively suitable for it, regardless of the employer’s intention. The case concerned a Facebook page, not calibration software. Not legal advice.
Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 (General Data Protection Regulation), Article 22. Official Journal of the European Union, L 119, 1–88., Publications Office of the European Union, EUR-Lex (2016). EU regulation. Cited for Article 22(1), the right not to be subject to a decision based solely on automated processing that produces legal effects or similarly significantly affects a person, and for the exceptions in Article 22(2). Not legal advice.
Court of Justice of the European Union. (2023). Judgment of 7 December 2023, C-634/21, OQ v Land Hessen (SCHUFA Holding (Scoring)), ECLI:EU:C:2023:957., Publications Office of the European Union, EUR-Lex (2023). Cited for an automated decision existing where a third party draws strongly on an automatically generated probability value. The case concerned credit scoring; applying it to performance ratings is an analogy, not a ruling. Not legal advice.