Performance calibration is the step in a performance review cycle where managers and HR compare draft ratings across teams, against the same standards and evidence, before grades are final. That is the HR meaning — not engine or instrument calibration. It catches generous or harsh graders, ratings with thin evidence, and outliers, so people doing similar work at the same level get similar grades.
This guide is for People leaders and HRBPs who are setting up calibration, or buying a tool for it, in a mid-market team that may not have a separate talent function. It extends the performance management framework, where calibration is step 4. The meeting itself is covered in how to run a performance calibration meeting and the calibration meeting template. Disclosure: Ash Taherian is Product Sales Executive at Revolut People, which sells performance management software. Treat product mentions as coming from an interested party.
What is performance calibration?
Performance calibration is the step in a performance review cycle where managers and HR compare draft ratings across teams, against the same standards and evidence, before grades are final. It catches generous or harsh graders, ratings with thin evidence, and outliers, so people doing similar work at the same level get similar grades.
You will also hear “talent calibration”. In this playbook that is the same idea: checking that a grade means the same thing in every team, against a published rating scale and talent bar, not against whichever manager happened to write it. A calibration meeting is how most teams do that work; it is not the whole system. The meeting has a running order and a record. The system — a spreadsheet or a product — is what gathers the drafts, shows the evidence and, once you agree, publishes the grades. If that conversation still happens on a performance-versus-potential grid, see the 9-box grid: how it works, a free template, and when to replace it.
Why scale-ups calibrate
Scale-ups calibrate for three practical reasons: fairness, pay and trust.
Fairness. Once you have several managers, draft grades drift. One manager is generous, another is harsh, and two people doing similar work at the same level leave the cycle with different grades. Calibration is the check that each level on the scale is being used the same way.
Pay. If pay, bonus or equity follows the grade, an uncalibrated grade becomes an uncalibrated rise. That is why calibration belongs in the pay cycle, not as a side meeting. For a wider discussion of linking review outcomes to pay, see Gartner on pay for performance. Once ratings are calibrated, a merit matrix is the usual way to turn a grade and position in range into a recommended increase.
Trust. People notice when one team sails through on top ratings and another is marked hard for the same standard of work. A short, evidence-based calibration is how you keep the scale credible. The customer story How Ebury cut calibration time by 75% states that Ebury replaced disconnected grading spreadsheets and cut departmental calibration from 4 hours to 1 hour. Quote that figure only as the story states it; it is one company’s published result, not a benchmark.
How a calibration cycle works
In the playbook, reviews run quarterly. Managers complete scorecards (about one to two weeks). Grades are then calculated from those scorecards, and the performance team calibrates them before they are final. Senior leadership reviews and approves the grades. Results are discussed in each person’s weekly 1:1, and can lead to a promotion path, development, a performance improvement plan or a pay change. In the playbook’s quarterly cycle, grade calculation and calibration take about one week at most, including senior leadership review.
The meeting itself should not discuss everyone. It should run three checks, which are set out in full in how to run a performance calibration meeting:
| Check | What to look for | What to do |
|---|---|---|
| Generous graders | A manager rates too many people at the top, for example half of one team as A-players. | Ask the manager to identify the strongest performers so the top band stays credible. |
| Thin evidence | Scorecards not filled in, or weak justifications. | Follow up with the manager straight away so the cycle is not delayed. |
| Outliers | Unusual highs, lows, or sudden jumps — for example someone previously graded above bar who receives an Exceptional score in one area. | Examine whether the change is genuine or a grading mistake. |
For the invite, pre-read, agenda and decision log, use the calibration meeting template for mid-market tech companies. For how the five-level scale should be read in the room, see employee performance rating scales.
Calibration methods
You do not need a large talent function to calibrate. You do need a method that matches your size, and a rule that grades change only with evidence.
| Method | How it works | When it fits |
|---|---|---|
| Committee | Managers and HR review a pack of draft grades together, against the talent bar, and agree changes in the room. | The default for most mid-market teams. Keep the session to the three checks, not a tour of every person. |
| Manager pairs | Two managers review each other’s drafts before a wider session, and bring only the remaining questions to the committee. | Useful when the group is small, or when you want a first pass without filling a long meeting. |
| Structured stages | Drafts are reviewed in stages — for example by team and then department, or by role and then function — so the same standard is applied across the organisation. | Fits once you have several departments or a functional structure, and one room can no longer see every draft. |
Software can support those methods without replacing them. Revolut People’s calibration documentation, as checked 13 November 2026, describes three modes: Admin, Manual (with delegated calibrators whose access is limited to the people they are assigned) and Structured. Structured mode supports multi-stage flows, including organisational paths (departments or teams) and functional paths (function, role or specialisation), with up to five primary stages. See the calibration process overview and the structured mode page. That is how one product implements the methods; it is not a reason to buy it, and it is not the only way to run a staged process.
What calibration software should do
The meeting still needs a person in the chair. The software should make the three checks faster to run, and harder to skip. Use this as a buying checklist, then confirm each point in a walkthrough of the product you are evaluating.
| What the software should do | Why it matters | Maps to |
|---|---|---|
| Flags | Surface generous or harsh grading patterns, thin evidence and unusual results automatically, so the room starts with the exceptions. | All three checks |
| An evidence view | Open the scorecard, comments and history beside the draft grade, without hunting through a separate pack. | Thin evidence |
| Distribution views | Show draft ratings side by side by team, role and level, not only as a flat list. | Generous graders, outliers |
| A recommended grade | A starting point the calibrator can accept or change, with the logic visible. It should not silently overwrite a manager’s draft. | A prompt, not a decision |
| Who can change grades | Named roles, with a limited scope for anyone you delegate to. | Governance |
| Locking and publishing | Once grades are agreed, they cannot be quietly edited, and they are published in a controlled order — typically to managers first, then to employees. | Trust |
| A record of changes | Who changed a grade, when, and why. Ask to see this in a walkthrough; do not assume every product keeps a full history. | Trust |
A tool flags; people decide. No product “removes bias” by itself. If a vendor says otherwise, ask who can still change a grade and what is recorded when they do.
Here is how that checklist looks in Revolut People’s own documentation, as checked 13 November 2026. These are descriptions of what the docs say, not a claim that the product is the only way to meet the list, and not a claim that it removes bias.
- Flags. When the calibration stage opens, the system generates calibration flags (the docs say they are populated overnight). They are grouped as concerns, strengths and insights. Concerns include patterns such as line-manager leniency and a grade not supported by deliverables; insights include a first performance review. See calibration flags.
- An evidence view. The Head of Department and Head of Function calibration pages let you open a person’s scorecard and performance history from the calibration table, including reviews, skills, values and bar-raiser signals. See calibrating grades as Head of Department.
- Distribution views. Those same pages include a target distribution and a grade-distribution view for the department or function.
- A recommended grade. Admins can choose one of four recommendation logics: no recommendation, a manager recommendation, a system grade recommendation, or Ranking Grade, which ranks a composite score and maps it to a predefined distribution. The recommendation is a starting point; calibrators assign the final grade. If you use a distribution-based recommendation, treat it as a prompt for the three checks, not a curve you must hit — see common mistakes. The four logics are described in customising the recommended grade.
- Who can change grades. The docs describe Performance Admins, Heads of Department, Heads of Function, and delegated calibrators whose access is limited to the people they are assigned. The Head of Function page notes that, unlike Heads of Department, Heads of Function are not shown a system recommended grade.
- Locking and publishing. Publishing grades states that calibration must be completed before grades can be published. Scheduled cycles can publish on the timeline; in a manual cycle a Performance Admin publishes them. Grades can be published to managers first, then to employees.
- A record of changes. The Head of Department and Head of Function pages describe comments and a calibration justification when a grade is assigned or changed. The public docs reviewed for this page do not describe a separate feature titled an audit trail, so this article does not claim one. If a complete history of who changed what matters to you, ask to see it in a walkthrough.
Questions to ask in a walkthrough
- Can it flag generous graders, thin evidence and outliers without us building those rules in a spreadsheet?
- Can we open the evidence next to the draft grade?
- Can we see ratings by team, role and level, not only in a flat list?
- If there is a recommended grade, can we see how it was produced — and can we change it with a reason?
- Who is allowed to change a grade, and is that scoped by team or role?
- Once we agree, can grades still be edited, and in what order are they published?
- Can we see a record of who changed what, and why?
If you want a second view of your scorecard or shortlist, book a 15-minute review with Ash. You walk through your cycle; you leave with specific fixes, whichever vendor you pick.
Calibrating in spreadsheets, and where it breaks
Yes, you can calibrate in spreadsheets. Plenty of mid-market teams start that way, and a well-run committee with a complete pack can do the three checks without a product. The work breaks in predictable places: the pack is out of date before the meeting starts, flags are built by hand and missed, a shared file is a poor model for who may change a grade, and there is no controlled publish — someone is still copying final grades into another sheet. That is the point at which software should carry the drafts, the flags, the permissions and the publish step, while the committee still makes the decisions. The practical move is in switching from spreadsheets to performance management software.
The Ebury customer story, already linked above, is one published example of leaving disconnected grading spreadsheets: it states that departmental calibration time fell from 4 hours to 1 hour, a 75% reduction. Do not treat that as a typical result.
Common mistakes
- Forced curves. Applying a distribution without reading the evidence turns calibration into arithmetic. The playbook publishes its own guideline weights, not an industry benchmark, and research on capped curves is mixed — see stack ranking vs calibration and the framework evidence and caveats. If a tool recommends grades by mapping people onto a predefined distribution, use it as a starting point for the three checks, not as a target the room must hit.
- Calibration theatre. The meeting happens, everyone nods, and the grades that walked in are the grades that walk out. If grades rarely change, one team’s top-heavy ratings go unchallenged, or the pack is never opened, you are running theatre. The fix is in calibration theatre (and how to fix it).
- Skipping evidence. Arguing about the person, or about last quarter’s reputation, instead of the scorecard in front of you. A flag is only useful if someone opens the evidence. The performance review bias guide and the free bias checker help you spot the patterns before the meeting.
Related guides
Stack ranking vs calibration: why forced ranking fails and what to use · Year-end performance review: HR run sheet and template · Performance management framework: scorecards, talent bar and calibration · how to run a performance calibration meeting · calibration meeting template for mid-market tech companies · employee performance rating scales explained · how to calculate regretted attrition · job architecture and levelling framework · upward feedback questions · performance review examples by rating level · staff appraisal template (UK) · 9-box grid: how it works, a free template, and when to replace it · calibration theatre · switching from spreadsheets to performance management software · performance management software buyer’s guide · free tools · succession planning for People teams · People Performance Playbook home
FAQ
What is performance calibration?
Performance calibration is the step in a performance review cycle where managers and HR compare draft ratings across teams, against the same standards and evidence, before grades are final. It catches generous or harsh graders, ratings with thin evidence, and outliers, so people doing similar work at the same level get similar grades.
How does a calibration process work?
After managers submit scorecards, the performance team reviews the draft grades against shared standards, then senior leadership reviews and approves them. The playbook’s three checks are generous graders, thin evidence, and outliers or sudden jumps. In the playbook’s quarterly cycle, grade calculation and calibration take about one week at most, including senior leadership review. Results are then discussed in each person’s 1:1.
What should calibration software do?
It should show draft ratings side by side by team, role and level; flag generous graders, thin evidence and outliers automatically; let the right people adjust grades with a reason; and lock and publish grades once agreed, with a record of who changed what. A tool flags; people decide. No product removes bias by itself.
Can you calibrate in spreadsheets?
Yes. Many teams start that way. Spreadsheets break when the pack goes stale, flags are built by hand, access is a shared file, and there is no controlled way to publish grades. That is usually the point to move the cycle into software. See the Playbook guide on switching from spreadsheets to performance management software.
Does calibration software remove bias?
No. Software can flag generous or harsh graders, thin evidence and outliers so the room spends time on the exceptions. People still decide whether a grade should change. Treat any vendor claim that a tool removes bias as marketing, not a substitute for the three checks.
Should we force a distribution of grades?
The playbook gives its own guideline weights, not an industry benchmark, and research on capped curves is mixed. Keep calibration evidence-based. If a tool recommends grades by mapping people onto a predefined distribution, treat that as a starting point for the three checks, not as a curve you must hit.