Most performance management projects begin with the form and end with the software. Neither is the system. The system is four decisions: what the rating is for, who owns it, how disagreement gets resolved, and what actually happens to someone after it is assigned. Organisations that answer those four can run a decent process on a spreadsheet. Organisations that skip them will not be rescued by a platform.
TL;DR: Design the decision, not the document. Fix what the rating drives before you fix the scale it uses, because a rating that feeds pay, promotion and exit simultaneously will be gamed on all three. In a GCC, settle the dual-reporting question explicitly: when a local site leader and a global functional lead both assess the same person, someone has to own the final rating and it should be written down before the cycle opens. And in India there is a legal dimension that almost no performance management content addresses: under the Industrial Relations Code, 2020, a dispute arising from the termination of an individual worker is deemed to be an industrial dispute in its own right, which makes your performance documentation the evidence base rather than an HR formality.
Why Most Systems Fail at Scale
Small organisations do not need a performance system, because the person deciding your rating has watched you work. That breaks somewhere between fifty and two hundred people, and it breaks in three specific ways.
The rating stops being evidence and starts being currency. Once a rating drives pay, managers rate to protect their people rather than to describe them. This is rational behaviour and no amount of training fixes it. It is a design problem.
Calibration becomes negotiation. With enough managers in a room and no agreed standard, the loudest advocate wins. What emerges is a distribution that reflects managerial confidence rather than employee performance.
The system starts producing paperwork nobody uses. Ratings are assigned, filed, and have no consequence that anybody can point to. Employees notice this faster than leadership does, and the credibility loss is very hard to recover.
The common root is that the organisation designed a form, a cycle and a rating scale, and never decided what the output was for. Our piece on the blind spots that appear only after scaling makes the same argument at leadership level: what breaks at scale is not effort, it is institutional design.
The Four Decisions That Are the System
1. Decide what the rating is for, and be willing to give something up.
A single rating cannot optimally serve pay, promotion, development and exit decisions. Pay wants differentiation. Development wants honesty. Exit wants documented consistency. Promotion wants forward-looking potential, which is a different construct from past performance entirely.
Most organisations attach all four to one number and then wonder why managers inflate. Pick a primary purpose. If the rating primarily drives pay, accept that it will be a weaker development tool and run development conversations separately, on a different cadence, with no number attached. Separating the two is the single highest-return change I see organisations make.
2. Decide who owns the rating, in writing, before the cycle opens.
Not who inputs it. Who owns it when two people disagree. This sounds procedural and is the most consequential thing on the list, particularly in a matrixed structure.
3. Decide how calibration actually works.
Calibration means comparing people at the same level across managers to ensure a rating means the same thing everywhere. It requires a defined population, an agreed standard described in behaviours rather than adjectives, a facilitator with authority to challenge, and a record of what changed and why.
A word on forced distribution. Imposing a fixed curve on a small population produces absurd outcomes, because a team of eight does not contain a statistically representative spread of performance. If you use guidance ranges, treat them as a prompt to justify a departure, not a quota to fill. And if you use a curve, understand that in India it will eventually be examined in a forum where you have to explain the reasoning.
4. Decide what happens next, for each rating.
Every rating band needs a defined consequence: what it means for pay, what it triggers in development, and what the process is at the bottom of the scale. If the bottom band has no defined path, managers will avoid using it, the distribution will compress upward, and within two cycles the system will be describing a workforce in which everybody is above average.
The GCC Problem: Dual Reporting and Who Owns the Rating
This is where generic performance advice stops being useful, because the structure is genuinely different.
In most capability centres an engineer has two bosses. A local site or delivery leader who sees them daily, manages their working environment, and knows whether they are struggling. And a global functional lead, often in another time zone, who owns the work, sets the priorities and is best placed to judge technical output.
Both have a legitimate view. Neither has the whole picture. The functional lead sees the deliverable and not the conditions; the site leader sees the effort and not always the standard the work is judged against.
Three failure patterns follow, and I have seen all three repeatedly.
The global lead rates without local context, and someone gets marked down for a delay caused by a dependency they did not control. The site leader rates protectively, and the functional lead quietly stops trusting the ratings and forms private views instead. Or nobody owns it, both fill in a form, and the outcome is whichever was submitted last.
The fix is not complicated but it has to be explicit. Name one owner of the final rating, usually the person closest to the work. Give the other a defined and recorded input, not a veto. Run one calibration that includes both, at a time that is actually workable across the time zones involved rather than one that quietly excludes India from the conversation. And write down where the tie breaks before the cycle starts, because deciding it during the cycle means deciding it in favour of whoever is more senior.
That last point matters for a reason beyond fairness. A centre whose people believe their rating is set by someone who has never seen them work will not retain its best engineers, whatever the pay band says. This is the same governance discipline we set out in the GCC governance blueprint, applied to people decisions rather than board ones.
In India, Your Performance System Is Also Your Termination Evidence
Here is the dimension that international performance management content almost never covers, and it changes how the system should be designed.
Start with who counts as a “worker.” Under the Industrial Relations Code, 2020, the definition excludes a person employed mainly in a managerial or administrative capacity, and a person employed in a supervisory capacity drawing wages exceeding eighteen thousand rupees per month, or such amount as the Central Government notifies.
Read that carefully, because the common assumption is wrong. The exclusion for high earners applies to those in a supervisory capacity. A senior individual contributor who earns a great deal and supervises nobody does not obviously fall within either exclusion. In a capability centre staffed largely by senior engineers with no reports, a significant part of the population may well be within the definition of worker.
That matters because of what follows. The Code provides that where an employer discharges, dismisses, retrenches or otherwise terminates the services of an individual worker, any dispute arising from that termination is deemed to be an industrial dispute, notwithstanding that no other worker and no trade union is a party to it. The worker can apply directly to the Tribunal after forty-five days from applying to the conciliation officer.
So an individual exit for poor performance is not a private HR matter that concludes when the person leaves. It is a matter that can be adjudicated, and the question in that forum will be whether the process was fair and documented.
Two further points sit alongside this. Standing orders, which must cover classification of workers, termination of employment and grievance redressal, are required for industrial establishments employing three hundred or more workers, up from one hundred previously. And the four labour codes came into force on 21 November 2025, so this is the current framework rather than a forthcoming one.
The design consequences are practical rather than legalistic.
Write ratings that describe behaviour and output, not personality. “Did not meet the agreed delivery standard on three of five committed items” survives scrutiny. “Attitude issues” does not.
Make the improvement process real and dated. If someone is placed on a performance plan, it needs specific objectives, a defined period, recorded review meetings and a documented outcome. A plan created two weeks before a decision that had already been taken is worse than no plan at all, because it evidences the predetermination.
Keep the record contemporaneous. Notes written after the decision are visible as such.
And ensure sexual harassment and misconduct are treated under their own frameworks rather than folded into a performance rating, which is a separate governance requirement.
None of this is a reason to avoid managing performance. It is a reason to manage it properly, and the organisations that do have less exposure, not more.
What to Measure, and What to Stop
Three things are worth measuring about the system itself, as opposed to about the people in it.
Distribution by manager, over time. A manager who has rated everyone highly for three consecutive cycles is telling you something about the manager.
Rating movement after calibration. If nothing changes in calibration, it is not calibration. If everything changes, the managers are not equipped.
Correlation between rating and regrettable exit. If your top-rated people are leaving at the same rate as everyone else, the rating is not identifying what you think it is.
Two things are worth doing less of. Cascading objectives through five organisational layers, which produces goals nobody can influence by the time they arrive. And running a mid-year review that is a lighter copy of the annual one; either it is a real checkpoint that can change the year’s outcome, or it is theatre.
A Design Checklist
-
State the primary purpose of the rating in one sentence, and separate development conversations from it.
-
Name the owner of the final rating for every role, especially dual-reporting ones.
-
Define calibration: population, standard, facilitator, and a record of changes.
-
Define the consequence of every rating band, including the bottom one.
-
Write behavioural standards per level, so that “exceeds” means the same thing across sites.
-
Train managers on writing evidence, not on using the software.
-
Diarise the cycle across time zones so no location is structurally excluded from calibration.
-
Make the improvement process a real process with objectives, dates and recorded reviews.
-
Check your worker classification against the IR Code definition before you rely on assumptions about who is exempt.
-
Review the system annually against its own metrics, not against whether the forms were submitted on time.
Frequently asked questions
- What makes a performance management system work at scale?
- Four design decisions, none of which are about software. First, what the rating is primarily for, since a single rating cannot optimally serve pay, promotion, development and exit at once and will be inflated if it tries. Second, who owns the final rating when two assessors disagree, which must be settled in writing before the cycle opens. Third, how calibration works in practice, meaning a defined population, a standard described in behaviours rather than adjectives, a facilitator with authority to challenge, and a record of what changed. Fourth, what actually happens after each rating band is assigned. Organisations that answer these can run a credible process on simple tools; organisations that skip them are not rescued by a platform.
- Who should own the performance rating in a GCC with dual reporting?
- Usually the person closest to the work, which in most capability centres is the global functional lead who sets priorities and judges technical output, with the local site leader holding a defined and recorded input rather than a veto. What matters more than which way you decide is that it is decided and written down before the cycle opens, because settling it mid-cycle means settling it in favour of whoever is more senior. Both parties should attend a single calibration scheduled at a time that genuinely works across the time zones involved, since a calibration held in a European or US morning slot systematically excludes the India view while appearing inclusive.
- Is a bell curve or forced ranking legal in India?
- There is no statutory prohibition on using a distribution guideline, but the risk is practical rather than definitional. Applying a fixed curve to a small team produces outcomes that cannot be justified on the evidence, because a group of eight does not contain a representative spread of performance. That becomes a problem where a rating contributes to an exit decision, since under the Industrial Relations Code, 2020 a dispute arising from the termination of an individual worker is deemed an industrial dispute and can be adjudicated. In that forum the question is whether the assessment was fair, consistent and documented. A rating driven by a quota rather than by evidence is harder to defend than one that departed from the guideline with reasons recorded.
- How should a performance improvement plan be documented in India?
- Treat it as a genuine process rather than a procedural step before a decision already taken. That means specific and measurable objectives, a clearly defined period, scheduled review meetings that actually happen and are minuted, support or training that was offered and recorded, and a documented outcome at the end. Feedback should describe behaviour and output rather than personality, since a statement about missed delivery commitments is evidence while a statement about attitude is opinion. Records should be contemporaneous, because notes written after the decision are recognisable as such. A plan created shortly before an exit that was already decided is worse than no plan, since it evidences predetermination.
- Are highly paid employees exempt from Indian labour protections on termination?
- Not automatically, and this is the most common misreading I encounter. Under the Industrial Relations Code, 2020, the definition of worker excludes those employed mainly in a managerial or administrative capacity, and those employed in a supervisory capacity drawing wages exceeding eighteen thousand rupees per month or such amount as notified. The high-wage carve-out attaches to supervisory roles. A senior individual contributor who earns substantially and supervises nobody does not clearly fall within either exclusion. In a capability centre staffed largely by senior engineers without direct reports, a material share of the population may fall within the definition, so classification should be checked against the Code rather than assumed from salary level.
A Closing Note
The best performance system I have worked with was not sophisticated. It had four rating bands, a one-page evidence form, a calibration meeting the leadership team genuinely attended, and a rule that no rating could be assigned that the manager could not defend in that room with examples.
The worst had a nine-box grid, a competency framework running to sixty behaviours, an expensive platform, and no meaningful difference between what happened to the top-rated and bottom-rated employees afterwards.
People strategy is business strategy in execution, and performance management is where that claim gets tested most directly. It is the mechanism through which strategy reaches individual behaviour. If the mechanism produces a number nobody trusts and nothing that follows from it, the strategy is not reaching anyone, however good the deck was.
Design the decision. The form is the easy part.
If you are rebuilding a performance framework, or dealing with a dual-reporting structure that is producing ratings nobody trusts, Astravise Services works on organisation design, performance frameworks and HR governance for organisations scaling across sites, including through our fractional CHRO advisory. Start by asking what the rating is for. Most of the disagreement in the room resolves once that is answered.