Performance Management

Performance management is not the annual form. The form is the artefact people remember, and it is the least important part of the system. What matters is the cycle it sits in: expectations set at the start, feedback given while the work is happening, a review that summarises rather than surprises, a calibration step that makes one manager’s “good” mean the same as another’s, and a consequence that actually follows.

Most companies have the form and one or two of the other five. That is why performance reviews have the reputation they do.

The cycle#

Expectations. What good looks like for this person in this role, agreed in advance. Without this the review is one person’s impression against another person’s assumption.

Ongoing feedback. The reason the no-surprises rule exists: nothing in a review should be new information. A manager who saves feedback for the form has not managed, they have collected evidence.

The review. A summary of a period, written against the expectations set at the start.

Calibration. Managers compare their ratings side by side before anything is final. This is the step that makes ratings mean anything at all, because “exceeds expectations” from a generous manager and a harsh one are otherwise different currencies. It is also the step most often skipped, since it is the only one that requires managers to defend their judgements to peers.

Consequence. Pay, promotion, development, or exit. A review with no consequence trains everyone to treat the whole exercise as paperwork, and they are right to.

The frameworks you will meet#

Rating scales. Three-point and five-point are both common. Five-point scales reliably collapse into three in practice, because almost nobody uses the bottom rung without an existing plan to act on it, and the top rung is rationed. If you build for five, expect the data to arrive as three.

Forced distribution, also called stack ranking: a mandated curve, so a fixed percentage must land in each band. Made famous at GE, adopted widely, and largely abandoned since, including by GE and Microsoft. It reliably produces two behaviours: managers hiring weak people to protect strong ones, and teams avoiding collaboration with people they are ranked against.

The 9-box grid. Performance on one axis, potential on the other, nine cells for succession planning. Useful for a conversation about who is ready for what. Dangerous when it leaks, because “potential” is the most subjective judgement in the whole system.

MBO and OKRs. Goal frameworks that predate and feed the review. The warning worth carrying, and Doerr is explicit about it: do not wire OKRs directly to ratings or pay. Goals set to be ambitious and goals set to be achieved are different instruments, and tying them to compensation converts every ambitious goal into a sandbagged one.

Continuous versus annual. The industry has moved toward lighter, more frequent check-ins with a thinner annual summary. That shift is real and healthy, and it does not remove the need for a periodic written record, for reasons the next section makes concrete.

The part that is specific to here#

In Indonesia a performance record is not only an HR instrument. It is the evidence chain behind a lawful termination, and behind the money that termination costs.

Dismissing someone for underperformance requires a documented process, normally the surat peringatan sequence: SP1, SP2, SP3, each issued for a stated reason and each typically valid for six months unless the company regulation says otherwise. The grounds and the process have to be set out in the peraturan perusahaan or the collective agreement, and the sequence has to actually have been followed.

Two consequences follow, and the second is the one people miss.

First, an undocumented performance problem is not a performance problem the law recognises. A manager who has been unhappy for a year but issued nothing has no route to act.

Second, the reason for termination determines the severance calculation. Under PP 35/2021 the entitlements, uang pesangon, uang penghargaan masa kerja and uang penggantian hak, carry different multipliers depending on why the employment ended. So the performance record does not stop at HR: it selects which formula the final payroll run uses.

Regulations move, and every citation here is our current understanding, to be re-verified against JDIH Kemnaker before it drives code.

One record, two destinations. The left-hand loop is talent work and it stops inside HR. The right-hand loop is a legal evidence chain, and its last step is a number our engine has to compute. Only one of these is ours.

Why this matters to our strategy#

The employee lifecycle page says plainly that we should not build the middle of the lifecycle, because develop and retain are real stages we are not the right product for, and a shallow module in each would cost us the credibility payroll earned. That still holds, and performance management is the case that tests it properly rather than the exception that breaks it.

The test resolves by splitting the subject, which is what the figure above is doing. Ratings, calibration and 9-box grids are talent tooling. That market is crowded, we have no advantage in it, and it is exactly the shallow module the lifecycle page warns about.

The warning-letter chain is not talent tooling. It is a dated, evidenced document trail whose final step selects a severance formula and lands in a payroll run. That is payroll-adjacent, it is legally consequential, and getting it wrong produces either an unlawful termination or a wrong final settlement, both of which are our kind of failure.

So the honest answer to “should Interwise do performance reviews” is no, and “should Interwise hold the performance record that a termination and its severance depend on” is yes. Those sound like the same question and are not.

Each letter carries a validity window, commonly six months unless the company regulation sets otherwise. A letter that expires before the next one is issued does not carry forward, so a chain assembled carelessly is worth nothing at the moment it is needed.

What this means for what we build#

Hold the record, not the rating. Store warning letters as dated, reasoned, acknowledged documents with a validity period the system understands. That is the artefact with legal weight.
Make expiry visible. A letter is only useful while it is live. If SP1 lapses before SP2 is issued, the chain has restarted, and the product should say so rather than leaving somebody to discover it during a dispute.
Wire the record to the final run. The reason for termination selects the severance formula, so the document trail and the calculation must be the same system. Two systems here means the number and its justification can disagree, which is the exact failure we exist to remove.
Do not build ratings, calibration or a 9-box. If a customer needs those, they are better served elsewhere, and saying so protects the credibility the payroll work earned.
Do carry the no-surprises rule into the interface. Anything that lets a manager issue a formal warning without a record of prior conversation is a product helping somebody build a weak case.

The Employee Lifecycle sets the position this page tests: do not build the middle.
Gross to Net is where a termination reason becomes a settlement figure.
Compliance as a Calendar covers the regulation stack these rules sit in.
Domain Standards & Learning Map lists what to learn before building here.