know.2nth.ai biz HR Performance
biz · HR · Skill Leaf

Goals, KRAs, and the record that a commissioner reads.

Performance management has two halves that most teams treat as one. There's the HRMS mechanics — goals, KRAs, appraisal cycles, the records in Frappe HR's Appraisal — and there's the human craft of feedback and coaching that lives in the People domain. This leaf joins them, and adds the part neither covers: in South Africa, the performance record is legally load-bearing. Under the LRA, poor performance is incapacity — and that's a fair reason for dismissal only if you can prove a fair reason and a fair process. The record is the proof.

Goals · KRAs · Appraisal LRA · Schedule 8 Feedback craft Frappe HR Not legal advice

Two halves of one discipline.

Performance management is where what someone is asked to do (their goals and KRAs — key result areas) gets set, tracked, and judged across an appraisal cycle. Split it cleanly and two distinct things appear.

The first is mechanics: the structured artefacts — goal records, KRA weightings, appraisal scores, sign-off dates. That's what an HRMS is for; in this tree it's Frappe HR's Appraisal DocType holding the cycle. The second is craft: the conversation. Good feedback, useful 1:1s, the coaching that actually changes behaviour — none of which a form captures. That half belongs to the People domain, where development, typologies and leadership live.

This leaf joins the two — and adds the third thing they each assume away: the legal weight. In South Africa a performance record isn't just an HR artefact. It's the evidence that decides an unfair-dismissal case. Get the mechanics and the craft right but the record wrong, and a fair decision can still be found unfair.

Goals vs KRAs vs the rating

A goal is a target for a period ("ship the migration by Q3"). A KRA is the standing area of accountability a role is measured on ("system reliability"), usually weighted. The appraisal rating is the periodic judgement against both. Frappe HR's Appraisal models all three; the judgement itself — the rating — is a human act, not a calculation.

Goal-set → check in → review → calibrate.

A cycle runs in four stages. Each produces a structured output — and that output is also the record. Skip a stage and you don't just lose rigour; you lose the evidence trail the law will later ask for.

Stage 01 · set

Goal-setting

Agree goals and KRAs for the period, with weightings and a clear standard of "good". Output: a signed-off goal/KRA record both parties can point back to.

Stage 02 · run

Check-ins & 1:1s

Regular conversations against the goals — coaching, course-correction, surfacing blockers. Output: contemporaneous 1:1 notes. This is where the craft lives, and where evidence accrues.

Stage 03 · judge

Review / appraisal

The periodic rating against goals and KRAs, with narrative. Output: the appraisal record — score, evidence, and the manager's reasoning, dated.

Stage 04 · fairness

Calibration

Ratings compared across a cohort to check consistency and remove bias before they're final. Output: a calibrated, defensible set of ratings — and a reason the spread looks as it does.

The order matters legally as much as managerially. A poor rating in stage 03 that wasn't foreshadowed by honest check-ins in stage 02 is a rating that "came from nowhere" — exactly the pattern a commissioner treats as procedurally unfair.

The record is the evidence.

When a poor-performance dismissal is challenged, it goes to the CCMA — and the performance record you built (or didn't) is the case. A commissioner doesn't take the employer's word for it; they read the trail.

An unfair-dismissal dispute is referred to the CCMA (Commission for Conciliation, Mediation and Arbitration) and must be referred within 30 days of the date of dismissal. The process runs conciliation first, then arbitration. At arbitration the commissioner determines both substantive and procedural fairness — was there a genuinely fair reason, and was the process fair? — exactly the two limbs the LRA sets out in s188.

The performance record is the evidence a commissioner reviews. The goal-setting that defined the standard; the check-in notes that show the employee was told, honestly and in time, that they were falling short; the counselling and training offered; the improvement period given. A complete, contemporaneous record makes the case. A thin or back-filled record loses it — a counselling note dated the week of the dismissal, or a sudden first-ever poor rating, reads as exactly what it is.

The discipline this demands is simple to state and easy to neglect: document contemporaneously. Write the 1:1 note when the 1:1 happens. Record the improvement plan when it's agreed, not when it's needed in evidence. Under POPIA (Act 4 of 2013) that record is personal information — keep it accurate, secured, and held on a lawful basis. The record's value as evidence and its lawfulness as data are the same discipline seen from two angles.

The cost of a back-filled record

A performance record assembled after the decision to dismiss has already been made is worse than no record — it signals the process ran backwards. Commissioners are practised at spotting it. If the trail isn't there before you act, the fair move is to build the trail (evaluate, guide, give time to improve) first — not to manufacture one.

Where the mechanics and the craft live.

The two halves map to two parts of this tree, and an agent can help with the mechanics — within a hard boundary.

The mechanics live in Frappe HR's Appraisal record: the goals, the KRA weightings, the appraisal cycle, the scores and sign-off dates. That's the structured artefact — and, in SA, the evidence trail. The craft lives in the People domain: how to give feedback that lands, run a 1:1 that's worth the half-hour, coach toward improvement rather than just rating it. The HRMS records the cycle; the People work is what makes the cycle mean something.

An agent sits usefully on the mechanics. It can draft a review from goal data and notes, summarise a series of 1:1s into a coherent narrative, and surface goal drift — flag where someone's tracking against a KRA is slipping while there's still time to course-correct. All of that is assistance, and all of it keeps a human deciding.

The hard line

An agent must never author a dismissal or disciplinary decision, and never produce the final incapacity finding or the performance rating sign-off. It can prepare the inputs a human reasons over; it cannot be the reasoner of record. The moment the output is a decision with legal consequence — a rating that goes on file, an incapacity conclusion, a step toward dismissal — a named human owns it, with an advisor where the stakes warrant it.

What to automate. What to keep human.

Safe to assist with AI

  • Goal & KRA tracking against the cycle
  • Summarising a series of 1:1s into a narrative
  • Drafting a review for a human to edit & own
  • Surfacing goal / KPI drift while there's time to fix it
  • Prepping the data pack for a calibration session
  • Flagging when a check-in is overdue

Where performance links in the tree.

Primary sources only.