Performance reviews for an IT team: light process, real outcomes

Why heavyweight reviews fail in a 20-person company

Most review processes at small companies are borrowed from large ones — forty-cell competency matrices, nine-box grids, calibration committees — and they fail the same three ways:

• Theatre. That machinery coordinates judgment across hundreds of managers. At 20 people, with two or three, the forms coordinate nobody. Everyone senses it's ceremony; the writing turns generic and the meeting becomes reading the form aloud.
• Recency bias. An annual review is a review of October and November. The migration someone carried in February is gone — not from malice, from memory — and the review ends up measuring the loudest recent project.
• No follow-through. Development goals get written in the review meeting, into a document opened a year later at the next review — where everyone is briefly embarrassed, then writes new ones.

The reflex conclusion — "we're too small for reviews" — is wrong too, you just pay differently: promotions become vibes-based, raises go to whoever asks loudest, quiet solid engineers stall invisibly, and when someone underperforms there's no written trail of expectations. The answer isn't no process; it's a process sized to reality.

A light cycle that actually works

Twice a year, four questions, goals set in advance. The whole design:

1. Set goals at the start of the cycle, not the end. Three or fewer, concrete enough to score honestly — SMART (specific, measurable, achievable, relevant, time-bound) is the checklist; the spirit is simpler: written so that in six months neither of you can argue about whether it happened. Goals reconstructed at review time aren't goals; they're a narrative.
2. Self-assessment first. The person writes their own view before seeing the manager's: what went well, what didn't, how the goals scored, where they want to go next. It surfaces work the manager forgot and shows where self-image and reality diverge — the most useful conversations live in that gap.
3. Manager assessment, same four questions. Written before reading the self-assessment, or it becomes an echo.
4. One conversation, about an hour. Compare the two documents, spend the time on the gaps and the future, and end with next cycle's goals written down.

The prerequisite that keeps all of this cheap: no surprises. If 1-on-1s happen all year, the review is a summary of conversations already had — twenty minutes of writing, not an anxiety event. If a review ever genuinely shocks someone, the process that failed was the fortnightly one, not the biannual one.

Connecting reviews to grades and pay — honestly

The tempting shortcut is a formula: review score in, raise percentage out. It fails at both ends — reviews turn into salary negotiations conducted in code (every score inflates), and pay decisions pretend to an objectivity they don't have. The honest architecture has three parts:

• A public grade ladder. Written criteria per level: what a middle vs a senior engineer is expected to own, decide and deliver. Promotion cases then argue from criteria — "has been doing the senior list for two quarters, here's the review trail" — instead of vibes ("feels senior"). Criteria beat vibes not because they're perfect but because they're arguable: a person can see the gap, target it, and measure progress against something that doesn't move.
• Reviews as the evidence, not the verdict. The review documents what happened; the grade defines what's expected; comparing the two is the promotion conversation. Raises follow the same shape: a band per grade, position in band informed by the review history.
• Separated conversations. The review is about the work. The pay decision follows on its own schedule — informed by the review, not held in the same meeting, because nobody hears feedback over the sound of waiting for a number.

In the EU this stops being merely good practice: pay-transparency rules require pay and progression to rest on objective, gender-neutral criteria that employees can ask about. A grade ladder plus a documented review history is that criteria trail — growing one now beats retrofitting it under a deadline.

360° feedback: when it helps, when it hurts

360° — structured feedback from peers, reports and adjacent managers — is a scalpel, not a default.

Where it earns its cost:

• Leads and managers. Their manager sees a fraction of their impact; the team experiences the rest daily. For anyone whose output is other people's working environment, single-source assessment is close to blind.
• Promotion cases with real stakes, where one manager's view deserves corroboration before a grade change the whole company will see.
• Cross-team roles — a QA lead or platform engineer whose actual customers sit outside their reporting line.

Where it hurts:

• Tiny reviewer pools. Two peers reviewing a person is anonymity theatre — everyone knows who wrote what, so everyone writes accordingly. Below a handful of honest reviewers, ask openly instead, or don't ask.
• As ammunition. The first time 360° input gets quoted in a dispute or a performance case, the whole company recalibrates toward blandness — permanently.
• As routine. Running 360° for everyone every cycle taxes a 20-person team's calendar heavily for feedback that mostly repeats what the manager already knew.

Keep it request-based — a specific person, a specific cycle, a handful of focused questions — and treat it as input to the manager's judgment, never a score that decides by itself.

How Helia HR does this

The Reviews & Goals pack ($1.50/employee/mo — see pricing) implements the light cycle end to end:

• Review cycles with periods — quarterly or biannual, with a clear start and end, so the ritual has a calendar instead of an initiator.
• Self-assessment + manager assessment on shared templates — the same questions, asked the same way, every cycle.
• SMART goals set at cycle start and scored at the end, so scoring is a comparison against something written in advance, not a memory exercise.
• 360° feedback requests for the cases above — requested per person per cycle, not blanket.
• An AI review draft the manager edits — it assembles the period's goals and material into a first draft; the manager rewrites, confirms, and owns every word — assistive, human-owned, a time saver rather than a judgment delegator.
• Full history per employee — every cycle, goal and outcome in one trail — exactly what promotion cases and pay conversations need.
• Links to career ladders and grades (Career paths pack, $1/employee/mo), so "ready for senior?" is answered against written criteria with the review history attached.

FAQ

Quarterly or biannual?

Biannual for the full written cycle; quarterly only if goals genuinely change that fast. Quarterly full reviews at a services company usually collapse into copy-paste by the third round — the cadence for continuous feedback is the 1-on-1, not more paperwork.

Should the review score set the raise?

Mechanically — no. Reviews inform pay through the grade ladder: the review evidences the grade, the grade carries the band. Couple score to money directly and every review becomes a negotiation and every score an argument — the feedback content drops to zero.

Do we need calibration at 20 people?

A committee, no. A 30-minute conversation between the two or three people who wrote assessments — "are we applying the same bar?" — yes, especially before promotion decisions. Calibration's purpose is fairness across managers; buy it at the smallest size that delivers it.

Is it OK that AI drafts the review?

Draft — yes; verdict — no. Assembling six months of goals, notes and outcomes into a first draft is exactly what assistance is for. Deciding what it means is the manager's job, and the person being reviewed should never receive a sentence the manager wouldn't defend line by line.