How to run a compensation review at a small company

Small companies usually do not have a compensation process. They have a series of exceptions: a counter-offer in March, a raise negotiated by someone who threatened to leave in June, a promotion that came with a number nobody benchmarked. Each decision was reasonable on its own. Together they produce a pay structure nobody designed and nobody can now explain, which is exactly the thing you get asked to explain.

A compensation review is how you convert that into something defensible. It does not require market survey subscriptions, a compensation function, or a heavy tool. It requires a cycle, a budget, a structure, and a rule about who decides.

Start with the structure, not the raises

You cannot review pay without something to review it against. That something is a set of bands.

A band is a minimum, a midpoint and a maximum for a role at a level. It is not a promise, it is a container: it says what this company pays for this kind of work at this level of experience. Two people doing the same job at the same level should be inside the same band, and if they are not, you have found the first thing the review is for.

Building your first bands does not require external data. Take your existing salaries, group them by role and level, and look at the spread. You will find roles where the spread is wide and nobody can say why, and levels that have drifted because each new hire was priced against the last one. The salary band calculator turns a minimum and a maximum into quartiles and places a salary with a compa-ratio, in your browser, with no market data invented and nothing sent anywhere.

Compa-ratio is the tool that makes the review tractable. A person’s salary divided by the midpoint of their band gives you one comparable number for everyone in the company. Sorted, that column tells you more in ten minutes than a week of individual conversations: who is drifting low, whether a team is systematically underpaid, and whether the same job is being paid two different ways.

The cycle

Run one annual review covering everyone, on a fixed date, plus a defined off-cycle route.

A fixed date matters more than which date. When people know the cycle, they stop asking, and managers stop making informal promises to bridge the gap. Pick a month that is not your busiest and not immediately after your financial year end, when the budget is not yet settled.

Everyone in the cycle matters because the alternative is reviewing only the people who ask. That pays negotiation rather than contribution, and it correlates with confidence rather than performance in ways that show up as a pay gap you did not intend and will struggle to justify.

A defined off-cycle route matters because promotions and genuine market events do not wait for your calendar. Write down what qualifies, who approves it, and what evidence is needed. An undefined off-cycle route becomes every route.

The budget comes before the decisions

Decide the total before anyone proposes anything. Managers asked to recommend increases without a constraint will recommend increases that add up to more than you have, and the correction that follows is a series of conversations in which people are told their manager’s recommendation was overruled. That is worse for trust than a constraint stated up front.

Express the budget as a total pot rather than as a percentage everyone gets. A flat percentage is not a review, it is an indexation, and it preserves every inequity already in the structure while costing the same money.

Then decide the split. Most of the pot goes to correcting people who are low in their band relative to their contribution and to sustained performance. Some goes to promotions, which are a band change rather than a raise. And a portion should be held back for the corrections you have not found yet, because you will find one.

Keep pay decisions separate from rating conversations

Performance input feeds the pay decision. The two conversations should not be the same conversation.

When feedback and a number arrive together, the person hears the number and remembers nothing else, so the development conversation is wasted. It also forces the manager to defend a budget constraint as though it were a judgement about the person, which is both untrue and unwinnable.

Run performance first, on its own, with a few weeks before pay decisions are communicated. A light performance review process sets out a cadence that works for a small team, and the separation of ratings from pay is one of its explicit design choices. If you use the Performance and OKR module, the review cycle lives there and the compensation cycle is its own thing.

The review itself, in five steps

1. Assemble the data. For every person: current salary, band and level, compa-ratio, time in role, time since last change, and the performance input. This should be a report, not an assembly job. If it takes a week to gather, that is a finding about your record-keeping rather than about compensation.

2. Look at the distribution before any individual. Sort by compa-ratio. Look at it by team, by level, and by how long people have been in role. Patterns first, people second. This is the step that catches the systemic problems, and it is the step most often skipped.

3. Have managers propose within the constraint. Give them the band, the compa-ratio, the budget and the rules. Ask for a proposal and a sentence of reasoning per person. The sentence is the important part, because it is what you will read again next year.

4. Calibrate across managers. Put the proposals side by side. You are looking for one manager being systematically more generous than another, for proposals that are really promotions in disguise, and for the person nobody proposed anything for who has been flat for three years.

5. Decide, and write down why. Every change and every non-change gets a reason recorded. The non-changes matter as much, because “why did nothing happen for me” is the question you will actually be asked.

Communicating it

Tell people the outcome individually, from their manager, with the reason.

Say the new number, when it takes effect, and what drove it. If someone is getting nothing, say that clearly and say what would change it, rather than implying that something is coming. Vagueness here is not kindness; it produces an expectation you did not agree to.

Be consistent about what you disclose. Whether you publish bands is a real decision with arguments on both sides, but the version that fails is publishing them to some people and not others. Whatever you choose, apply it to everyone.

Then put the change into the record with its effective date, and generate the letter from the record rather than retyping it. In the Capstan core, compensation changes are effective-dated, so a revision agreed in April takes effect in April and is still readable as history in 2029, and letters are generated from record fields and filed against the person automatically. The Compensation and Equity module is where a structured review cycle lives if you want the process in the system rather than in a spreadsheet.

Equity is a separate conversation

Refresh grants, promotion grants and new-hire grants are compensation, but they follow different rules and a different clock. Do not fold them into the cash review as an afterthought, and do not use equity to cover a cash problem, because the two are not substitutes for someone paying rent.

If equity is part of your package, the thing that matters most is that people understand what they have. Startup equity and ESOPs covers the mechanics, ESOP vesting explained covers cliffs and schedules, and the ESOP vesting calculator will show someone their own schedule without asking them to trust a spreadsheet.

The records that make next year easier

Three things, kept properly, halve the work next cycle.

Effective-dated compensation history. Not the current salary, the sequence of salaries with the dates they took effect. This is what lets you answer what someone earned in a given month, which an auditor, an acquirer or a tribunal will eventually ask.

The reason for each decision. One sentence per person per cycle. It costs nothing at the time and it is the difference between a defensible pattern and a pile of unexplained numbers.

The band definitions as they stood. Bands move. If you cannot reconstruct what the band was in 2026, you cannot explain a 2026 decision in 2028.

All three are the same underlying property: a record that can be read as it stood at a date rather than only as it stands today. That is the argument in an HR system is a record of the past, and compensation is where it pays off first.

If you have never done this

Do a small version this year. Build rough bands from your own salaries, compute compa-ratios, look at the distribution, fix the two or three cases that are obviously wrong, and write down the reasoning.

That is a week of work, and it produces the thing you actually need: a structure to review against next year, instead of another twelve months of exceptions.