Guide · Pay reviews

Merit matrix or manager judgment?

Matrices fail at the edges and judgment fails at scale. A sound process uses both and reviews the result.

Short answer: Neither alone. A merit matrix gives consistent raises and a predictable budget but breaks for people it fits badly: Top performers high in their range, hires below market, people mid-promotion. Manager judgment fits individuals but compresses raises and can produce gaps nobody intended. Use a guide with ranges, named exceptions and a review of outcomes before letters go out.

Two cycles, two problems

Picture a 220-person fintech company in Chicago. Two years ago it ran its pay review with a merit matrix: A grid that crossed each person's performance rating with where their pay sat in its range and gave a suggested increase for each cell. Finance liked it. The budget landed within a few thousand dollars of plan.

Managers did not. Their best engineer sat near the top of her range, so the matrix gave her 2%. A strong analyst hired a year earlier below market got the same suggestion as a colleague paid far more. A team lead halfway into a promotion fell between the rules. By the end of the cycle a third of all increases were exceptions, each argued by email.

So last year the company switched to manager judgment. Each manager got a budget and a short note on principles. This time the managers were happier and the results were worse. Most teams spread their budget almost evenly, so nearly everyone got 3% to 3.5%. One team gave everyone 5% and went over budget. And when the people team looked at the outcomes afterwards, women rated "exceeds" had received smaller raises on average than men with the same rating.

The CFO wants the matrix back. The CEO wants managers to own pay decisions. Both are describing a real failure.

This company is a composite of situations common at this size. It is not a client and the details are illustrative.

Why the choice is hard

A merit matrix and manager judgment are two answers to one question: Who turns a performance rating into a number? A matrix answers with a rule set in advance. Judgment answers with a person who knows the employee. Each protects something the other gives up.

The first goal is consistency. People with the same rating and the same pay position get similar increases. The budget is predictable. The company can explain any decision by pointing at the rule. The second goal is fit. The person who knows the work adjusts for what a rule cannot see: A flight risk, a role that grew, a hire who came in too low.

Both goals matter more as pay becomes visible. Employees compare raises. In the EU, the criteria used to set pay progression have to be accessible to workers and objective and gender-neutral [6]. US equal pay law accepts a merit system as a reason for pay differences only where it really is a system and really explains the difference [7]. A rule is easy to explain and hard to fit. Judgment is easy to fit and hard to explain.

The budget sharpens all of this. Typical US merit budgets for 2027 sit around 3.2% [5]. With that much money, the gap between a strong raise and an average one is small. Any extra for one person comes from someone else. Our note on what a 3.2% merit budget really means works through the arithmetic.

Timing adds pressure. In many companies ratings are settled and raises are set within the same few weeks, sometimes in the same meeting. When the two decisions sit that close, people start choosing the rating that produces the raise they already want. A matrix makes that temptation stronger, because the link between rating and money is visible. Judgment hides it, which is not the same as removing it. Either way, a rating chosen for its pay effect stops telling the company anything about performance.

Five ways merit decisions break

Matrices tend to fail at the edges. Judgment tends to fail at scale. These are the patterns behind both.

"The matrix makes it fair"

What it looks like: The company adopts a matrix because it removes manager discretion from the pay step. Fairness is treated as solved.

Why smart teams fall into it: There is good evidence that discretion in the pay step can create gaps. A matrix closes that door. Closing it feels like enough.

What it costs: A matrix is only as fair as the ratings it is fed. It turns ratings into pay without bias, but it cannot correct a rating that was itself skewed. If calibration meetings rate some groups lower for reasons that are not about the work, the matrix pays out that difference faithfully and with an air of objectivity. The bias has not gone. It has moved one step upstream, where the pay team is less likely to look.

The pay-position axis has its own blind spot. Most matrices give smaller raises to people high in their range. That is sensible when high position reflects long tenure in a stable job. It is not when it reflects a recent market adjustment or a role that has grown beyond its band. The matrix cannot tell the difference. It treats a person whose range is out of date as a person who is overpaid. A matrix makes the rating the most important pay decision in the company. It deserves the scrutiny that implies.

"Managers know their people best"

What it looks like: Each manager gets a budget and full discretion. The pay team checks only that budgets balance.

Why smart teams fall into it: It respects managers, fits a culture of ownership and avoids the email war over exceptions. Managers really do know things a grid does not.

What it costs: Knowing people is not the same as paying them consistently. A widely cited study of merit pay followed nearly 9,000 employees at one large company. It found that women and minorities received smaller raises than white men with the same performance score, in the same job and unit, under the same manager [1]. The gap arose in the step where ratings became raises, the step judgment controls. A later study of a large company that added accountability and transparency to that step found the gaps narrowed [2]. The lesson is not that managers are biased. It is that unreviewed judgment at scale produces patterns no single manager intended.

Discretion also compresses. Most managers find it easier to give everyone a little than to give some people much more and some nothing. The fintech's even spread is the usual result.

There is also a capacity problem. A manager with eight reports and a two-week window has perhaps an hour to set eight raises. Without a guide, each one starts from nothing. The decisions that get the most thought are the ones someone pushes for, which tend to be the confident and the well connected. That is not bias in the usual sense. It produces the same kind of pattern.

"The edges will sort themselves out"

What it looks like: The matrix covers the middle of the workforce well. Nobody decides in advance what happens to the people it fits badly.

Why smart teams fall into it: The edge cases are a minority. Designing for them feels like overengineering.

What it costs: The edges are where the money and the risk are. A top performer near the top of her range gets a small suggested raise and starts taking recruiter calls. Someone hired below market sits low in range for years, because a matrix that rewards position in range moves them slowly. A person halfway to a promotion has no cell. Each of these becomes an exception. With no process, exceptions are decided by whoever argues hardest. When a third of increases are exceptions, the exceptions are the real system. They are also the part nobody can explain afterwards.

Many edge cases are not really merit questions at all. A below-market hire needs a market adjustment. A promotion needs a promotion increase. Folding them into merit hides what the company is paying for.

"We pay for performance, so differentiation will happen"

What it looks like: The policy says pay follows performance. The outcomes show strong and solid performers a percentage point or so apart.

Why smart teams fall into it: Differentiation sounds right in a policy. In a conversation it means telling someone they get less than a colleague. With a small budget it means some people get nothing.

What it costs: Under a typical budget the spread is thin. A compensation software provider's data for the first half of 2026 showed median raises of 4.7% for people rated above expectations and 3.3% for those meeting them [4]. Some sectors go much further. A 2025 to 2026 planning survey of 563 organizations found that in software and consulting services, nearly 55% of companies give high performers 150% to 300% of the average increase [3]. Real differentiation of that kind is a decision. Neither a matrix nor judgment produces it by default. A matrix produces it only if its cells are far enough apart. Judgment produces it only if managers are willing to give some people very little.

"It's consistent, so it doesn't need explaining"

What it looks like: The matrix is applied rigidly and every increase matches its cell. But employees have never seen the logic. Managers can only say "that's what the system gave".

Why smart teams fall into it: Consistency feels like its own defense. And publishing the grid feels like inviting arguments.

What it costs: A rule nobody understands earns none of the trust a rule is meant to build. Employees see the number, not the logic. They assume the worst. Managers who cannot explain a raise lose credibility with their team. And in the EU, the criteria for pay progression must be accessible to workers, not just applied consistently [6]. The fix is not to publish every cell. It is to be able to say, in plain words, what moves a raise up or down and why. Our guide on what to do when an employee asks about pay looks at those conversations.

What a sound merit process looks like

Most sound processes are neither pure matrix nor pure judgment. They use a rule for the many and judgment for the few. Then they review both. In our view a sound process has these qualities.

  • The guide sets a range, not a point. Each combination of rating and pay position has room for judgment inside it. Managers decide within the range without asking anyone.
  • Exceptions are named and few. Going outside the range needs a reason in writing and one approval. The share of exceptions is tracked. If it climbs, the guide is wrong, not the managers.
  • Merit pays for merit only. Market adjustments, promotions and pay equity fixes have their own budgets and their own decisions. The merit guide is not asked to repair what it was not built for.
  • Ratings are checked before they become pay. The distribution of ratings is looked at by team and by gender before any raise is set. A skewed distribution is investigated, not paid out.
  • Outcomes are reviewed before letters go out. Someone looks at the proposed raises by rating, team and gender while they can still change. Research on merit pay suggests accountability works better before decisions than after [1].
  • Managers see their own pattern. Each manager can see how their spread compares with others. Compression and outliers become visible to the person who can fix them.
  • The logic can be said in a paragraph. Employees know what moves a raise and why, in words a manager can repeat.

None of this is a template to copy. Each quality is a test a reviewer can put to a company's last cycle.

One more quality matters at this size: The guide is rebuilt each year. Budgets, ranges and the rating distribution all move. A guide set two cycles ago with a 4% budget will overspend or underpay with a 3.2% one. Rebuilding it is a short piece of work. Skipping it is how a matrix drifts out of touch with the money behind it.

The calls only you can make

Several of the choices in a merit process are about philosophy, not mechanics.

How much to differentiate. A wide spread rewards top performers clearly and tells solid performers they are average. A narrow spread feels fair and pays off the wrong people over time. There is no neutral setting.

Whether a solid performer can get nothing. With a small budget, real differentiation may mean a 0% raise for someone meeting expectations who is already well paid. Some companies accept that. Others treat a raise for meeting expectations as a floor.

How strongly to pull pay toward the middle of the range. A guide that gives more to people low in range closes gaps faster. It also means two people with the same rating get different raises, which some employees read as unfair.

Who owns the final number. Central ownership gives consistency and slows things down. Manager ownership is faster and needs a review step to stay consistent. Where you sit says something about how you want managers to lead.

How much managers should know. Some companies show managers the full guide and the budget math. Others give each manager only a pot and a suggested split. More openness builds manager confidence in the conversation. It also makes the guide easier to game.

What to do above the range maximum. A strong performer at the top of the range can get a lump sum instead of a raise, or the range can be questioned. Each sends a different message about what the range means.

The research and the numbers

  • Budgets. US employers planned merit increases of about 3.2% for 2027, according to a July 2026 survey of about 1,000 organizations [5].
  • Spread. A compensation software provider reported median raises in the first half of 2026 of 4.7% for above-expectations ratings, 3.3% for meets and 2.0% for below [4].
  • Differentiation by sector. In a 2025 to 2026 planning survey of 563 organizations, nearly 55% of software and consulting services companies gave high performers 150% to 300% of the average increase. Health care showed the least differentiation [3].
  • Bias in the pay step. A study of about 9,000 employees at one company found smaller raises for women and minorities with the same performance score, job, unit and manager as white men [1]. At a company that then added accountability and transparency to its pay decisions, the gaps narrowed [2].
  • Law. US guidance treats a merit system as a defense only to the extent it accounts for a pay difference [7]. In the EU, pay progression criteria must be accessible, objective and gender-neutral [6].

Our view

We think the choice between a matrix and judgment is a false one. Matrices fail at the edges and judgment fails at scale, so a sound process needs both: A guide with ranges for most people, named exceptions for the few it fits badly and separate budgets for the things merit should not fix.

What makes either approach trustworthy is the review. Ratings are checked before they become pay. Raises are checked by team and gender before letters go out. Exceptions are counted. Without that, a matrix pays out whatever bias the ratings carry and judgment produces patterns nobody chose.

We would argue hardest against switching wholesale from one to the other after a bad cycle. The fintech above did that twice. Each switch fixed the last cycle's problem and created the opposite one.

Sources

  1. Castilla, Gender, race and meritocracy in organizational careers, American Journal of Sociology 113(6), 2008, checked 2026-10-25
  2. Castilla, Accounting for the gap: A firm study manipulating organizational accountability and transparency in pay decisions, Organization Science 26(2), 2015, checked 2026-10-25
  3. Pearl Meyer, 2026 compensation planning: Data and insights (November 2025; 2025 to 2026 Compensation Planning Survey, 563 organizations), checked 2026-10-25
  4. Pave, merit increase data for the first half of 2026 (26 June 2026), checked 2026-10-24
  5. Mercer, Results of the July 2026 QuickPulse US Compensation Planning Survey (31 August 2026), checked 2026-10-24
  6. Directive (EU) 2023/970 on pay transparency, Article 6 (EUR-Lex), checked 2026-10-25
  7. US Equal Employment Opportunity Commission, Compliance Manual Section 10: Compensation discrimination, section 10-IV.F.1, checked 2026-10-25

Compa-ratio or range penetration?

The same pay can be a 0.90 compa-ratio and sit below the band minimum. What each measure answers and where each one misleads.

Which pay gaps can you defend?

An explained gap is not a defensible one. Five ways a pay equity review explains away a real gap and what a review you can rely on looks like.

How should you split the salary budget?

When one budget pays for promotions, market fixes and range moves, merit gets what is left. Five traps and what a sound budget split looks like.

See Resources for every guide and the self-check.