TL;DR
- Core decision: whether your bands will hold under hiring pressure, or remain a spreadsheet nobody defends.
- When doing nothing is right: when exceptions are vanishingly rare, the path is already slow on purpose, and your governance gives a single owner real authority.
- What has to be true: a hiring manager under pressure at 5pm on a Friday can find the answer, believe it, and not be able to walk around it without it being visible.
- How the options split: into structures that absorb pressure (wide ranges, review thresholds), structures that distribute pressure (many levels, matrices), and structures that translate market pressure into a number your organisation can defend (market-anchored bands).
- Decision rule: if exceptions are climbing faster than headcount, the structure is the problem, not the people using it.
- Outcome to expect: fewer overrides, slower hires you can defend, and a documented record of the times you said no on purpose.
The 5pm Test
The reward lead is at her desk on a Friday, halfway through a half-drunk coffee, when a recruiter pings. There's a candidate for a senior platform engineer. The recruiter has just told the candidate the role is interesting. The hiring manager has already said yes in principle. The candidate's current base is forty thousand above the top of the band, and the recruiter wants to know, by end of day, what the company is going to do. The reward lead opens the band sheet. She can see the range. She can see the candidate is outside it. She has no exception form, no defined approver, no script for what evidence makes the case, and no record of what the company did the last three times this happened. She guesses.
That guess is the system. Not the spreadsheet she spent three years building. Not the benchmark study she refreshed last quarter. The system is what happens at 5pm on a Friday when a hiring manager has found someone they want, the structure says no, and nobody has decided in advance what the answer is.
The usual instinct is to assume the bands are wrong. Maybe the market moved. Maybe the role got re-scoped. Maybe the band is too narrow. All of these are possible. None of them is the most likely cause. The most likely cause is that the band was built from data, and the exception was made from urgency, and urgency wins every negotiation where nobody owns the structure. The test of a banding framework isn't whether it's analytically sound. It's whether a hiring manager under pressure at 5pm on a Friday can find out what they're allowed to offer, and believe the answer. If they can't, the framework is decoration.
Best tools for Payroll & Compensation
When You Genuinely Don't Need to Act Yet
Most articles on this subject open with urgency. This one doesn't. Some teams don't have a problem, and the most useful thing a reward lead can read is the case for leaving a working setup alone. Four honest stages follow.
Stage 1: The current setup is genuinely fine. A mid-sized professional services firm, two hundred and fifty people, has run a single salary band per level for six years. Exceptions last year: three. All three had a written reason. All three went to the same approver, the chief people officer, who said yes to one and no to two. Hiring managers know the rule. They push the recruiter when a candidate is above band; they don't push the reward team. The structure is small, the exceptions are few, and the system has not been challenged. A reward lead in this position should not rebuild anything. They should write down what they have, because the current setup depends on institutional memory that walks out the door when the CPO retires next spring.
Stage 2: Friction is starting to show. A growth-stage SaaS company, eight hundred people and rising fast, has a band structure that worked at three hundred and is now straining. Exceptions have roughly doubled in eighteen months. The reward team can name every one. Approvals still go to a single person, but that person is starting to ask the same questions every time: is this market, is this role, is this a retention play. The pattern is recognisable: the bands are still respected, but only because a single human is enforcing them, and that human is visibly tired. The structure is intact but personal. Action is warranted, but the action is to formalise what already works, not to redesign it.
Stage 3: Real risk has arrived. A larger employer, several thousand staff, has bands that look fine on paper but are quietly being ignored. Recruiters have learned that "market adjustment" gets a faster yes than "above band". Hiring managers ask the recruiter to make the case rather than the reward partner. The reward team has lost track of how many exceptions were approved in the last year because they were made in chat messages, not in a form. Compensation is becoming a private negotiation between recruiter and hiring manager, with the reward function as a witness rather than a participant. This is the stage at which the structure has already failed. The question is whether the failure is recoverable or whether a redesign is needed.
Stage 4: The edge case. A small number of organisations have already redesigned twice in three years, and each redesign has not solved the problem. The bands are sophisticated. The governance is documented. The exceptions still climb. The issue is almost never the band. It's that the organisation is paying for the same role three different ways because the role itself is unstable: re-scoped every quarter, hired against a different brief each time, with a different person owning each hire. No band can survive a moving target. The right move here's not another redesign. It's to fix the job architecture underneath.
Five Questions You Ask Yourself at 11pm
Q1: Do my hiring managers actually know what is in band? The honest answer is usually no. The band sheet lives in a shared drive. The hiring manager has not opened it. They ask the recruiter. The recruiter asks reward. By the time the answer arrives, the candidate has taken another call. The reasoning behind this question is that a band is a behaviour, not a document: it only works when the people using it can act on it without escalation. If a hiring manager can't, in under a minute, find the band for a role they're hiring for, the structure is theoretical. If the reader gets this wrong, they will keep paying above band without knowing it, because every hire will quietly route around a structure nobody can find.
Q2: What did we do the last three times someone was above band? If you can name the cases, the approver, the outcome, and the reason, you've a system. If you can name the cases but not the reasons, you've a habit. If you can't name the cases, you've a problem that has been quietly delegated to recruiters, and they're not equipped to defend it. The reasoning: precedent is what governs pay when the structure doesn't. Without a written trail, the next exception is decided in a vacuum, and each one becomes the new floor. If this question is answered wrong, the reward function loses the ability to say no to anything, because no-one can point to the last time no was said and why.
Q3: Who has the authority to say no? Not the soft no. Not the "let me check". The actual no, with the documentation to back it up, given to a hiring manager of significant standing, in a meeting where the cost of losing the candidate is on the table. If the answer is "the reward team can object but the hiring manager can override", the answer is no-one. The reasoning: authority has to live somewhere specific, and it has to be the kind of authority that survives a recruiter's complaint and a CFO's raised eyebrow. If the reader cannot name the person with the standing to refuse a senior hire, the structure is decorative. If this is wrong, every exception gets approved by default, and the bands become the floor rather than the ceiling.
Q4: What does my exception rate look like over twelve months? Not the count. The rate, against hires. A low headcount year and a high exception count is a different signal to a high headcount year and the same exception count. Without the rate, you can't tell whether exceptions are normal friction or a structural tell. The reasoning behind this is that exceptions look small in absolute terms and devastating as a proportion: a handful of overrides against fifty hires is a structure under strain, and a handful against five is a structure that has stopped working. If the reader treats count as the metric, they'll miss the early warning and discover the failure only when a board or a regulator asks. Getting it wrong means the structure fails slowly enough that nobody intervenes until it's already gone.
Q5: If my reward lead left tomorrow, what would break? Be specific. Which decisions would stall. Which approvals would go to someone who has never made one. If the answer is "everything", the structure depends on a person, not a process. That person is your biggest single point of failure, and the band is one quiet resignation away from collapsing. The reasoning: a structure held together by one person's memory is a structure that survives only as long as that person stays, and a good resignation from a reward lead should not produce a compensation crisis. If this is wrong, the reader discovers on the day the reward lead hands in their notice that the bands, the exceptions, and the precedents all lived in one inbox. The cost of getting it wrong is paid by the next hire, the next counter-offer, and the next audit.
Three Honest Categories the Approaches Split Into
Category 1: Structures that absorb pressure. Wide bands, generous ranges, and review thresholds are designed to make exceptions rare by giving the structure enough room that most candidates land inside it. The bet is that variance is cheaper than governance. When it's right: a fast-growing organisation in a tight labour market, where the cost of losing a candidate exceeds the cost of paying above the median. When it fails: when the band is so wide it stops being a band and starts being a personal negotiation with a number on either end. The structure has not been defended, it has been abandoned. The example to picture: a Series C company with a senior engineering band wide enough to cover three market percentiles, where every hire is technically in band, and the band has stopped communicating anything at all. The mechanism that produces this failure is simple: a band that contains every plausible number contains no number. A hiring manager can offer anything inside the range and call it in-band, the recruiter has no anchor for negotiation, and the candidate has no signal about what the company actually thinks the role is worth. The structure looks compliant on paper because no exception was filed; in practice, internal equity drifts, the reward team can't reconstruct why two people at the same level are paid differently, and the next restructure exposes the inconsistency.
Category 2: Structures that distribute pressure. Many-level structures and job-family matrices work by giving the reward team more granular tools, so that an above-band candidate can usually be placed at a higher level that's in band, or in a different family with a different range, without an exception. When it's right: a large organisation with many roles, where role drift is common and a re-level is more honest than an override. When it fails: when the matrix becomes so complex that even the reward team can't find the right cell quickly, and hiring managers stop trusting the structure because they can't find their way through it. The example: a multinational with a nine-by-twelve matrix of job families and levels, where the right answer for a given hire requires a calculator and a half-hour. The mechanism behind the failure is that granularity only helps if it is navigable. A matrix with too many cells produces slow answers, and slow answers get routed around. The recruiter stops asking which cell applies and starts asking what number can be defended; the hiring manager stops waiting for the reward partner and starts offering; and the matrix, which was designed to make every hire fit the structure, ends up producing informal decisions that the structure can no longer describe. The cost of getting the granularity wrong is paid in lost recruiter time, slow hires, and a reward function that has to apologise for its own structure.
Category 3: Structures that translate market pressure. Market-percentile anchored bands take an external signal (a survey, a benchmark) and pin the band to it, so that the conversation about whether a candidate is expensive becomes a conversation about whether the candidate is above the 75th percentile, with a source. When it's right: a regulated or audit-heavy environment where the structure needs to be defensible to people who aren't in the room. When it fails: when the benchmark and the actual labour market for a specific role diverge, and the structure is defending a number that has stopped being true. The example: a financial services firm whose bands are pinned to a benchmark that doesn't cover the niche quantitative role they're trying to hire, where the structure is producing a defensible answer to the wrong question. The mechanism behind this failure is that an anchor is only useful if it points at the thing you actually need to govern. When the benchmark covers a broad category and the hire is in a narrow specialism, the band describes a market that doesn't exist for this role, and the structure produces confident, sourced, and wrong numbers. The cost is that the reward team ends up defending a benchmark rather than a pay decision, and the hiring manager ends up losing the candidate because the band is technically right and practically irrelevant.
Five Diagnostic Questions You Can Self-Assess Against
D1: Can a hiring manager find the right band for a role in under sixty seconds? Don't ask the reward team. Sit with a hiring manager and watch them try. Time it. If they can't find it without help, the answer to "is the structure usable" is no, regardless of how well it's designed. To answer this about your own organisation, pick a hiring manager who has hired in the last six months, ask them to open the band sheet for a role they have hired for, and count the seconds from the moment they start to the moment they name the band. Anything over a minute is a signal. Repeat with two more managers; if more than one struggles, the structure is theoretical.
D2: When was the last time someone said no to an exception, and was that decision written down? Look at the record. If the last "no" was years ago, either the structure is perfect (unlikely) or the system has stopped surfacing the cases that should have been turned down. Either way, the answer matters. To answer this for yourself, ask the reward team to pull the last twelve months of exception decisions and identify any that were turned down. If the file is empty, ask why. If the file doesn't exist, ask where the decisions live. The absence of a "no" is not, by itself, evidence of a healthy structure; it is evidence that the structure has either not been tested or has stopped being used.
D3: What proportion of your hires in the last year involved an exception of any kind? Calculate it against total hires, not against total headcount. A rate that's creeping up is a leading indicator. A rate that's suddenly high is already a problem. To get this number, ask the reward team for the count of hires in the last twelve months and the count of those that involved an exception, in any form, including a re-leveling, a sign-on, or an above-band offer. Divide. Repeat for the prior twelve months. The trend matters more than the absolute number.
D4: Who can approve an exception, and do they know they can? Look at your last twelve exceptions. If the approver is the same person every time, ask whether that person has a delegate who has actually approved one. If the answer is no, the structure has a person, not a process. To check this in your own organisation, look at the last twelve exception decisions. If one name appears every time, find out who that person's delegate is, and ask the delegate when they last approved an exception. If the answer is "never" or "I'm not sure", the structure has a single point of failure.
D5: If you had to defend every band exception from the last year to a regulator, a board, or a journalist, could you? Pull the file. Not the summary. The file. If you can't reconstruct the reasoning, the exception, the evidence, the alternative considered, and the outcome in writing for each one, you don't have a defensible structure. You have a pattern that has not yet been challenged. To answer this for yourself, pick five exceptions from the last year at random and try to reconstruct each one from the documents you hold. If you can't, the structure is held together by memory, and memory is not a defence.
Five Banding Approaches, Reviewed
Broad Bands With Wide Ranges
A band is defined per level, and the range within that level is wide enough to absorb most variance without an exception. The intent is that almost every candidate lands inside the structure, and the structure is rarely overridden. It earns a place because it's the lowest-friction option for fast-moving organisations: hiring managers can usually say yes without escalating, recruiters don't have to walk a candidate through a slow approval, and the reward team spends less time on individual cases. The weakness is that a band wide enough to absorb every candidate is a band that communicates nothing. It doesn't tell a hiring manager what the company thinks the role is worth. It doesn't tell a candidate what the company has decided they're worth. It doesn't tell the reward team whether a pay decision is consistent or arbitrary. In practice, wide bands tend to drift: the same level ends up paying materially differently for people doing materially the same work, because the band is so wide that any number inside it's defensible. The structure has not been defended, it has been replaced by personal negotiation wearing a band's clothing, and the reward function ends up watching pay decisions it cannot explain.
Narrow Bands With Many Levels
A band is defined narrowly, but the structure has many levels, so an above-band candidate can usually be re-leveled rather than excepted. The intent is granularity: every hire fits the structure, and the structure is rarely broken. It earns a place in organisations with stable role definitions and high hiring volume, where the cost of an exception is genuinely high, and where role drift is rare enough that re-leveling is a real option rather than a fiction. The weakness is operational. A structure with many levels is a structure that takes time to work through. Hiring managers need to know the difference between level four and level five, and they rarely do. Recruiters need to be able to argue for a level, and they often can't. The result is that re-leveling becomes the slow path, and exceptions still happen because the slow path is too slow. The structure is technically respected but practically circumvented, and the reward team ends up re-running the same debate on every hire.
Market-Percentile Anchored Bands
Each band is pinned to an external market signal, usually a survey or benchmark, with the band midpoint tied to a specific percentile and the range sized to absorb variance around that percentile. The intent is to translate a market conversation into an internal one: instead of arguing about whether a candidate is expensive, the conversation becomes whether the candidate is above the agreed percentile. It earns a place in organisations that need the structure to be defensible to people outside the room: auditors, regulators, boards, and in some jurisdictions, worker representatives. It's also useful in industries where the market is the dominant force and internal equity is less of a constraint. The weakness is that benchmarks lag. They describe a market that was, not a market that's. They're also broad: a percentile for "senior engineer" covers a wide range of actual specialisms, and the structure is producing a defensible answer to a question that has stopped being the right one. In jurisdictions where the first gender pay gap reports are due in 2027 and joint pay evaluations are required where the gap runs above five per cent, a structure anchored to a stale benchmark is the structure that fails the audit before the audit starts.
Job-Family Matrices With Levels
A two-dimensional structure: one axis is job family (engineering, product, design, sales, and so on), the other is level. Each cell in the matrix has its own band. The intent is to recognise that the same level in different families is paid differently, and that a candidate who is expensive for one family is reasonable for another. It earns a place in larger, more diverse organisations where the cost of a single-band structure is visible role-by-role distortion, and where the matrix reflects real, defensible differences between roles. The weakness is complexity. A matrix is only as good as the team's ability to read it. If the reward team can't quickly answer which cell a candidate falls into, the matrix is producing slow answers and the hiring manager will route around it. The structure has been respected in design and abandoned in practice, and the reward function ends up maintaining a document that no-one uses.
No Formal Bands With a Review Threshold Instead
There's no formal band structure. Every offer is reviewed against a threshold (a market reference, a budget, a guideline), and the review is what governs the decision rather than the band. The intent is flexibility: the organisation can pay what it needs to pay, when it needs to pay it, without re-engineering a structure for every market move. It earns a place in very small organisations, in fast-moving early-stage companies, and in any context where the formal structure would be more fiction than fact. The weakness is that without a structure, there's nothing to defend. Every decision is a precedent, and precedents drift. Review thresholds become personal, and personal review thresholds become a quiet, undocumented compensation system that no-one can explain. In jurisdictions with pay transparency posting requirements and bans on pay history questions, the absence of a band means there is nothing to put in the posting, and the threshold becomes the disclosure, with all the drift that implies.
The Decision Table
| Situation | Scale | Setup | Primary Pain | Recommended Starting Point |
|---|---|---|---|---|
| Exceptions rare, single approver, low variance | Under 500 | Working bands, light governance | Documentation risk if owner leaves | Formalise what you have; write the rules down |
| Exceptions rising, single human enforcing structure | 500 to 2000 | Bands respected by habit | Personal dependency on one approver | Build the exception form and delegate approval |
| Exceptions common, recruiter-mediated overrides | 2000+ | Bands theoretical | Structure ignored under pressure | Redesign for pressure-absorption: wider ranges or many levels |
| Audit-heavy industry, board scrutiny | Any | Pinned to external benchmark | Need for external defensibility | Market-percentile anchored bands |
| Multiple roles with materially different markets | 2000+ | Single band per level causing distortion | Internal inequity by family | Job-family matrix with levels |
| Sub-200 staff, fast-changing roles | Under 200 | Bands rebuilt twice in three years | Moving target | Review threshold instead of formal bands |
| Stable workforce, low attrition, low hiring | 500 to 1500 | Wide bands, rare exceptions | Inconsistent pay for similar work | Narrow bands within current level structure |
| Hot skill breaking the structure | Any | Specific niche roles above band | Pattern of exceptions in one family | Out-of-cycle review for that family; do not widen global bands |
Designing the Exception Process
The exception path is the band. A band without a defined exception path is a band that will be overridden informally, because the informal path is faster than asking. The design intent is to make the formal path visible, slightly slow, and well-evidenced, so that the cost of bypassing it's higher than the cost of using it. The path isn't there to say no. It's there to make the yes defensible.
The exception path should be a form, not a chat message. The form should require the hiring manager to articulate the reason in their own words, not to tick a box. It should require the recruiter to record the candidate's current pay and the candidate's expectation. It should require the hiring manager to record what would happen if the candidate were hired at the top of band instead. It should require the reward partner to record the alternative considered (re-leveling, a different family, a sign-on versus a base, a delayed review). And it should be reviewed by a single approver who has the standing to say no and the documentation to back it up.
| Scenario | Who Approves | What Evidence Is Required |
|---|---|---|
| Candidate above band, market move cited | Reward Director | Written reason, current pay, candidate expectation, alternative considered |
| Retention case, existing employee counter-offer | Hiring Manager's Skip-Level + Reward | Flight risk assessment, internal equity impact, replacement cost |
| Re-leveling instead of exception | Reward Lead | Updated job description, level justification, hiring manager sign-off |
| Sign-on bonus used in lieu of base | Reward Lead | Total package comparison, payback period, internal precedent |
| Hot skill, niche role above market | Reward Director + CFO sign-off | Market evidence (anonymised), business impact, equity implications |
| Long-tenured employee below band | Reward Lead + HRBP | Tenure, performance, internal equity, retention risk |
| Equity refresh used to bridge gap | Reward Lead + CFO | Total package comparison, dilution impact, vesting schedule |
The form isn't bureaucratic theatre. It's the record that turns a decision into a defensible one. In jurisdictions where bans on pay secrecy clauses are in force alongside the EU Pay Transparency Directive, a written record per exception is also the only way to demonstrate to worker representatives that the joint pay evaluation triggered by a gender pay gap above five per cent rests on something other than conversation. A reward lead who can pull twelve months of these forms and reconstruct every override has a structure. One who can't is one quiet quarter away from discovering the structure never existed.
Keeping Bands Current Without Rebuilding Them Annually
Annual rebuilds are expensive and disruptive. They also tend to produce structures that are technically correct but operationally unfamiliar, because hiring managers have spent twelve months learning the old one. The alternative is a refresh cadence that's light by default and triggered when something specific changes. The refresh isn't a redesign. It's a check: are these bands still pointing at the right market, and are they still governing the right decisions?
The cadence has three parts. First, an annual light refresh that confirms the structure is still the right structure, without touching the bands. This is a review, not a rebuild. Second, a rolling check on the bands that matter most: roles with high exception rates, roles in tight markets, roles whose scope has visibly shifted. Third, a defined set of triggers that force an out-of-cycle review of a specific band, not the whole structure. The triggers are what keep the system honest without forcing a yearly upheaval. Where postings must carry a salary range and a benefits description in good faith, an out-of-cycle refresh also protects the posting from advertising a band the company has stopped using internally.
| Trigger | What It Forces | Who Decides |
|---|---|---|
| Exception rate above a defined threshold for a band | Targeted review of that band | Reward Lead |
| Material market move for a specific role | Out-of-cycle refresh of that role's band | Reward Director |
| Role re-scope by hiring manager | Re-leveling review before next hire | Reward Lead |
| M&A or major restructure affecting a family | Family-wide review | Reward Director + CFO |
| Regulatory change affecting disclosure | Review of bands used in postings | Reward Director |
| Loss of three or more hires in a quarter to above-band offers | Targeted review with hiring manager | Reward Lead + Hiring Manager |
| Internal equity complaint upheld | Review of band and adjacent bands | HRBP + Reward Lead |
The point of the cadence is to make the structure a living thing that responds to specific signals, not a fixed thing that gets rebuilt every twelve months because that's the calendar. A structure that responds is a structure that survives.
What to Put in Writing
Written records are what turn a good decision into a defensible one. The artefacts below are the minimum. A reward function that has none of them is running on institutional memory and goodwill. One that has all of them has a structure that will survive the reward lead leaving, the CFO asking questions, and a regulator knocking. In jurisdictions where vertical direct effect now applies to public-sector employers and directive-consistent interpretation applies to private employers, the artefacts are also what allows a national court to read existing national law in conformity with the Directive without the reward function having to argue its case from scratch.
| Artefact | Who Owns It | When It Is Written | What It Prevents |
|---|---|---|---|
| Band structure document with rationale | Reward Lead | At design, refreshed on major change | Drift, inconsistency across teams |
| Band sheet per role, accessible to hiring managers | Reward Lead | At design, refreshed annually | Recruiter-mediated guesswork |
| Exception form template | Reward Lead | At design | Informal overrides, undocumented decisions |
| Exception log | Reward Lead | Per exception, as it happens | Loss of decision history, repeat mistakes |
| Approver delegation matrix | Reward Director | At design, refreshed on role change | Bottleneck on a single individual |
| Refresh cadence and trigger list | Reward Director | Annually | Surprise rebuilds, ignored signals |
| Annual light refresh memo | Reward Lead | Annually | "When did we last look at this?" |
| Out-of-cycle review record | Reward Lead | Per trigger | Untracked structural drift |
| Disclosure controls for postings | Reward Lead + Legal | Per relevant regulation | Misaligned internal and external bands |
| Internal communications on bands | Reward Lead + Comms | At launch, at major change | Hiring manager uncertainty, candidate confusion |
The artefacts aren't the work. They're the evidence the work happened. A reward lead who is asked in a board meeting why the engineering band is what it's, and who can produce a document that says when it was set, what data it was set against, who approved it, and what the exception rate has been since, is a reward lead with a defensible structure. One who can't is a reward lead with a spreadsheet.
Questions to Ask Before You Commit
On your current exception pattern. Ask your reward team: what proportion of our last fifty hires involved any form of exception, and how does that compare to the year before? A bad answer is vague. A good answer is a number, a direction, and a named cause.
On the approver. Ask who has actually said no to an exception in the last twelve months, and what the case was. A bad answer is "we don't really get to no". A good answer names a case, the reason, and what happened next.
On delegation. Ask who approves exceptions when the usual approver is on leave, and whether that person has actually approved one. A bad answer is "their deputy, in principle". A good answer is a name and a precedent.
On the band itself. Ask what would have to change about our market for this band to be wrong, and how we would know. A bad answer is silence. A good answer is a trigger, a source, and a cadence for checking.
On documentation. Ask what would happen to our banding if the current reward lead resigned next month. A bad answer is "it would be a rough quarter". A good answer is a named interim owner and a handover document.
On the hiring manager experience. Ask a hiring manager, not the reward team, how they find the right band for a role. A bad answer is "I ask [name]". A good answer is "I open [document]".
On internal equity. Ask how we would know if two people doing the same role were being paid materially differently for non-objective reasons. A bad answer is "we'd hear about it". A good answer is a check, a cadence, and a defined threshold for action.
On disclosure. Ask whether the bands we publish in job postings match the bands we use internally, and how we would know if they drifted. A bad answer is "they should be the same". A good answer is a control and a named owner.
On refresh. Ask when the bands were last reviewed, and what triggered the review. A bad answer is "last year, I think". A good answer is a date and a cause.
The Cost of Getting This Wrong
The invoice never arrives. The cost of a banding structure that fails under pressure shows up in three places, and none of them is on a purchase order. The first is the cumulative cost of exceptions that nobody tracked. Each one is small. A few thousand above band, a sign-on bonus, a re-leveling that quietly became the new normal. A year of them, across a hiring plan, is a material sum that no budget line item will show. The second cost is the loss of internal equity that compounds over time. Two engineers at the same level, paid materially differently for reasons neither of them can see, will eventually leave, or stay and resent, and both outcomes are expensive. The third cost is the slow loss of authority from the reward function itself. Every informal override is a precedent that the structure is advisory. Once enough of those accumulate, the reward team isn't governing pay, it's witnessing it.
So the question to put to a board, a CFO, or your own reward lead, isn't "are our bands defensible". It's "if we had to defend every pay decision we've made in the last twelve months, could we, and would the defence hold". If the answer is no, the structure isn't failing. It has failed. The remaining question is how long it takes for someone outside the function to notice.
When You Are Ready to Go Further
If your exception rate is climbing, if your approver is tired, if your hiring managers have learned to route around the structure, the next step isn't a vendor demo. It's a clear-eyed look at what your banding is actually doing, where it's breaking, and what a structure that holds under pressure would require. HROpsLab does independent comparison work across the tools reward teams use to design, govern, and refresh their structures. We don't sell software, we don't provide payroll services, and we don't give legal or financial advice. We publish what we find, and we let the reader decide. If you want to see how your current approach compares to the structures other reward teams have landed on, our comparison library is a reasonable place to start.
If you're further along and want a second opinion on a specific design choice, an exception pattern, or a refresh cadence, our editorial team will read what you send and tell you what we see. We won't pitch you a product. We will tell you what we would ask if we were sitting in your seat.
Frequently Asked Questions
How wide should a salary band be?
A band wide enough to absorb most variance without an exception is a band that has stopped communicating anything, because almost any number is defensible inside it. A band narrow enough to mean something specific is a band that will generate exceptions under pressure. The honest answer is that there's no universal figure, and the right width depends on how volatile your market is, how fast your hiring manager population turns over, and how much authority your reward team is willing to delegate. Start with what you've, measure your exception rate, and adjust the bands that are generating exceptions rather than the ones that aren't. A structure that has been adjusted on evidence is more durable than one that was set on a benchmark.
What do I do when a candidate is above band?
The first move isn't to say yes. The first move is to ask what would change if the candidate were hired at the top of band instead, and whether the answer to that question is genuinely "they would walk". If it's, the second move is to use your exception path: record the reason, the alternative considered, and the evidence. If you don't have an exception path, you've a decision to make about whether the candidate is the right one to build the path around. Saying yes once without a path is the moment the path stops being built, and the next above-band hire will be argued against that precedent rather than against the band.
How often should bands be refreshed?
Lightly, by default. An annual review to confirm the structure is still the right structure, without touching the bands, is the minimum. A rolling check on the bands that matter most: roles with high exception rates, roles in tight markets, roles whose scope has shifted. And a defined set of triggers that force an out-of-cycle review of a specific band. The calendar isn't the right trigger. Specific signals are, and the triggers should be written down so a future reward lead inherits the cadence rather than reinventing it.
Should I show bands to current staff?
Yes, in most cases. A structure that hiring managers don't know about is a structure that can't be defended. A structure that staff can't see is a structure that will be assumed unfair. The disclosure should match the bands you publish externally, because drift between the two is itself a problem. If your market is materially different internally and externally, the question to answer first is why, not whether to disclose, and the answer should land in a document a worker representative could read.
Who should approve exceptions?
A single named approver, with a documented delegate who has actually approved one. The approver needs the standing to say no to a senior hiring manager, the time to read the case, and the documentation to defend the decision later. A committee is slower but more defensible for material cases. A chat message isn't approval, and treating one as approval is how the exception log ends up empty while exceptions keep happening.
What about long-tenured staff who are below band?
Treat it as a band problem, not a person problem. A long-tenured employee below band is evidence that the band has moved, the market has moved, or both. The right move is usually a targeted adjustment for the affected cohort, funded and documented, rather than a quiet fix for one person. The exception is the same as any other: written, evidenced, and approved by someone with the standing to defend it, because a cohort adjustment done quietly is a precedent waiting to be challenged.
How do I handle a hot skill that breaks the structure?
Don't widen the global band. Widen the band for the specific role, or create a separate band for the specific niche, with a defined scope and a defined expiry. The hot skill is a signal that the market for that skill has moved, not that your entire structure is wrong. A targeted fix is defensible. A global widening is a precedent, and the next hot skill will inherit the wider band without the same justification.
What if my hiring managers simply ignore the bands?
The bands aren't the problem. The governance is. A hiring manager who ignores a band is a hiring manager who has learned that ignoring the band has no cost. The fix is to make the cost visible: a slower approval, a required justification, a record that the override happened and why. If the override is happening and the record isn't being kept, the structure has already failed, and the next audit, board paper, or journalist's question will be the moment that failure becomes visible to people outside the function.
HROpsLab publishes independent, evidence-led comparison work for reward, payroll and compliance teams. No vendor payola, no software sales, no advice dressed up as analysis.
