Job leveling framework: how to build job levels, a leveling matrix and the pay bands under them

A job leveling framework is the shared ladder every role in your company sits on. It defines a small number of levels by scope, autonomy and impact, so a Level 4 designer and a Level 4 accountant represent comparable value to the business. Titles, career paths, promotion decisions and salary bands all hang off it. Without one, every pay decision is negotiated from scratch and none of them are defensible.

Most companies build this the wrong way round. They benchmark salaries first, discover the market data assumes levels they never defined, and end up reverse-engineering a ladder from whatever titles people already have. That is how you get four different "Senior Manager" roles being paid three different amounts with no explanation. Leveling comes first. Pay follows from it.

This guide covers what belongs in a level definition, how many levels a company under 200 people actually needs, how to build the matrix, and how to attach salary bands to it once the levels exist. Build a band for a real role below while you read.

What is job leveling?

Job leveling is the practice of sorting every role into defined levels of scope, autonomy and impact, so that the same level means the same thing across functions. It answers one question precisely: how senior is this work, independent of what the job is called and independent of who currently does it.

That independence is the whole point. A title is a marketing artifact, negotiable at offer time and inflated by every competitive hire you lose. A level is a description of work someone is expected to own. When a candidate asks for a Director title, a level framework lets you answer the real question, which is whether they are being asked to set strategy for a function or execute inside one, and pay accordingly.

The five criteria that actually separate levels

Weak frameworks level on tenure and headcount. Strong ones level on what a person is trusted to decide alone. These five criteria are the ones that hold up when someone challenges a placement.

Notice what is absent: years of experience, number of direct reports and job title. Years measure tenure, not contribution, and a framework built on them cannot promote a fast learner or hold a coasting veteran in place. Headcount measures org chart shape, and using it forces your best specialists into management to get paid, which is the fastest way to lose them.

How many job levels should a company have?

Most companies under 200 employees work well with 6 to 8 levels across all functions. Below 6, the gaps are so wide that promotions become rare, enormous and politically charged. Above 8, you are defending distinctions your managers cannot articulate, and the ladder stops being believable to the people climbing it.

A workable default for a 50 to 200 person company: two entry levels, two established individual contributor levels, a senior or lead level, and two leadership levels split into a manager track and a senior individual contributor track that pay the same. The dual track matters more than the count. If your only route to a higher band runs through managing people, you will promote your strongest engineers into jobs they do not want and lose them within a year.

What is a job leveling matrix?

A job leveling matrix is a grid with levels down one axis and job families across the other, where each cell describes what a person at that level in that family is expected to own. The level definitions give you the vertical logic. The matrix makes it concrete enough for a manager to point at during a promotion conversation.

Keep each cell to two or three sentences of observable work. "Owns the quarterly close for one entity with review" is usable. "Demonstrates ownership and drives outcomes" is not, because two managers will read it two different ways and both will believe they are right. The test for every cell is whether a disagreement about a promotion could be settled by reading it.

How to build a job leveling framework in seven steps

This takes a focused week for a company under 200 people, not a quarter. The failure mode is not moving too fast, it is trying to design the perfect ladder before testing it against real humans.

1. List every role, not every person. Group duplicates. A 90 person company usually has 30 to 40 distinct roles, which is a manageable number.
2. Define job families. Engineering, product, sales, marketing, finance, operations, people. Six to eight families is normal. Families answer what kind of work; levels answer how senior.
3. Write the level definitions first, before you place anyone. Use the five criteria above. Writing them with names in mind guarantees you write the ladder around your current org rather than the work.
4. Place a calibration sample. Take 10 roles you are certain about, place them, and check that the levels look consistent across families. Fix the definitions, not the placements.
5. Place everyone else, then calibrate cross-functionally. Get department heads in one room to challenge each other's placements. This session is where a framework either becomes real or becomes a document nobody uses.
6. Attach salary bands to the levels. Benchmark each level against market data, then set a minimum, midpoint and maximum. This is the step that turns a ladder into a pay structure.
7. Publish what you can and run it for a cycle. Levels and expectations should be visible to employees. Whether you publish the bands is a separate decision, though state posting laws are steadily removing the choice.

A framework also changes hiring the moment it exists. You stop opening a requisition for "a senior person" and start opening one for a specific level with written expectations, which means you can screen every candidate against the same level criteria instead of relying on whoever happened to run the first call. It is also what stops offer negotiations from quietly creating a new level nobody agreed to.

Attaching salary bands to your levels

Levels describe expectations. Bands describe money. Keeping them separate is what lets you re-benchmark pay every year without renegotiating your entire career ladder, and it is the difference between a structure that survives three years and one that gets abandoned after the first market shift.

The mechanics are straightforward once the levels exist. For each level in each family, pull market data for the matching occupation and metro, set the midpoint at the market percentile your compensation philosophy targets, then build the range around it. A range spread of 30 to 40 percent suits most professional levels, widening at senior levels where individual variation is larger. Adjacent levels should overlap by roughly 20 to 40 percent so a strong performer can be paid well without a promotion they have not earned.

Once bands exist, level placement becomes measurable. Compa ratio tells you where each person sits against their level's midpoint, and a cluster of people at 80 percent in one level usually means the level is misdefined rather than that everyone is underpaid. Our compa ratio calculator runs that check in the browser, and the guide to building defensible salary bands covers the benchmarking step in full.

Four mistakes that kill a leveling framework

Leveling people instead of roles. The most common failure. Once you place a person rather than the work, the framework encodes your current team's quirks and cannot survive anyone leaving. Define the level, then see who meets it.

Too many levels. Twelve levels in a 70 person company means most promotions are cosmetic and most level distinctions are unexplainable. If two adjacent levels cannot be told apart in a sentence, they are one level.

No dual track. A single management-only ladder converts your best specialists into mediocre managers because that is where the money is. Senior individual contributor levels should map to the same bands as their management peers.

Never recalibrating. Titles inflate, scope drifts and a framework left alone for three years describes a company that no longer exists. Review placements once a year alongside the market refresh of your bands, and treat systematic drift in one function as a signal that the definitions need work.

Why leveling is now a compliance question too

State pay transparency laws changed the stakes. A posted salary range has to be a good faith range for the role, and the cleanest way to show good faith is a documented band tied to a defined level rather than a number invented for that requisition. When a range is challenged, "this is the L4 band, here is the level definition, here is the market data" is an answer. "That felt about right" is not.

Levels also make pay equity analysis possible at all. Comparing pay across a whole company tells you almost nothing, because the mix of roles explains most of the variation. Comparing within a level, across genders or ethnicities, is the analysis that finds real problems, and it only works if the levels are meaningful. See the guides to running a pay equity audit and the state pay transparency laws that trigger this work for most companies.

Frequently asked questions