Job family or level: Which comes first?
Levels first, families second. Most stalled projects are stuck on family boundaries that change no one's pay.
Short answer: Levels first. A company-wide level framework drives offers, promotions and posted ranges. It can be written before any job family boundary is settled. Draw broad families quickly by the work people do, not by department. Split them only where pay or career paths differ. Most job architecture projects stall by perfecting families first.
Eight months of boxes
Picture a 140-person health-tech company in Boston. Eight months ago it started a job architecture project. The first deliverable was a family tree: 14 job families, 52 sub-families and a color-coded chart that covers every role the company has or might hire.
Since then the project has held six workshops. Most of the time has gone on boundary questions. Does data science belong with engineering or with product? Is customer success a family or a sub-family of sales? Should clinical operations have its own family when it has four people? Every answer reopens another one.
Meanwhile nothing about pay has changed. There are still no levels. Managers set offers by comparing with whoever they hired last. A new law in a state where the company hires remotely now requires a pay range on every job ad. Recruiting is making up ranges role by role. The CFO asks the head of people a fair question: When will any of this help us pay people?
This company is a composite of situations common at this size. It is not a client and the details are illustrative.
Why the order matters
Job architecture has three building blocks that are easy to confuse.
- A function is where a job sits in the organization: The department and the reporting line. It answers "whose team is this?"
- A job family is a group of jobs that do similar work and draw on similar skills, wherever they sit. It answers "what kind of work is this?"
- A level describes the size of a job: Its scope, complexity, autonomy and impact. It answers "how big is this job?"
All three matter. They do not matter equally for pay. They do not take equal effort to settle either.
Levels carry most of the weight in pay decisions. Pay ranges attach to them. Promotions move people between them. Equal pay questions compare jobs of similar size and value across families, which is a question about levels more than about family boundaries [1]. Families matter mainly where markets differ, for example where engineers cost more to hire than people in operations. Functions matter for reporting and budgets. They hardly matter for pay at all.
The effort runs the other way. Families invite endless debate because there is no single right answer. Public occupation systems show this: The US Standard Occupational Classification groups jobs by the work performed [6]. A company building its own tree faces the same judgment calls with less time. Levels, by contrast, can be defined once for the whole company, in language that applies to every family.
So the tension is between two good goals. The first is completeness: A structure that covers every role and every boundary before anything is used. The second is momentum: A structure that starts helping with pay decisions this quarter. Most stalled projects chose completeness without deciding to.
None of this means families are unimportant. They become important in two places. The first is where markets differ enough that one range per level would overpay one group or underpay another. The second is where career paths genuinely differ, for example where a specialist can grow without managing people in one family but not in another. Both are real. Both can be handled once levels exist, by splitting a family where the evidence shows a difference. Neither requires the full tree to be drawn first.
Many companies are in the middle of this. In a survey of 525 organizations fielded in late 2025, a compensation software and data provider found that 51.4% had a formal job architecture and roughly 30% were planning or building one. It also found that 22% did not use job leveling to inform pay at all [2].
Five ways job architecture projects stall
Each of these feels like diligence at the time. Each delays the part of the work that changes pay.
"Start with the perfect family tree"
What it looks like: The project begins by mapping every job into families and sub-families, with definitions for each. Levels are planned for "phase two".
Why smart teams fall into it: It seems logical to sort jobs before sizing them. Families are visible and easy to discuss. Everyone has an opinion about where their team belongs.
What it costs: Family boundaries are judgment calls with no right answer. Whether data science sits with engineering or product changes very little about how anyone is paid. It changes a great deal about how people feel, which is why the debate never ends. Months go on boxes while offers, promotions and posted ranges keep being decided with no structure at all. When phase two finally arrives, the company has often reorganized and the tree needs redrawing.
The warning signs are easy to spot once you look for them. Workshops end with more open questions than they started with. The family chart has been revised more times than any job has been leveled. People ask about their family before they ask about their level. And nobody on the project can name a single pay decision the work has changed so far.
"Our families are our departments"
What it looks like: Each department becomes a family. Marketing, finance, engineering and sales each get one. Every job in them inherits it.
Why smart teams fall into it: It is fast and everyone understands it. The data already exists in the HR system.
What it costs: Departments follow reporting lines. Families should follow the work. A data analyst in marketing and one in finance do similar work and compete in the same market, but a department-based tree puts them in different families with different ranges. A program manager in engineering and one in operations may do near-identical jobs. And every reorganization rewrites the family tree, along with the ranges attached to it. Public classification systems group jobs by the work performed for this reason [6].
"Levels can wait until the families are final"
What it looks like: The team treats levels as something to write inside each family, so no level work starts until families are agreed.
Why smart teams fall into it: Level descriptions often mention family-specific skills, so it seems they cannot be written without knowing the families.
What it costs: Most of what defines a level is the same across the company: How much scope a job has, how complex its problems are, how much it decides alone, how far its impact reaches. That can be written once, for every job, before a single family boundary is settled. Family-specific examples can be added later. Waiting means the most useful part of the structure is held hostage by the least urgent. It also means the company has nothing to attach ranges to when a posting law or an employee question arrives.
Company-wide levels also make the family debate easier. Once two jobs are known to be the same size, arguing about which family they belong to loses most of its heat, because the answer no longer decides whether one of them is worth more. Many family disputes are really level disputes in disguise.
"Every family needs its own ladder"
What it looks like: Each family leader writes their own level ladder in their own words. Engineering has seven levels, sales has five, customer success has six. Nobody checks whether a level 4 means the same thing in each.
Why smart teams fall into it: Family leaders know their work best. Letting them write the ladder gets buy-in and saves time.
What it costs: Separate ladders cannot be compared. A level 4 in one family may be a level 3 in another. That makes cross-family promotions confusing, internal moves hard to price and equal pay questions hard to answer. In the EU, pay structures must allow a comparison of whether jobs are of equal value on common criteria such as skills, effort, responsibility and working conditions [1]. Our guide on whether the EU will make you evaluate jobs looks at that comparison. One company-wide level framework, with examples per family, keeps the comparison possible.
"More families means more precision"
What it looks like: The tree keeps growing. Every specialty becomes a sub-family, some with one or two jobs in them.
Why smart teams fall into it: Finer groups feel more accurate. Specialists like seeing their work named.
What it costs: Every family multiplies the work that follows. Each needs market matches, ranges at every level and someone to maintain it. A company of 140 with 52 sub-families has more boxes than it has groups of comparable jobs. Many families end up with one person, which makes ranges meaningless and pay comparisons impossible. Precision that does not change a pay decision is cost without benefit.
The cost keeps growing after launch. Every new role has to be placed in the tree, every market survey has to be matched to it and every reorganization tests it. A small number of broad families can absorb new jobs without new boxes. A large number of narrow ones needs a decision every time someone is hired into something slightly new. At 140 people, the team that maintains the structure is usually one person with other work to do.
What a sound order of work looks like
There is no single right architecture. There is an order that keeps a project moving and useful. In our view it has these qualities.
- Levels come first, for the whole company. One set of level definitions describes scope, complexity, autonomy and impact in language that applies to every job. It is drafted before family boundaries are settled.
- Families are broad and drawn in days, not months. A first set groups jobs by the work they do. It is good enough when each family holds several jobs and the boundaries do not change anyone's pay.
- Functions stay separate. Reporting lines live in the org chart. They do not decide families or levels.
- Every job gets a family and a level. The title follows. Titles are set from the structure, not the other way round. Our guide on how to level employees at a startup covers placing people on levels.
- Families split only where pay differs. A new family or sub-family is created when its market or its career path is genuinely different. Otherwise a level and a title are enough.
- Ranges can attach early. As soon as levels exist, pay ranges can be set for them, by family where markets differ. The structure starts paying for itself.
- Boundaries can be refined later without re-leveling. Moving a job between families changes its label and perhaps its range. It does not change its size.
- One person owns it. Someone decides disputes and records why, so the structure does not drift back into debate.
None of this is a build plan to copy. Each quality is a test a reviewer can put to a project already under way.
One practical check helps more than any chart. Take three real decisions the company faces this quarter, such as an offer, a promotion and a posted range. Ask whether the structure as it stands would make each one easier. If the answer is no for all three, the project is working on the wrong part.
The calls only you can make
Several of the choices in a job architecture are about the company you want to be.
How much families should drive pay. Some companies pay one range per level across all families and add premiums where markets demand it. Others set separate ranges per family. The first is simpler and easier to explain. The second follows the market more closely. Our guide on which salary data to trust covers the market side.
How many levels. Fewer levels mean bigger jumps and fewer promotions. More levels mean more frequent steps and finer distinctions to defend. Our guide on how many job levels you need looks at the trade-off.
How titles relate to levels. A strict title policy ties every title to a level and makes the structure visible. A loose one lets teams keep titles they like and hides the structure. Titles are promises to people, so this is as much about culture as structure.
How open the structure is. Sharing level definitions with everyone helps people see their path. It also invites comparison, which the company must be ready to answer.
Whether to buy a framework or build one. Commercial leveling frameworks offer tested language and easy market matching. A framework built for the company fits its work and culture more closely. Many companies buy the language and own the decisions about how it applies.
Who gets a say. Involving managers in drawing families builds buy-in and slows the work. Deciding centrally is faster and needs more explaining afterwards.
The law and the numbers
- How common it is. In a survey of 525 organizations fielded in November and December 2025, 51.4% had a formal job architecture, roughly 30% were planning or building one and 22% did not use leveling to inform pay [2].
- EU. Pay structures must allow comparison of equal value on objective, gender-neutral criteria. The criteria for pay and pay progression must be accessible to workers. Pay gap reports and joint pay assessments are built on categories of workers doing the same work or work of equal value [1]. All three rest more on levels than on family boundaries.
- US. Pay range posting laws require a range for the specific job being advertised [3] [4]. A level with a range behind it is the simplest source for one. Our note on what changed in US pay range laws in 2026 tracks the dates.
- EU guidance. The EU's 2026 toolkit on gender-neutral job evaluation assesses jobs, not the people in them, against objective criteria [5].
- Classification. The US Standard Occupational Classification is a task-based system that groups occupations by the work performed [6].
Our view
We think most job architecture projects at this size should build levels first and families second. A company-wide level framework can be drafted in weeks and starts helping with offers, promotions and posted ranges at once. A first set of broad families can be drawn in days and refined as the company learns where pay really differs.
We would argue hardest against perfecting the family tree before anything else. It is the most common reason projects stall and the part that changes pay the least. The test for any family boundary is simple: Would moving this job change what we pay it? If not, it can wait.
Levels first is not a shortcut. It is the order in which the work starts paying back. A company that has levels and rough families can price jobs, answer pay questions and post ranges. A company with a perfect tree and no levels can do none of those things.
Where we help
Our job architecture work starts with a company-wide level guide, then draws broad families and maps today's titles to both, so ranges can attach quickly. The entry point is our Leveling Starter Kit, for up to three job families. If a conversation would help first, Book a free call.
Sources
- Directive (EU) 2023/970 on pay transparency, Articles 3, 4, 6, 9 and 10 (EUR-Lex), checked 2026-10-25
- Salary.com, 2026 State of Pay and Compensation Practices Report, press release (14 April 2026; 525 organizations, November to December 2025), checked 2026-10-25
- Washington L&I, Administrative Policy ES.E.2, Job Posting Requirements (issued 3 December 2025), checked 2026-10-24
- Code of Virginia § 40.1-28.7:12 (Virginia Legislative Information System), checked 2026-10-20
- European Institute for Gender Equality, EU-wide guidelines on gender-neutral job evaluation and classification: Step-by-step toolkit (26 March 2026), checked 2026-10-25
- US Bureau of Labor Statistics, Standard Occupational Classification, 2018, checked 2026-10-25