By employee, by role, or by driver? The first decision in an HR model
Every workforce planning project starts with a question that looks technical and isn't: at what level do we model people?
It's usually answered in the first week, in passing, often by the consultant. And it is the most expensive decision to undo in the whole model — because dimensions, security, input screens and the integration with Finance all hang off it.
It's two questions, not one
What's normally called "level of detail" is two independent axes that we tend to treat as one.
First axis — what a row represents. It can be by employee (one row per person) or by role/position (one row per position). More detail on one side, more aggregation on the other.
Second axis — where the number comes from. It can be typed in (decided line by line), by driver (calculated from a rule), or inferred (deduced from history).
The two axes combine. By far the most common combination is by employee with typed movements — and it works well for annual budgeting. Role and driver gain weight as scale and horizon grow.
Before choosing: what does the company already use?
That's the question that should come first, and rarely does.
The model will be checked against something in the first month of use — the headcount report HR already publishes, actual payroll, the number the leadership already knows. A different level of detail from what already exists means reconciliation cost every cycle.
That's not a blocker. Sometimes the new level is exactly the project's value. But it's worth choosing with that cost on the table, not discovering it in February.
By employee: the default, and why
It mirrors what HR already has in its system, and it's what most clients expect from a budget process.
For:
- Maximum customisation. Real hire date, real salary, specific benefit, individual eligibility. Nothing has to be approximated.
- Direct reconciliation against payroll and the master file.
Against:
- Sensitivity. It's literally a person inside the system. That shifts the security conversation from "access control" to "who can see what, cell by cell".
- Loses relevance at scale and long horizon. Year 3 of a five-year plan with a person's name on the row is fiction that looks like precision.
By role: only works if standardisation is real
Modelling by role assumes standardised salary bands and job titles that exist in practice, not just on the org chart.
Without that, you build an average nobody recognises — and the model loses credibility the first time someone checks their own area.
There's also an argument that shows up all the time and almost never holds up: using role to hide individual salary. Where's the anonymisation gain when only one person in that cost centre holds that role? In practice, the protection only kicks in at the higher levels of the company — which is exactly where confidentiality matters most and where aggregation protects least.
By driver: needs a measured relationship, not an intuition
Drivers work where there's a stable, already-measured relationship between volume and people: tickets per hour, tonnes per day, beds per shift, cases per store.
Where that measurement doesn't exist, the driver is a guess in the shape of a formula — and that's worse than a guess, because the formula lends authority to the number.
A simple test before adopting one: does anyone already report that ratio today, without the planning model? If the answer is "we'd have to build it", the driver isn't ready to carry headcount yet.
By inference: today's default expectation
Today it's the most common expectation at the first meeting: that the system returns the number instead of receiving it.
It works where there's reliable history, a long-enough series, a stable relationship — and someone willing to validate what comes out.
Without that, the model still returns a number. It just starts deciding on its own.
Who sees the model
Before locking in the level of detail, three questions about the process (not about the tool):
- How many people take part in the cycle?
- Is the process collaborative or centralised in HR and FP&A?
- How confidential are the movement decisions?
Cell-level security here isn't a "nice to have" — it's part of the model's definition. Finance needs personnel cost per cost centre; Finance doesn't need individual salary. That has to be designed, not agreed informally.
The risk almost nobody puts on the table
Combine three decisions that, on their own, look reasonable:
by employee + long horizon + collaborative process.
The reduction plan now carries names inside it. People who are still working, in a system several teams can open, and the name stays there for the whole cycle.
A decision nobody has made yet, living inside the model.
You can mitigate this, and it's worth designing on purpose:
- Named detail with an expiry. Named until month X of the cycle, by position after that.
- A separate version with restricted access for reduction scenarios, outside the collaborative version.
- Reduction modelled by position and cost centre, not by person — the person shows up in execution, not in the plan.
After the decision, the model has predictable layers
Once the level is set, the rest of an HR model is reasonably stable across clients:
- Inputs — personnel master (employee id, role, level, cost centre, status), charges and provisions (social security, severance fund, 13th salary, holidays), benefits with eligibility and assumptions (collective agreement, inflation, reference date).
- Dimensions and versions — Person, Position, Role, Level, Cost Centre, Entity, Time, Version; plus the HR ↔ Finance mapping between cost centre, GL account, and payroll line.
- Core — headcount and movement (active, hires, terminations, open positions, FTE per month), salary and charges with configurable parameters, benefits applied by eligibility, and projections for collective agreement and promotion.
- Consolidation and governance — personnel cost per cost centre, approved-plan lock, cell-level security.
- Outputs — HR input screens, headcount and cost reports, and direct consumption by FP&A as OPEX and P&L input, without manual export.
Notice that the first-axis decision shows up in all five layers. That is why it's expensive to undo: it isn't one more dimension, it's the shape of everything that comes after.
There isn't a single right answer
There is a conversation that has to happen before the first cube. And it belongs to the client, not to the consultant.
The implementer's job is to put the trade-offs on the table — including the uncomfortable one — and make sure the choice is a decision, not an inheritance from the template the consultant used on the previous project.
We've been modelling personnel cost inside integrated planning for over 15 years — see how we handle HR and workforce planning or get in touch to discuss your case.