Organising compensation without losing its meaning

1. Define the question before selecting the amount

A useful compensation architecture separates what the employee is promised, what was paid or vested, and what employment costs the employer. These are different measurements. A recruitment offer, payroll reconciliation, market comparison and equal-pay review should not silently share the same number.

Official concepts demonstrate the problem. ILO earnings encompass gross remuneration in cash and kind but exclude employer social-security/pension contributions and severance. A broader employer-cost measure therefore has a different boundary. Eurostat's Structure of Earnings Survey distinguishes regular monthly earnings from annual earnings that can include irregular payments. Neither definition makes a survey's generic “salary” equivalent to this project's base salary. ILO definitions, Eurostat SES metadata.

For every measure, write a short measurement contract: eligible population. Included components. Gross or net basis. Cash/noncash boundary. Observation period. Working-time normalisation. Currency. Valuation method. And treatment of missing values. Give the contract an owner and version. A column name alone cannot carry all of that meaning.

2. A practical component dictionary

The following classification is our recommended data policy. Local contracts, collective agreements and survey definitions may require another boundary. Record the exception instead of relabelling it away.

ONS separately identifies basic, overtime, incentive, shift-premium and other pay in its earnings survey. That is useful corroboration for keeping components distinct, but it is a survey classification rather than a universal employer schema. ONS ASHE interpretation guide, pay components.

Total rewards is broader than a monetary pay total. Professional terminology includes benefits, wellbeing, career opportunities and recognition. These can belong in the employee proposition without inventing a euro value that is added to salary. This distinction is a professional taxonomy, not an empirical finding about what motivates every employee. WorldatWork model.

3. Five levels, eight useful menu choices

The normal dropdown now uses this ladder. It is deliberately a project convention. “total compensation” and “total remuneration” vary across countries and survey providers.

The last two levels are not total employment cost. They exclude employer social contributions, payroll taxes, recruitment, training, business facilities and termination costs. A statutory reporting definition may include items excluded from a particular market measure. Produce separate outputs for those separate purposes.

In production, current base must not be used as a shortcut for last year's fixed cash if salary changed during the year. The simplified table accepts prepared totals. Historical payroll detail remains upstream. It need not be squeezed into the visual. The synthetic demo assumes fixed pay and benefits were unchanged throughout its actual year so its displayed components reconcile directly.

Variable-pay subtotals are analytical conveniences, not extra components. Here VariableActual = StiActual + CommissionActual and VariableTarget = StiTarget + CommissionTarget. They exclude LTI and overtime. Another organisation may legitimately define variable compensation more broadly. Change the label and documented mapping together.

4. Actual, target, earned, granted, vested and realised

These terms answer different questions:

For equity, do not add the grant, vesting value and later sale proceeds of the same award. They are alternative observations of its lifecycle. Our demo uses annual target award value for target LTI and vesting-date value for actual LTI. This is a compensation-analysis convention, not an accounting treatment. IFRS 2 distinguishes equity-settled and cash-settled measurement. Its accounting valuation is not automatically the employee's realised proceeds. IFRS 2 standard, especially §§10–15 and 30.

For multi-year grants, do not universally divide by vesting years. An annual grant that vests over three years and a once-in-three-years award have different annual opportunity patterns. Select a convention consistent with the benchmark, document overlap of outstanding awards, and retain award-level audit records upstream. Do not treat an option's number of shares multiplied by share price as option fair value.

Target percentage needs a denominator: 15% of base is different from 15% of fixed cash. “On-target earnings” (OTE) often communicates base plus target sales incentive. Never assume OTE includes LTI, pension or every allowance. Ask the provider for the included items.

5. Normalise only what can defensibly be normalised

For ordinary salary at one reference date, an explicit example is:

Annual FTE base = contractual part-time monthly base × contractual annual salary instalments / contractual FTE fraction.

Thus €3,000 per month × 12 / 0.75 = €48,000. If a guaranteed 13th salary is included, the corresponding figure is €52,000. This is a constructed illustration, not a rule that every bonus should be annualised or that every German employee has 13 instalments.

For a year with salary or working-time changes, use effective-dated segments. Build the ordinary-pay rate for each segment and weight it by the relevant fraction of the year. Reconcile actual paid cash separately. Dividing a whole year's payments by the FTE fraction at year end can distort the result. A new joiner's signing award, overtime, a fixed benefit or a pension cap may not scale linearly with FTE. Mechanically dividing everything by 0.5 can invent an entitlement.

Three outputs should therefore remain distinguishable in the source preparation: actual cash received, a comparable annual/FTE analytical amount, and employer cost. Keep transformations traceable from payroll codes to these outputs. Where an annual/FTE equivalent is not meaningful, leave that analytic field missing and retain the raw amount upstream or as a clearly labelled tooltip field. The present visual expects annual FTE EUR. It does not calculate working-time conversions or foreign exchange.

A gross hourly measure needs an explicit hours denominator. Hours paid, contractual hours and hours actually worked differ. Paid leave and overtime treatment matter. Eurostat's SES hourly definition uses gross reference-month earnings divided by paid hours, with its prescribed boundaries. An annual FTE chart is not an equivalent statutory hourly-pay calculation. SES 2022 metadata.

6. Reconcile components before aggregating employees

Constructed example, EUR, full-year FTE, constant fixed pay:

Fixed cash is €63,000. Actual cash is €73,500. Target cash €75,000. Actual direct is €78,500. Target direct €81,000. Actual remuneration is €83,500. Target remuneration €86,000. The €9,000 actual variable subtotal is already represented by STI and commission: adding it again would overstate actual cash by €9,000.

Our recommended validation order is identity → dates/currency/working time → atomic amounts → totals → employee population → statistics. Round source monetary components to the required accounting precision before reconciliation and specify a tolerance. Reconciliation needs a complete component boundary: a mismatch may reflect an omitted payroll item rather than a wrong total. Never silently overwrite an authoritative total to make a check pass.

Blank means not supplied/unknown. Zero means a known zero. A null bonus may reflect unknown eligibility or incomplete extraction. Adding nulls as zero creates false completeness. Genuine negative payroll corrections also require an explicit upstream netting policy: the current visual excludes negative analytic pay amounts, so it is not a payroll transaction ledger.

For each person, construct the desired total first. Then compute distribution statistics across persons. In general median(base + bonus) is not median(base) + median(bonus). Example: three employees have base [40, 50, 100] and bonuses [60, 0, 0], in thousands. Median base + median bonus = 50. Median of total cash [100, 50, 100] = 100. The paired employee data determine the answer.

For an employee-weighted average across disjoint groups, combine their totals and valid counts. Do not take the unweighted average of department averages unless each department is deliberately meant to have equal weight. Pooled medians and percentiles need the underlying observations or a suitably documented estimation method. Subgroup medians are insufficient. Headcount, sum of actual payroll and sum of annual FTE amounts are also different quantities.

The visual's internal percentiles use linear interpolation: for sorted values, position (n−1)p in zero-based indexing, interpolating between neighbouring observations. This is the type-7 convention documented by R. It is not a claim that a survey provider uses the same method. R quantile documentation.

7. Organise jobs and groups independently of pay components

Use employee identity as the row grain and keep company, location, department, job family, job, position, band, grade and step as separate attributes. A job describes a type of work. A position is an organisational seat. A person occupies it. A grade is a job-value category. A band is a pay-structure grouping. A step is a progression point. A particular employer may merge grade and band, but that is a local design decision.

Use stable keys. Labels can change. Scope reused keys, for example a grade code that means different things in two countries. A single employee row cannot simultaneously represent several assignments without a policy. Choose the principal assignment or build a documented consolidated employee record upstream. Duplicating employees by cost centre changes medians and headcounts. The current visual intentionally rejects duplicate employee identities in one snapshot.

Internal job evaluation should consider the work rather than the incumbent's current salary or negotiating power. A structured, gender-neutral evaluation can support equal-value comparisons across differently named jobs. Market rates alone cannot establish equal value. ILO, Promoting equity: gender-neutral job evaluation.

The grouping dropdown changes the population being summarised, not the definition of pay. Company, location and job analysis therefore remain available for the eight headline metrics. Market comparisons need their own valid job/grade matching. A company average is not automatically a market benchmark.

8. What incentive research can and cannot justify

Lazear's Safelite study reports increased output following a switch to performance pay and distinguishes incentive and workforce-selection effects. It is evidence from a particular setting, not proof that every job needs a bonus or that its effect size transfers to professional services. Lazear (2000), AER 90(5), 1346–1361.

Holmström and Milgrom's multitask principal-agent model explains why rewarding measurable output can divert effort from other valuable tasks. It provides a theoretical reason to consider balanced measures or less intensive incentives, not a formula for an optimal universal STI percentage. Holmström & Milgrom (1991), JLEO 7, 24–52.

A defensible practitioner decision consequently records the business objective, employee influence over results, measurement quality, time horizon, risk, affordability and potential unintended behaviour. Distinguish that judgement from descriptive salary data. A pay architecture should make decisions explainable. It cannot remove the need to make them.

9. Operating policy to approve before production use

Approve a pay-code dictionary and total definitions, period rules, FTE policy, LTI and benefits valuation, job architecture, market matching, exclusion rules and access permissions. Give payroll ownership of payment reconciliation, reward ownership of definitions and matching, and the analyst ownership of reproducible transformations. Record who reviews exceptions. A material change to included pay, reference population or normalisation deserves a new methodology version even if the visual's software is unchanged.

Before a compensation committee sees a chart, provide the measure contract, valid employee count, missingness by group, dates, reconciliation results and a worked employee example. Preserve the source extract and transformation version in an appropriately restricted location. The published demo contains no real employee or survey data.