How to conduct a wage compliance review

Many underpayments do not begin with an obvious payroll failure. They begin with a payroll that runs cleanly every fortnight, produces pay slips without error messages, and reconciles to the general ledger. What the system cannot tell anyone is whether the rules it applies are the right rules.

The usual causes are unglamorous. The wrong instrument was picked up when a business line was acquired, or a classification set in 2019 was never revisited as duties grew. Time data is incomplete — shifts approved in bulk, breaks assumed rather than recorded, start times rounded to the roster. Pay rules drift out of date after an annual wage review or award variation. Base rates and their satellites interact badly: a loading calculated on the wrong base, an overtime trigger that ignores hours outside the span, an allowance paid monthly where the award pays per occasion. Annualised arrangements are set up and never reconciled. Leave and final pay omit loadings the instrument requires.

None of that shows up in a payroll run. It shows up in a review designed to look for it.

A wage compliance review is a planned assurance exercise. It asks a question payroll processing never asks: for a defined population and period, does what we paid match what the governing instruments and the law required? Payroll executes rules; a review interrogates them.

It is also distinct from a complaint investigation, which is bounded by one employee's allegation and often resolved without testing whether the same fault affects others; a review starts from the fault, not the complainant. Nor is it a remediation program. Remediation may follow once findings are quantified and accepted; conflating the two produces rushed payments built on unverified numbers.

Scale is a matter of judgement; a small business testing one award against one roster pattern is running a review just as much as a national employer testing four agreements across eleven entities. What makes it a review is that scope, method and conclusions are set deliberately and written down.

Appoint a senior sponsor with authority to fund the work and act on its findings — typically a CFO, COO or general counsel, rarely the payroll manager, who may be reviewing their own configuration. Build a team with industrial interpretation, payroll systems knowledge, data capability, and someone who understands how the workforce is actually rostered and supervised.

Agree decision rights up front. Who resolves an ambiguous award clause? Who approves an assumption that moves the number by seven figures? Who decides a sample finding warrants full-population testing? Set confidentiality expectations and an escalation path to the board or audit committee.

Consider legal advice early where exposure may be substantial or interpretation is genuinely contestable, and be realistic about privilege. Involving a lawyer does not make everything privileged: whether privilege attaches depends on the purpose for which each document was created, and payroll data, system extracts and business records are generally not privileged simply because a lawyer received them.

Finally, preserve the material. Once a review is contemplated, stop routine deletion of time records, rosters, payroll extracts, system change logs and relevant correspondence, and confirm superseded payroll configuration can still be reconstructed for historical periods.

Each dimension of scope should be a decision rather than a default: legal entities; awards, agreements and contract populations; workforce cohorts, locations and employment types; pay elements; and periods.

Let the risk signals drive it. Pay complaints, even those resolved individually, point at a rule. So do payroll implementations and upgrades, acquisitions where a workforce arrived with its own arrangements, restructures that moved people between classifications, award variations implemented late or partially, cohorts with thin time records, and anomalies internal audit or the payroll team has already raised.

Sampling is legitimate and often necessary, but it needs a rationale: how the sample was drawn, why it represents the cohort, and what error rate would be treated as significant. Full-population testing is warranted where the population is small, where the calculation can be automated, or where a sample has already shown a systemic fault. A sample finding that traces to configuration or interpretation is a rule error: test every employee and period that rule touched.

Before anything is calculated, be clear about what governs each cohort. The National Employment Standards apply to national system employees as a floor. Above that sits a modern award, an enterprise agreement or, in limited cases, a workplace determination — and where an agreement applies, the award generally does not apply to that employee while it covers them, though the NES still does. Contracts and policies sit on top and can confer more generous entitlements, enforceable as contractual promises.

The distinction matters. A review that tests only against the award floor will miss a contractual promise of six weeks' notice or an above-award allowance the business has stopped paying; a review that tests only the contract will miss the award terms the contract never mentioned.

Coverage and classification deserve their own analysis, and we have set out that methodology separately in our guide to award interpretation and classification risks. In short, coverage must be established by reference to the instrument's coverage clause and the employer's actual industry or the employee's actual occupation, classification turns on duties performed rather than internal titles, and both need to be established for each historical period under review, not just for today.

One qualification on jurisdiction: this guide principally addresses national-system Fair Work obligations. Some employers and employees fall within a State workplace-relations system, and long service leave, payroll tax, unclaimed money and similar obligations may be governed by State or Territory law.

Expect to gather employee master data and employment histories; contracts and variations; classification records; rosters; actual start, finish and break records; leave records; payroll transaction data at line level; pay slips; superannuation contributions and payment dates; allowances and deductions; individual flexibility arrangements; annualised wage agreements and any reconciliations; termination calculations; and manual or off-cycle payments.

Then validate it. Completeness first: reconcile headcount across systems, look for employees with no time records in periods they were paid, and check period boundaries align. Then meaning. Field names are unreliable guides to content — an "ordinary hours" field may include paid breaks in one entity and exclude them in another, and a "base rate" may be loaded. Trace several employees end to end, from roster to timekeeping to pay slip. Where the data cannot support a conclusion, record the limitation rather than modelling around it.

Under the Fair Work Act and its regulations, employers must make and keep prescribed time and wages records. The Fair Work Ombudsman's record-keeping guidance summarises the core obligations: records must generally be kept for seven years, be readily accessible to an inspector, be legible and in English, and must not be altered unless the alteration corrects an error. It is unlawful to make or keep records the employer knows to be false or misleading.

Two consequences follow. Gaps are not merely an evidentiary inconvenience — failing to make or keep required records can itself be a contravention, separate from any underpayment. And in proceedings alleging a contravention, where an employer has not complied with record-keeping or pay slip obligations and cannot give a reasonable excuse, the Fair Work Act can shift the burden onto the employer to disprove the allegation about the matters those records should have covered.

Keep the seven-year retention rule separate from the limitation period for recovery. Section 545(5) generally limits orders for contraventions to those occurring within six years before the application is made. Confusing the two produces look-back periods either too short to be safe or longer than the evidence can support.

The model calculates what should have been paid, and its central requirement is historical accuracy: each pay period must be tested against the instrument version and rates in force at the time. Annual wage review increases, award variations, and agreement replacement or expiry all create effective-date boundaries, and a model that applies today's rates backwards will be wrong in both directions.

Within each period the model must handle age-based and apprentice or trainee progression where it applies, ordinary hours and the span in which they may be worked, overtime triggers, penalty rates, casual and shift loadings, allowances, break entitlements and the consequences of missed breaks, higher duties, leave and leave loading, public holidays, permitted deductions, and termination payments including accrued leave and notice.

Where the instrument is genuinely ambiguous — and some clauses are — take a position, document the reasoning and the alternative reading, and quantify the difference. Assumptions should be listed in one schedule with their basis and their financial sensitivity. Reviews that bury assumptions inside formulas cannot be audited or defended later.

Salaried arrangements are a recurring source of exposure because two different mechanisms are often treated as one. An award annualised wage arrangement is made under a specific award clause and carries that clause's conditions. A common-law set-off arrangement relies on a contractual term applying a higher salary against identified award entitlements, and its effectiveness depends on the drafting and on the relationship between the payment and the entitlement it is said to satisfy.

Where an award annualised wage clause is used, the Fair Work Ombudsman's guidance on annualised salaries is a useful orientation, but the operative terms differ between awards. Depending on the clause, the employer may need to record the arrangement in writing, specify the award provisions satisfied and the outer limits of penalty or overtime hours covered, record start and finish times and unpaid breaks with the employee confirming them, and reconcile against actual entitlements annually and on termination — paying any shortfall.

Two errors are worth naming. The first is assuming every award contains the same annualised wage provisions; they vary, and some awards have none. The second is treating a salary as extinguishing award entitlements by itself. It does not: if the conditions are not met, or the salary does not in fact cover what was worked, the underlying entitlements remain payable.

Reconciliation compares modelled entitlement against amounts actually paid, at the level of the individual entitlement and the individual pay period — aggregation hides exactly the errors a review exists to find.

Resist casual offsetting. An overpayment of one entitlement in one period does not automatically cancel an underpayment of a different entitlement in another. Whether set-off is available depends on the instrument, the contract and the basis of the payment; it is a question to be advised on, not assumed by whoever built the model.

Take a salaried employee whose annual salary comfortably exceeds the total award amounts they would have earned across the year; read annually, the arrangement looks generous. Now look at a fortnight in which they worked two late shifts, a Sunday and eleven hours of overtime covering a vacancy. In that fortnight the penalties, overtime, shift loading and allowance exceeded the salary paid for the period. Whether the annual surplus can answer a pay-period or entitlement shortfall depends on the applicable award clause and on the contractual or set-off arrangement relied on; it should not be assumed.

Model output should be tested by someone who did not build it. Re-perform a selection of calculations by hand against the instrument. Test the boundaries — the first shift after a rate change, a public holiday on a rostered day off, a mid-period termination. Check rounding and the treatment of missing values, because a null read as a zero silently suppresses entitlements. Interrogate the exceptions list; the records that would not process often hold the largest errors.

Where a sample was used, say plainly what the extrapolation does and does not support. A range with stated assumptions is more useful to a decision-maker than a single figure with hidden ones.

Separate isolated transactions from patterns. A keying error affecting one employee once is a different problem from a payroll rule that mis-set an overtime threshold for four years across a cohort. Group findings by cause: configuration, interpretation, data capture, management practice such as unapproved off-the-clock work, and governance failures such as award variations with no owner.

Quantify each finding by employees affected, periods covered and entitlement categories. Address the whole affected population, not only those still employed — former employees may form a substantial part of the affected population and are easily overlooked — and consider adjacent cohorts governed by the same rule at other sites or entities.

Where a finding suggests that people with financial or advisory responsibility knew of a problem and let it continue, that is a governance escalation, and the involvement of managers, advisers and payroll providers carries its own legal significance. See our article on accessorial liability for advisers and payroll providers.

Verify the calculations again before paying. The Fair Work Ombudsman's guide to fixing an underpayment sets out the practical sequence: work out how long the employee was underpaid, calculate the shortfall, back-pay them, and keep records of what was done.

Back payments carry tax, superannuation and payroll tax consequences turning on the nature and timing of the payment; those belong with qualified tax and superannuation advisers. Superannuation on remediated ordinary time earnings, interest where it applies, and PAYG treatment and reporting should be settled before payment rather than corrected afterwards.

Communicate in plain language: what happened, which periods and entitlements were affected, how the amount was calculated, when it will be paid, and who to ask. Pay promptly, and keep the calculation, approval, communication and payment evidence together for each employee. For former employees, record genuine attempts at contact and take advice on unclaimed money obligations. Correct current pay prospectively at the same time.

Back payment does not automatically end the matter. It is relevant to how a regulator responds, but it does not by itself prevent enforcement action or remove an employee's ability to pursue a claim.

Whether to report to the Fair Work Ombudsman turns on scale, duration, cause and conduct. The regulator indicates it is best to report where an underpayment has been occurring over a long period or is extensive in quantity, and its compliance and enforcement policy explains how self-reported non-compliance is treated. Neither reporting everything by policy nor a reflex against reporting serves an employer well; decide on the facts, with advice, and record the reasoning.

Since 1 January 2025, intentional underpayment of amounts payable under the Fair Work Act, a modern award or an enterprise agreement can be a criminal offence. The Fair Work Ombudsman's guidance on criminalising wage underpayments is explicit that honest mistakes are not captured; the offence targets intentional conduct. The distinction between a configuration error and a decision to withhold matters, which is one reason contemporaneous records of what was known, and when, are worth keeping. Small business employers should look at the Voluntary Small Business Wage Compliance Code: where the regulator is satisfied a small business employer has complied with it, that employer cannot be referred for criminal prosecution.

Where issues are large, systemic, long-running or capable of being characterised as deliberate, take tailored legal advice before deciding what to disclose, to whom, and in what sequence.

A review that ends with a payment run has done half the job. Correct the award and rate libraries. Rebuild the payroll rules that were wrong and test them against known cases before go-live. Fix time capture where breaks or actual start times were not recorded, and tighten approvals where shifts were waved through in bulk. Build exception reports for the specific patterns the review found, so the fault is caught by the system rather than by the next review.

Give the recurring obligations an owner by name: implementing the annual wage review outcome, monitoring variations to the awards and agreements that apply, reviewing classifications when duties change, and reconciling annualised wage arrangements where a clause requires it. Then test again — a control configured but never verified is an assumption.

As for frequency, there is no single legally mandated interval and an annual review is not automatically sufficient. Set the rhythm by risk and by change: a payroll migration, an acquisition, an agreement replacement, a significant award variation or a cluster of pay complaints each justify a review of their own.

Australian Workplace Strategies designs and runs wage compliance reviews for employers who need an independent view of their exposure and a method that will withstand scrutiny. We scope to the risks that apply, build entitlement models on period-correct instruments, test payroll configuration against what the rules require, and set out findings, quantification and remediation options a board can act on. Our workplace advisory and compliance team can be engaged for a single review or ongoing assurance.

Where an organisation wants the obligations, controls, evidence and remediation actions from a review kept in one auditable place across years, Strobe is available for that purpose.