Most not-for-profit leaders do not think much about payroll until something goes wrong. The pay run is processed, Single Touch Payroll reports lodge, staff get paid, and the question of whether every hour was paid correctly under the right award sits well below funding acquittals and board papers on the list. Then an employee queries an overtime payment going back eighteen months, and a payroll review becomes the most urgent thing in the organisation. AI has changed what is possible in that moment, because it can read far more payroll data than a person can check by hand. What it has not changed is who decides what the data means.
Why payroll is harder than it looks
Payroll looks like arithmetic. Hours times rate, less tax, plus super. In practice it is the point where employment law, award conditions, enterprise agreements, organisational policy and software configuration all meet, and they do not always agree with each other.
A single shift can attract an ordinary rate, an overtime rate, a penalty or an allowance, depending on the time it started, how long it ran, what the employee had already worked that week and what their agreement says. Not-for-profits often run rosters that make this harder: part-time staff across multiple programmes, casuals covering shifts at short notice, people funded from more than one source, sleepovers and on-call arrangements in community and care settings. Each of those is a rule that has to be set up correctly once and then applied correctly thousands of times.
That is the part that catches organisations out. Payroll errors are rarely dramatic. They are small, systematic, and repeated every fortnight until someone looks closely. A penalty that applies from 6pm but was configured to start at 7pm does not announce itself. It produces a slightly wrong number, consistently, for everyone on that roster, until an employee asks a question or a new provider takes a proper look.
What AI is good at in a payroll review
Given a few thousand lines of exported timesheet data, AI can work through all of it. It can compare recorded start and finish times against what was paid, and it can spot shifts that look inconsistent with the pattern around them. It can then find the same anomaly repeating across a roster and group the lot into a shortlist a person can work through.
We have used AI to review exported timesheet data and flag potential payroll issues that would otherwise have required a significant manual review. That is the honest description of the benefit. The manual review still happens, but it starts from the shortlist instead of the whole dataset. The same applies to the slower parts of month-end, where we have used AI to process financial reports and rapidly prepare calculations such as accrued wages, reducing manual spreadsheet work.
For a finance manager who has been sampling a portion of each pay run and hoping the rest looks the same, that is a real change in coverage. Sampling finds errors by luck. Reviewing everything finds them by design.
Where a payroll review still needs a person
AI produces flags, and a flag is a question. It is not an answer, and the gap between the two is where the expertise lives.
When a review surfaces forty shifts where the paid rate does not match what the tool expected, each one has several possible explanations. The award might provide for that rate in those circumstances. The employee might be on an agreement with different terms. The roster might have been changed after the fact for a sound reason. Or the payroll system might be configured incorrectly, in which case the forty shifts in front of you are the visible edge of something much larger.
Telling those apart takes someone who understands the award, the organisation’s agreements and how the payroll and timekeeping systems have been set up. The consequences of getting it wrong run in both directions. Treat a correct payment as an error and you create work, confusion and an unnecessary conversation with a staff member. Miss a genuine underpayment and you leave an obligation sitting there, accruing against the organisation and unrecorded.
A caution belongs here too. Payroll obligations depend on individual circumstances. Responding properly to a flagged issue can involve employment law and industrial relations questions that sit outside accounting, and where that is the case, specialist advice is needed. A good adviser will say so rather than guess.
What a payroll review involves
A review that produces useful answers works through four layers, in order. Running them out of order is why some reviews find symptoms and never find causes.
Start with the configuration. Pay items, award mappings, penalty windows, allowance rules and leave accruals are where systematic errors originate. If overtime is set up incorrectly, every downstream figure inherits the problem, and checking individual payslips will only tell you that something is off.
Then test the data against the rules. This is where AI earns its place. Export the timesheet and payroll data, compare what was recorded against what was paid, and let the tool surface everything that does not reconcile.
Verify each flag against the entitlement. Someone who knows the award and the agreements works through the shortlist, confirms which flags are real and discards the rest. The output of this step is a much shorter list with a reason attached to each item.
Decide what the findings require. Corrections, back payments, a configuration fix, updated reporting, or specialist advice. That is a judgement call about obligations and risk, and it belongs with people.
The pattern underneath those four steps is simple to state. Automation expands what gets looked at, and expertise decides what it means.
The systems question sitting underneath
Hopscotch works across platforms including Deputy, Tanda and Employment Hero. These are capable systems, and most of the time they do exactly what they have been told to do. The quality of the output still depends heavily on how each system has been configured.
That sentence is easy to read past, so here is what it means in practice. A timekeeping system records hours. A payroll system turns hours into dollars using the rules someone entered when it was implemented, possibly years ago, possibly by a person who has since left. If those rules were set up with an incorrect assumption, the system will apply that assumption perfectly, every pay run, without complaint. No error message appears, because from the software’s point of view nothing has gone wrong.
When Hopscotch took over a client’s payroll function, it identified that overtime may not have been configured correctly in the client’s timekeeping or payroll system. That was one engagement rather than a typical result. The shape of it is familiar enough, though, that configuration is the first place we look.
Questions worth asking about your own payroll
You do not need to be a payroll expert to probe this. If you lead an organisation and want a sense of whether the process underneath your pay runs is sound, these are the questions to put to whoever handles it.
- When was the payroll configuration last reviewed against the awards and agreements that apply to our staff?
- How much of each pay run is checked, and how is the sample chosen?
- What happens when an anomaly is found, and is the cause investigated or just the symptom corrected?
- If an employee queried a payment from two years ago, could we reconstruct how it was calculated?
- Who decides when something needs specialist employment advice rather than an accounting answer?
The answers matter more than their detail. A process with clear owners and a real review cadence will produce confident answers. One that relies on the software being right will produce vague ones, and the vagueness is itself the finding.
Why this reaches the board
Employee costs are a substantial line in most not-for-profit budgets, so an error in how they are calculated does not stay in the payroll ledger. It flows into programme budgets, funding acquittals and the reports your board uses to make decisions. A systematic overtime issue is a liability that has not been recognised, and the board carries responsibility for recognising it.
That does not put payroll on every agenda. It does mean the board should know that payroll is reviewed properly, how often, and by whom. If your directors are asking sharper questions about financial information generally, it connects to the broader question of how often a board should review financial reports and what level of detail is useful at that table.
A payroll review also produces something boards value in its own right: evidence. Documented checks, a record of what was tested, and clear reasons for the decisions taken. That is governance material, and it is considerably more reassuring than an assurance that the system handles it.
One more thing about providers. We have found situations where previous providers may have processed information without sufficiently questioning whether the underlying setup or calculations were correct. Processing and reviewing are different jobs. If the arrangement you have only covers the first, the second is not happening anywhere, and the gap stays invisible until it does not.
Where AI fits in a well-run payroll function
Used well, AI shifts the economics of checking. Work that was too slow to do at scale becomes practical, which means more of your payroll data gets examined and anomalies surface earlier. For organisations with limited finance capacity, that is a real gain, and it shows up as fewer surprises rather than as a headline.
The review still needs a person who understands awards, overtime rules and system configuration to verify what has been flagged and decide what action is required. AI is a tool experienced accountants use and verify. The Fair Work Ombudsman’s guidance on pay and entitlements is a fair reminder of how much context sits behind a single hour of work, and software does not hold that context.
If you would rather not carry the question internally, that oversight is part of what an outsourced bookkeeping and payroll arrangement should include: someone reviewing the setup, not only running the numbers through it.
Frequently asked questions
No. AI can analyse payroll and timesheet data at scale and flag potential issues, but each flag needs to be verified by someone who understands the relevant awards, agreements and system configuration before any action is taken.
Whenever something changes: a new award rate, a new employment arrangement, a payroll system upgrade or a change of provider. Configuration errors repeat every pay run until someone finds them.
Processing applies the rules already in the system to this fortnight’s hours. Reviewing asks whether those rules are correct in the first place, which is a different question and needs different expertise.
Both, often. The calculation and reporting side is accounting work, while questions about entitlements, back payments and obligations to employees can require specialist employment or industrial relations advice.
Start with the setup, not the spreadsheet
If you take one practical step from this, make it the configuration review. Ask whoever runs your payroll when the pay items, award mappings and overtime rules were last checked against the awards and agreements that apply to your staff. If nobody can give you a date, that is where to begin, because every other check you run sits on top of those settings.
Hopscotch Accounting works with not-for-profit organisations across Australia on bookkeeping, payroll and reporting, including the kind of review that looks past the numbers to how the system producing them has been set up. We use AI where it makes the work faster and more thorough, and we verify what it finds, because that part cannot be automated.
If you would like someone to look at how your payroll is configured and what your current review process is covering, Start a conversation.
Liability limited by a Scheme approved under Professional Standards Legislation.
