How long do I have to keep employee records?

Seven years for time and wages records under the Fair Work Act. Former employees can ask for copies of their records after they leave, so the clock matters more for people who have gone than for people still on the payroll. Check whether your award, tax or super obligations add anything for your business.

Can my timesheets just be generated from the roster?

Not safely. The roster is a plan and the record has to show what happened. Where an employee's pay depends on time worked, such as a casual or someone paid overtime penalties, a timesheet copied from the roster can end up recording hours that were never worked or missing hours that were. Use the roster as the starting point and have the actual start and finish times confirmed.

Is it a problem if we change a record after the fact?

Records can only be altered to correct an error. Correcting a genuine mistake is fine, and keeping a note of what changed, who changed it and why is what makes the correction defensible later. A system that overwrites the old value with no history leaves you unable to show the change was a correction.

Records that survive a wage claim

What does Fair Work actually require you to record?

More than most payroll exports capture by default. The Fair Work Ombudsman's record-keeping guidance groups it like this:

General details. Employer name and ABN, employee name, start date, and whether they are full-time, part-time or casual, permanent or temporary.
Pay. The rate paid, gross and net amounts, every deduction, and any loading, penalty rate, allowance, bonus or incentive paid.
Hours. Hours worked for casuals and irregular part-timers paid by time. For other employees, overtime hours each day or the start and finish of overtime where a penalty or loading applies. A copy of any written agreement to average hours.
Leave. Leave taken and the running balance, plus copies of any agreement to cash out leave or take it in advance.
Super. Amounts, dates, the period covered, the fund, and the basis for paying into it, including the employee's fund choice and when they made it.
Agreements and endings. Individual flexibility arrangements and any notice ending them. How employment ended, whether notice was given and how much, and who terminated it.

Pay slips sit beside all of this. They go to each employee within one working day of payday, in hard copy or electronically, and carry their own list of required details.

Why do small businesses lose wage claims they could have won?

Usually because the business paid correctly and can't prove it. The money went out, the employee was happy at the time, and three years later nobody can produce the timesheet that shows the overtime was worked on a Saturday morning and paid at the right rate.

The gaps tend to be predictable:

Hours that were never recorded, because the employee was on a set weekly figure and the overtime "evened out".
Agreements that lived in a text message, such as a cash-out of leave or a change to hours that was agreed and never written down properly.
Records that left with a system, when the business changed payroll software and the old subscription lapsed.
Records that left with a person, when the bookkeeper who kept the spreadsheets moved on.
Corrections with no trail, where a figure was fixed months later and nobody can show what it was before or why it changed.

None of these involve underpaying anyone. They involve being unable to show you didn't.

Where does automation help?

At the point of capture, and in the dull work of keeping things where they belong. A few places it earns its keep:

Start and finish times recorded when they happen, by the employee, rather than reconstructed at the end of the week.
A daily or weekly comparison of actual hours against the roster, with differences sent to a manager to approve or query. The system finds the gap. A person decides what it means.
Pay slip checks, confirming every employee paid in a run was sent a slip within the working day.
Agreements filed against the employee, so a leave cash-out or flexibility arrangement is attached to the person's record, not sitting in someone's inbox.
A yearly archive export, in a format you can open without the original software, stored somewhere you control.

That last one is the cheapest insurance in this article. It takes an hour a year and covers the most common way records disappear.

Where does automation quietly destroy the record?

When it makes the record tidier than what happened. This is the side most owners don't see, because the output looks better than before.

Timesheets filled from the roster. It saves everyone typing, and it turns the record into a copy of the plan. When someone stays forty minutes late and the timesheet shows the rostered finish, the record is now wrong, and it was wrong the moment it was written.

Rounding. Clock-in rules that round to the nearest quarter hour can look harmless. If the rounding consistently runs against the employee, it adds up across a team and a few years, and the record will show it.

Syncs that overwrite. Two systems passing data between them can update a figure in place with no history. If a rate or an hours figure changes, you need to see the old value, the new one, who changed it and why. Without that, a legitimate correction looks the same as an alteration.

Short retention settings. Some tools default to deleting or archiving old data after a set period, and plenty of businesses never check. Seven years outlasts most software choices.

AI that cleans up records. Using AI to summarise a week's hours for a manager is fine. Letting it rewrite, merge or tidy the underlying record is not. AI is good at producing something plausible, and a plausible timesheet is the wrong goal. The limits of what AI should be trusted with apply here more than almost anywhere: the source record stays as it was captured, and a person approves any change to it.

How do you check whether your records would hold up?

Run a test on someone who has left. Pick a former employee from three or four years ago, ideally a casual or someone who worked overtime, and try to produce their full record in a day. Every hour worked, every rate paid, every leave balance, their super fund choice and how their employment ended.

Then do the arithmetic on your exposure. Count how many people have worked for you in the last seven years, including casuals who stayed a few weeks. Every one of them is a record you are expected to be able to produce. Most owners guess a number well below the real one, because the short-stay casuals are the ones nobody remembers.

If the test took a day and everything turned up, you are in good shape. If it took a week, or something was missing, you now know which of the five gaps above you have.

What to do next

Run the former-employee test this week and write down what was hard to find.
Check whether your timesheets record actual hours or copy the roster, and change it if they copy.
Look at the rounding and retention settings in your time and payroll software.
Export a full archive from your current payroll system and from any system you have stopped paying for, if you still can.
Decide who approves corrections and make sure each one carries a reason.

If you want a second set of eyes on which parts of your payroll and admin are worth automating and which should stay manual, book a free fit chat. If changing payroll systems is on the cards, the payday super changes are worth reading first, because they affect the same records.