AI Accounting for Retail

It handles the consistency problem, which is most of the real-world damage, but it is not a substitute for a considered view on the hard categories. Australian GST treatment of food is famously fiddly: basic food is generally GST-free, but prepared food, confectionery, savoury snacks and most bakery products are taxable, and the line between them produces genuinely difficult calls. A plain bread roll and the same roll with a filling are not treated the same way. What goes wrong in practice is rarely that a retailer does not know the rule — it is that the same product gets treated one way in March and another way in September because two different people coded it. The AI applies your determined treatment consistently across every store and every period, and flags new products it has not seen before rather than assuming. The determination itself should come from your accountant or BAS agent.

Because they are measuring different things, and the gap is structural rather than a mistake. Your POS records gross sales on the day of sale. Your bank records a net deposit some days later, after the payment processor has taken its merchant fee, batched multiple days together, or held funds over a weekend. Buy-now-pay-later settles on its own schedule with its own fee. Refunds net against the deposit rather than appearing separately. So a deposit of $4,812.37 on Tuesday might represent three days of gross sales, minus fees, minus two refunds, from two different processors. Reconciling that manually is a job nobody has time to do daily, so it gets done monthly, badly, or not at all. The AI does the unpicking every day and surfaces only the genuine variance.

It is one of the most common and most damaging shortcuts in retail bookkeeping, and it is worth being blunt about. Booking only the net deposit understates your turnover by the amount of the fees, which distorts every metric built on revenue and can matter for thresholds and reporting. It also means the merchant fees never appear as an expense — so your cost base is understated too, and the GST on those fees is never claimed. The business ends up with a profit figure that is roughly right by accident, built from two offsetting errors, and a revenue figure that is simply wrong. Automated settlement reconciliation books the gross sale and the fee separately because that is what actually happened.

Both, and the distinction matters more than most retailers expect. Each store is reconciled on its own — its own takings, its own banking, its own float, its own variances — because a discrepancy at one store is a store-level issue and averaging it across the group is how it stays hidden for eight months. Reporting then rolls up, so you get a group position without losing the store-level detail underneath it. The pattern worth knowing about: takings variances are rarely random. When one store shows a consistent small variance in one direction, that is a signal, and it is one that only shows up if you were reconciling per store rather than per group.

It is the normal case for retail and hospitality, and it is exactly the shape of work automation suits — high volume, low complexity, high tedium. The AI reads invoices from email, PDF and phone photos, extracts the detail, matches against the supplier and against deliveries where you record them, and codes to your chart. The two things it does that a human at volume genuinely cannot: catch duplicates reliably across a month of near-identical invoices from the same supplier, and notice that a supplier’s price for a line item has moved. The second one is not an accounting function at all — it is margin protection — but it falls out of the same processing for free.

This is where the capacity argument for automation is strongest in retail, because the December problem is not a December problem. Your transaction volume might triple in the lead-up to Christmas while your bookkeeping capacity stays exactly the same, so the backlog builds through the peak and gets cleaned up in February — which means you are flying blind on margin during the single period where it matters most. Automated processing scales with volume without a hiring conversation, so the ledger stays current through the peak. You get to see what is actually happening to margin in December, in December.