Cash and till reconciliation: where AI pays back for NZ teams

Cash handling is the fastest place a small NZ business gets paid back for a bit of AI. The process is high-volume and money-sensitive, so a little more reliability compounds every week. And the assists that do the work are small: auto-categorisation in Xero or MYOB, an anomaly flag, a reminder that fires whether or not you remember, a weekly check that runs on its own. None of it is the "AI transformation" you read about.

This is the playbook I hand to small teams that still move real cash, juggle mixed payment types, or lean on one person for reconciliation.

Why cash handling breaks first when staff change

The pattern is consistent enough that I assume it before I ask. When a team has steady staff, cash handling is fine. The same person opens up, banks, and reconciles, and the rules live in their head. Then that person leaves, or goes on leave, or the shop gets busy enough that the rules get bent on a Saturday and nobody writes the bend down.

After that it drifts. A coin float stops getting topped up. A refund gets processed a new way. EFTPOS settlements stop matching daily takings cleanly. Say a few weeks pass, the numbers are off by a couple of hundred dollars, and there's no clean way to walk back through which day it came from.

This isn't really a cash problem. It's a process-resilience problem with cash as the medium. That's why I raise it before any AI talk: the fix order matters. You name the owner and the cadence first, then you add tooling. AI doesn't fix a process that has no owner. It just makes a worse process run faster.

Which assists actually pay back

By "small AI" I don't mean an agent that runs your accounts. I mean a few narrow capabilities that sit inside tools you already pay for, so they're easy to set up and easy to govern.

Auto-categorisation is the obvious first one. Both Xero and MYOB now suggest a category for each bank line and score how confident they are, Xero through bank rules and remembered coding, MYOB through Smart Reconciliation that matches on a schedule and learns from what you accept or reject. Let the high-confidence routine lines apply on their own, keep the rest as a suggestion you approve, and you stop clicking through the boring ones every week.

Anomaly flagging catches more than a quarterly review ever will. A weekly summary that highlights the outliers is enough: takings well down on the same weekday a month ago, a settlement that never landed, a supplier amount that jumped. Pick the threshold that means something for your shop.

Reminders tied to the cash process, not generic ones. "Bank drop not logged by 4pm Friday." "EFTPOS settlement variance over the amount you set." "Till float not topped up before open." The steps that fail silently otherwise.

First-pass reconciliation drafts. The tools match the obvious lines, settlements, recurring suppliers, regular customer payments, and flag the rest. You still approve, you just stop spending an hour on the easy ones.

None of it is glamorous, and that's the point. The teams that win aren't running anything experimental, just narrow assists where the process was already fragile.

A weekly rhythm that catches the timing problems

Here's the cadence I'd suggest if you're starting from scratch. It assumes a basic accounting tool and some kind of POS, nothing custom.

1. Daily, end of trade. Cash counted, bank drop logged in a shared sheet or your POS, with the name of whoever counted. Five minutes, owned by whoever closed.
2. Weekly, Monday morning. A 20-minute pass over the previous week. Bank lines reconciled with the auto-categorisation handling the routine matches, variances over your threshold flagged and investigated, the anomaly summary read.
3. Monthly, ahead of your GST return. A clean review with the bookkeeper or accountant. By now the noise is gone, and the conversation is about the real exceptions instead of the data quality.

The weekly pass is the one most teams skip. They jump from daily counting to a monthly catch-up, and that gap is where the timing problems hide: payroll funding shortfalls, supplier double-payments, missing settlements. Twenty minutes a week with a flag or two catches most of them while they're still small.

AI doesn't fix a process that has no owner. It just makes a worse process run faster.

What I'd leave alone for now

I get asked this often enough that it's worth saying what I'd not touch yet in a small team.

Cash forecasting. Tempting, and there are tools that promise it. But for a 2-to-15-person business with steady patterns, a spreadsheet built once with your accountant beats anything AI will generate this year, and the false confidence it can breed is a real risk.

Fraud detection. Most small businesses don't have the transaction volume for the patterns to mean anything. The fraud that hits small NZ teams is usually a person, not a pattern, and your weekly anomaly flag will catch it long before a model would.

Payment chasing via AI-written emails. Templated reminders, sure. But personal-sounding emails going out under your name that you didn't write are a trust risk and a relationship risk. Get one wrong in a small NZ town and it costs you. Stick to clear, factual reminders that look like reminders.

The rule I use: if an AI assist runs in the background and a person reviews the exceptions, that's fine. If an AI assist is generating something that goes out the door under your name without a human reading it first, that's where the wheels come off.

The bit most teams skip: a recovery plan

This part isn't about AI at all. It's what makes any cash process hold up, and most small teams don't have it written down.

When something breaks, a missed bank drop, a system outage, the bookkeeper sick on reconciliation day, a power cut at close, what's the fallback? Who steps in, and what's the manual process until normal service resumes?

I'd write a single page. Three scenarios, a couple of paragraphs each, pinned next to wherever cash gets counted, and walk a new hire through it on day one. It doesn't need to be impressive. It just needs "what do we do when X breaks" answered once, by the owner, rather than improvised at 4:55pm on a Friday by whoever's around.

This is what makes the small assists worth doing. They pull enough off the day-to-day load that there's room to follow the recovery plan when you need it, instead of being permanently behind on the basics.

Where to start this week

If your team handles cash and you've nodded along, the smallest useful step is three things. Write down who reconciles and when, one sentence, stuck somewhere visible. Turn on whatever auto-categorisation your accounting tool already has, set the confidence threshold conservatively, and watch it for two weeks before expanding. Then set one weekly anomaly flag on the metric you'd most want to know about, and have the tool email it to you Monday morning.

That's the whole first month. No platforms, no agents, no transformation, just the boring controls that stop the slow leaks.

If you want help picking the one or two automations worth doing and getting the cadence to stick, that's the kind of thing I do. The framework I use for picking the first automation covers the broader thinking, MYOB's AI features get their own rundown, and you can see how this fits a wider workflow setup.