If you sell online and to trade customers, money leaks in the gaps between your systems. A refund done in the online shop but never recorded in the accounts. A customer who paid twice and is still waiting. A return nobody chased. Each system looks fine on its own. The problem is that nothing joins them up except people’s memory.
We recently automated three of the jobs that catch those leaks inside a real UK equipment supplier selling online and to trade: the monthly debtors review, the refund tracker and the returns process. Here’s what we did, and the steps you can copy.
What we found when we checked properly
The first time the AI tied the debtors report back to the accounts system, it found 9 customers had dropped out of the hand-built analysis. Nobody knew, because the total had never been checked against the source.
It also found old refunds marked as “bank transfer” in the online shop that had never reached the bank refund run. Ticking “bank transfer” only meant someone had marked it, not that money had left the bank. Those customers were most likely still owed. And it found the opposite problem too: refunds already paid but still showing as open in the accounts, which is exactly how businesses end up paying the same customer twice.
On the refund tracker, comparing every shop refund and every credit note with the spreadsheet turned up more than 200 missing refunds, 16 wrong amounts and 36 duplicates. On returns, the first week of runs found 7 jobs that had already been done and skipped them, instead of doing them twice.
Step 1: write the process down
Before any AI, write down how the job works today: who does each step, which systems they open, what they look for and what they must never do. The AI follows that document, not memory. Every time you correct something, add it to the document the same day. In our returns process, that one habit meant the same mistake never happened twice.
Step 2: check against the source, not the spreadsheet
A spreadsheet that has only ever been checked against itself will always look right. The real gaps only show up when you compare it line by line with the systems the money actually moved through: your online shop and your accounts system. Start with your debtors. Does the report total match the debtors control account to the penny? If not, find out why before doing anything else.
You can start with an AI assistant such as Claude, Microsoft Copilot or ChatGPT and a couple of exports. Try this:
“Here is our aged debtors report as at [date] and our trial balance for the same date. Check the report total equals the debtors control account. If it doesn’t, find the difference: customers missing, documents posted after the report was run, or unposted batches. Don’t do anything else until we agree it ties.”
Step 3: sort every line by rule, then cross-check
Most lines on a debtors report need nothing: store credits that show in both systems, trade invoices still within terms, web orders paid today. Agree those rules with whoever owns the debtors, and let the AI sort every line into “needs action” or “no action”. Then, for each action item, check the online shop (was it sent, was it refunded, how was the refund paid?) and your bank refund sheet (paid, waiting for the next run, or never added?).
Finally, ask for a suggested fix for every line, grouped by type, so all the “raise a credit note” items can be done together. The most important group is the first one: refunds already done somewhere else. Record them, and do not pay again.
“For each web order on the action list, check the online shop: sent or not, refunded or not, and how the refund was paid. For any refund marked as a bank transfer, find it on our refund sheet by order number, name and amount, and say whether it’s paid, waiting, or missing. Read only: don’t change anything.”
Step 4: keep the people steps
In all three automations, a person signs in to each system, a person clicks Refund, and nothing in the accounts is changed until the finance lead approves it. The AI never types a password. It reads everything, checks everything and suggests the fix. That’s what keeps the risk low and makes the team comfortable trusting it.
Step 5: check before every action, and leave a note
Has this already been refunded? Has this job already been done? Did we email this approver today? Every run should check before it acts. Take a dated backup before every save, and leave a short carry-over note at the end of each run so the next one starts exactly where the last one stopped. On our refund tracker, someone saved an old copy over the updated file eight minutes after the AI finished. The backup is the only reason nothing was lost.
Your quick checklist
1. Does your debtors total tie to the control account to the penny?
2. Is every refund in your online shop on your refund tracker, once, at the right amount?
3. Are refunds marked “bank transfer” actually on your bank refund run?
4. Is anything open in the accounts that has already been refunded?
5. Has every customer who paid twice been refunded?
6. Does every return have a note saying who was asked and when?
7. Has the stock from every refunded return been booked back in?
8. Is there a backup before every change to your shared spreadsheets?
One safety point: use a business account for your AI assistant and check its data settings before sharing customer or financial details.
Build it yourself, or have us build it
If you want to build it yourself, our guide Get Paid and Stop Refund Leaks walks through all three automations: every rule, every check, the problems we hit, 21 copy-ready prompts and the full run procedures.
If you’d rather have it done for you, our Cash and Refunds Setup builds the debtors review, the refund tracker check and the returns run inside your business, connected to your own systems, with your team trained to run it. It’s £1,495 + VAT, fixed price, done in three weeks.
We’re not IT people. We’ve sat in the MD’s chair and built the systems behind it, and we’ll talk to you in plain business terms. Not sure where to start? Take the free Growing Pains Check.
Leave a Reply