AR chasing on autopilot — how Singapore SMEs collect 30 days faster
We replaced a finance assistant's Tuesday-morning chase ritual with an agent that knows when to nudge, when to escalate, and when to leave a customer alone. Here's what it actually looks like — and the three places it goes wrong if you let it run unsupervised.
Every SME in Singapore has the same Tuesday-morning ritual. Someone — usually the owner's most overworked finance person — opens Xero or QuickBooks, sorts the AR aging report by days overdue, and spends the next two hours sending the same five emails to the same five customers. By Wednesday they've forgotten who promised what. By Friday the chase has gone cold. The next week, they start over.
The numbers under that ritual are uglier than most owners realise. Across the SME books we manage, average days sales outstanding sits between 52 and 71 days against contract terms of 30. That's roughly six weeks of working capital frozen in your customers' bank accounts because nobody has the time, the patience, or the institutional memory to chase consistently.
We built the AR agent to replace that ritual — not the relationship. After running it across a dozen SMEs over the past year, we've cut average DSO by 28 days and freed up roughly six hours per week of finance-team time. Here's how it works, and where it doesn't.
What the agent actually does
The AR agent connects directly to your accounting system — Xero, QuickBooks, ABSS, whichever you use — and watches the AR aging report in real time. Every weekday morning at 8:30am it scans for invoices that have crossed a threshold and decides what to do with each one.
Decision rules are written in plain English on a page your team can edit. A typical setup looks like:
- At
+3 days, send a soft nudge from the finance inbox — friendly tone, PayNow QR attached, no urgency. - At
+10 days, escalate to a phone call task assigned to the account owner, with talking points pre-written. - At
+30 days, draft a formal demand letter and route it to the partner for sign-off before sending. - If the customer has paid late three quarters in a row, switch their default term from net-30 to net-15 on next contract renewal.
The agent doesn't just send emails. It reads replies. If a customer writes back saying "we'll pay by Friday," it parses the promise, sets a calendar follow-up for Saturday, and only re-engages if the payment hasn't arrived. If they raise a dispute — wrong line item, missing PO number, GST query — it pauses the chase, tags the invoice, and pings whoever owns that customer relationship. It will not argue with a customer. That part is still your job.
The hardest part of building this wasn't the AI. It was getting comfortable with what to do when the agent reads a reply that says "your service was rubbish, we're not paying."
Where it goes wrong
Three failure modes show up consistently. None of them are fatal, but ignoring them turns a useful agent into a liability.
1. The agent over-chases relationships you've already saved
A customer calls your sales lead, has a long conversation, agrees to pay next month. The agent doesn't know about the call. On day +10 it fires a stiff escalation email and undoes the goodwill. The fix is a single field in your CRM — a "chase paused until" date the agent reads before every send. Boring, mandatory.
2. The agent mistakes a quiet customer for a bad customer
Some clients in Singapore — especially older family-run businesses — simply don't reply to email. They pay, eventually, by cheque or PayNow, but they will never write back. If your rules treat "no reply" as a signal to escalate, you'll alienate good payers. We override the default behaviour after the first late payment with a manual flag, and the agent reverts to single-touch monthly nudges.
3. The agent drafts something legally awkward
We don't let it send a Letter of Demand without human review. Ever. A demand letter is a legal artefact in Singapore — wrong wording can complicate IRAS GST treatment of bad debt later, or muddle a small-claims case if it goes that way. The agent drafts; a human signs.
What "30 days faster" looks like in practice
A construction sub-contractor we work with — eight staff, ~SGD 4M revenue — was running 68 days DSO when they came to us. Three months in: 41 days. Six months: 38 days. The cash effect was about SGD 290k freed from receivables, which is what eventually paid for their new estimator hire. None of that money was new — it was already theirs. It was just stuck.
That's the honest pitch for AR automation. You aren't buying more revenue. You're buying back time and unsticking cash that was always yours. Whether it's worth the monthly fee depends entirely on how much of your AR is currently stuck and how much you value not having to do the Tuesday-morning chase ritual yourself.
If you want to know whether your AR is actually a candidate for this — some books are, some aren't — book a 30-minute call and we'll walk through your aging report together. No deck, no demo. Just numbers.