01 the business
Commercial cleaning company
220 staff, 350 business clients
Chasing payment happens in bursts when cash gets tight. Everyone gets the same template. Reliable clients feel nagged, while the accounts that need a phone call get the same polite email.
02 before
The problem
Chasing payment is the job nobody wants, so it happens in bursts when cash gets tight. Every late customer gets the same template, good customers feel nagged, and the accounts that really need a call get the same polite email as everyone else.
- Reminders go out irregularly.
- Every late payer gets the same message.
- Disputes are chased by mistake.
03 the build
What the AI does
- Step 1: Reads open invoices and payment history from Xero, QuickBooks or NetSuite every day.
- Step 2: Groups overdue accounts by risk using days overdue, amount, past payment habits and any open disputes noted in the CRM.
- Step 3: Drafts a reminder for each account that matches its group, from a light nudge for a reliable payer to a firm note for a repeat late payer.
- Step 4: Attaches the invoice and a payment link from Stripe or GoCardless so the customer can pay straight away.
- Step 5: Logs every reminder and reply against the customer record and suggests who should get a phone call this week.
paused: waiting for a person to approve
How it was put together
- Read open invoices and payment history daily.
- Grouped overdue accounts by days overdue, amount, past habits and disputes noted in the CRM.
- Drafted reminders that fit each group, with the invoice and a payment link attached.
- Logged every reminder and reply, and suggested who needs a call this week.
04 human in the loop
Where a person stays in charge
Your finance lead or account owner reviews the drafted reminders in a daily batch and approves, edits or holds each one before it is sent. Any account marked as in dispute or on a payment plan is held for a person to handle.
- The finance lead approves, edits or holds each reminder in a daily batch.
- Accounts in dispute or on a payment plan are held for a person.
- Tone stays polite and factual, with no threats.
05 rollout
From kickoff to live in 3 weeks
- Week 1Agree risk groups, message tone and hold rules.
- Week 2Draft reminders for the current overdue list and review together.
- Week 3Start the daily approve-and-send routine.
06 after
What changes
Follow-up happens every week instead of whenever someone finds time. Good customers get a lighter touch, and the team puts its phone time into the accounts that matter most.
Assume $900,000 of monthly billing, 45 days sales outstanding, and steady reminders bringing it down to 38 days.
$900,000 ÷ 30 days = $30,000 billed a day. Collecting 7 days sooner frees 7 × $30,000 = $210,000 of cash.
What to measure
- Days sales outstanding.
- Invoices over 60 days overdue.
- Time spent chasing each week.
- Customer complaints about reminders.
07 evidence
What the research says
About 75% of generative AI's potential value sits in four areas: customer operations, marketing and sales, software engineering, and R&D.
Gartner research put the average cost of poor data quality at $12.9 million a year per organisation.
Gartner, Data quality: why it matters and how to achieve it (opens in a new tab) · 2020
08 lessons
Mistakes to avoid
- Automating without holds. Disputes need people.
- Same message for everyone. Tone should match the relationship.
- No payment link. Make paying the easiest step.
09 questions
Common questions
- Will customers find it pushy?
- Good payers get light nudges or none. Firmer messages are kept for repeat late payers, and every message is approved.
- Can it take card or direct debit payments?
- It adds a Stripe or GoCardless link to each reminder.