Overdue invoices grow when follow-up depends on someone remembering. And when the message finally goes out, it is late, generic or written in a tone that strains the relationship with a good customer.
Who it is for
Owners and managers of small businesses that sell on credit or in installments (services, distribution, schools, clinics, agencies) and today collect with a spreadsheet and hand-written WhatsApp messages.
AI does not collect: it prepares. It detects what is due, drafts a message with that customer's context and logs everything. A person decides on sensitive cases. You can start today with a spreadsheet and an AI chat; once the flow works by hand for a week or two, then it makes sense to automate it with a tool like n8n.
The protocol, step by step
-
Build the receivables table
One sheet with one row per invoice or installment. Without it there is no system: it is the memory that today lives in someone's head.
Copy and pasteColumns: customer · contact · amount · due date · days overdue · stage (upcoming / overdue / promised / escalated) · previous promise (date and amount) · last contact · next action · owner
-
Define the segments
Segment by history, amount, stage and previous promise. A customer who always pays and is three days late is not the same as one who has already broken two promises. Write each segment's rule in one line.
Copy and pasteA · Friendly reminder: due in 3 days or up to 7 days late, good history. B · Firm follow-up: 8 to 30 days late or one broken promise. C · Sensitive case (goes to a person): high amount for your business, strategic customer, open complaint or two broken promises.
-
Write your business voice
Before asking AI for anything, write down how you talk. This is what keeps it from sounding like a collector.
Copy and pasteWe speak informally, warm and direct. We never threaten or mention legal consequences. We always offer an easy way out: payment link, transfer details or the option to propose a date. We sign with a person's name, not "Collections Department".
-
Generate the message with context
Every morning filter the rows whose next action is today and paste each one into your AI chat with this prompt. The message comes out different for each customer because it uses their history.
Copy and pasteYou are the collections assistant for [BUSINESS]. Voice: [PASTE YOUR VOICE FROM STEP 3]. Customer data: [PASTE THE ROW]. Segment: [A/B]. Write a WhatsApp message of at most 4 lines that: 1) greets them by name, 2) mentions the amount and the exact date, 3) if there was a previous promise, mentions it without blame, 4) ends with one clear action (pay at [LINK] or reply with a date). Do not use words like "debt", "delinquent" or "final notice".
-
Escalate before sending
Every segment C case goes to a person before anything is sent. AI can leave a draft, but someone with a name approves the send. Same if the customer replied with anger, a complaint or a personal situation.
-
Log the reply, commitment and next action
When the customer replies, paste their reply into the chat and ask it to update the row. That way nobody loses a payment promise between conversations.
Copy and pasteCustomer reply: [PASTE]. Return only: commitment (date and amount, or "none"), new stage, next action with date, and whether it should escalate to a person (yes/no and why).
-
Review once a week
Ten minutes every Monday: how many promises were kept, which customers moved to C and which message got the best response. Adjust the voice and segments with that. Once the flow works the same two weeks in a row, automate steps 4 and 6 with n8n or Activepieces, keeping step 5 always human.
Common mistakes
- Automating sending from day one. First check by hand that messages sound right and reach the right person.
- Sending the same message to everyone. A good customer who gets a generic collections text reads it as distrust.
- Not logging the payment promise. Without it, the next message ignores what the customer already said.
- Letting AI decide sensitive cases. High amounts, complaints or personal situations are always seen by a person.
- Pasting customers' bank or card data into the chat. Name, amount, dates and history are all it needs.
Who this is NOT for
It is not for legal collections or for debt sold to third parties: those have their own legal rules in each country. It also will not help if you have very few invoices a month; a calendar reminder is enough there.
Want it applied to your business?
If you would rather not build it alone, book 30 minutes with me and we will review it with your real case. If you still do not know where to start, the map tells you what to automate first in 5 questions.