Invoice Follow-Up AI Agent
A services business was leaking cash to late payments. This was not because clients refused to pay, but because no one had the time to chase. I built a Relevance AI agent that reads the invoice ledger every morning, drafts polite follow-ups in the client's voice, and escalates the few that need a human conversation.
AI agent dashboard · agent decisions, drafts, and escalations
The problem
The business had eight overdue invoices on any given day. Following up took half an hour each — pulling the invoice, finding the right contact, writing a polite-but-firm email, logging the follow-up. Nobody had the half-hour. Cash sat unclaimed.
A blanket auto-reminder felt cold. They wanted the chase to sound like them.
What the agent does
- Reads the ledger dailyConnects to QuickBooks each morning, identifies invoices past their due date, and groups by age (1–7 days, 8–30, 30+).
- Drafts in the founder's voiceReads past correspondence with each client to match tone — relaxed for long-term clients, more formal for new ones.
- Escalates strategicallySends gentle reminders automatically up to a threshold. For older invoices, drafts a message and flags it for human review before sending.
- Logs every actionUpdates the CRM record with date, message sent, and any reply received — so the founder has full visibility.
- Handles repliesWhen a client responds, the agent categorises the reply (will pay, dispute, needs more time) and routes accordingly.
The guardrails
The agent never escalates a tone without human approval. It never replies to a disputed invoice. And every message it sends is logged with the reasoning behind the choice — so the founder can review and correct the agent's judgement.