An AI agent that runs collections like your best analyst
Reads an email thread, judges the right escalation level, and drafts the follow-up — with the accounting nuance built in.
Context
Collections were reactive and inconsistent. Tone and timing depended on whoever happened to send the email; escalation was ad hoc; and sensitive cases — related-party balances, export receivables with regulatory realisation timelines, genuinely disputed amounts — were handled like routine ones, sometimes making the situation worse.
Challenge
Make collections aggressive and consistent without a follow-up landing wrong on a relationship-sensitive or genuinely disputed account. The hard part isn't writing a reminder — it's knowing *which* reminder, at *what* tone, for *this* customer, today.
What I built
A GenAI agent built as a structured master prompt with a defined workflow:
1. Extract the facts from the thread — invoice, amount, due date, days overdue, prior follow-ups, any dispute or tax-mismatch signal.
2. Auto-detect an escalation level on a five-tier matrix, from soft reminder through to pre-legal.
3. Apply customer-type rules — contract customers, related-party, export / FEMA, domestic — as modifiers on top of the base level.
4. Draft the email at the right tone, with one specific time-bound ask and a level-appropriate consequence.
The judgement a senior person makes automatically is encoded explicitly: a related-party balance gets a "is this a reconciliation-timing issue, not a collections problem?" check before any firm language; export accounts reference realisation timelines; pre-legal drafts are flagged for human sign-off and never sent on the agent's own authority. A human always confirms the level before anything goes out.
Outcome
Consistent, escalation-aware collections instead of tone-by-mood. The risky edge cases handled correctly by default. Drafting time cut from minutes per email to seconds — with the judgement built in, not stripped out.