A Python pipeline that sends personalised confirmation emails to thousands of sellers, then polls the inbox and auto-classifies every reply — completely unattended.
Each cycle meant sending personalised confirmation emails to between 500 and 5,000 sellers — then manually reading every reply, working out whether each said yes, had a question, or wanted to defer, and updating a spreadsheet row by row. It was a full day of repetitive, error-prone inbox work, every cycle.
What We BuiltEach seller gets a unique 16-character cryptographic token, written into a working Excel file — re-runs are safe and never reissue existing tokens.
Outlook-safe HTML emails with three one-click reply buttons (YES / QUERY / LATER) that pre-fill the subject line. Rate-limited, with each send committed immediately so a crash is safe to resume.
A scheduled job polls the IMAP inbox every few minutes, matches the token in each reply's subject, records the intent, and updates the seller's row — no human reading required.
Every event is written to an append-only audit log. Duplicate clicks, expired tokens, and mismatched senders are detected and routed to manual review.
What used to consume a full working day now runs in about five minutes of hands-on time. Every reply is classified automatically, every action is logged, and the whole pipeline runs on a schedule with no one watching the inbox. The kind of automation that pays for itself in the first cycle.