The cron jobs were the business.

What we did.

1. 1

   We made renewal the central job

   It finds listings approaching expiry, sends reminders, and — for listings more than seven days late — automatically downgrades them, updates the invoice, and re-indexes the record so the change is visible in search immediately.

   Seven days, then downgrade
2. 2

   A downgrade you cannot see in search is not a downgrade

   It is a downgrade in the database only, and the advertiser keeps the placement they stopped paying for. That is why re-indexing belongs in the same job rather than in a nightly batch.

   Visible immediately
3. 3

   We handled recurring payments with all three outcomes

   Success, failure and upcoming charge each have their own message. Payment failure is not an error state to log — it is a conversation with a customer who probably wants to keep the service.

   A conversation, not a log
4. 4

   We timed re-engagement to the annual cycle

   A three-step campaign at roughly 179, 182 and 186 days from a listing’s last update. Not arbitrary: half a year out from renewal, a listing with stale information is a listing whose owner has stopped thinking about it — and that is who fails to renew.

   179 · 182 · 186 days
5. 5

   We chose the channel per job, and kept one route out

   Reminders about unfinished draft listings go in-app rather than by email: someone halfway through creating a listing doesn’t need to be chased into their inbox, they need to be met where they left off. Every job followed the same shape — query the database, take a template from object storage, write a record, put a message on a queue — and one service did the actual sending. Outside production that service redirected all mail to a test address, which made emailing real customers from staging structurally impossible.

   One route out, one redirect