The alert nobody sets up
Most stores have alerts for downtime, payment failures, and stock-outs. Almost none alert on the scariest signal of all: zero sales. Your checkout can be technically green and still dead. If nobody bought for 12 hours, no monitor screams — you find out in the monthly report, 3 weeks late.
The dumbest possible watchdog wins
You don't need ML. You need a loop: poll your sales API every 5 minutes, compare against the last known number, and page yourself when the counter doesn't move. Ours is 20 lines of bash + Python:
- Fetch order count and revenue.
- Log one line: timestamp, revenue, sales count.
- If revenue is unchanged for N cycles AND you had traffic — alert.
- Break the loop on first dollar: a sale is the bell, everything after it is noise.
That's it. The value isn't sophistication; it's that the silence becomes loud. A dead checkout, an expired token, a broken product page — all of them collapse into one visible number stuck at zero.
What the zero actually means
When the number is stuck, it's usually one of five things: the store itself is down (check with a status probe), the payment rail expired (tokens die quietly), the product 404s (rename/regression breaks links), traffic went to zero (distribution died, not the store), or the offer stopped converting (price, page, proof). Each has a different fix, and the watchdog buys you the hours you'd otherwise lose.
Then make the alert boring
Run it on a schedule that survives restarts, log to a file a human can read, and treat "monitoring loop is down" as its own emergency — an unmonitored store is just a store you've decided to check monthly. The cheap version takes an afternoon and pays for itself the first time it catches a dead checkout at 2am instead of at month-end.
If you want the plug-and-play version — the watchdog script, checkpoint tooling, and the incident playbook behind this post are in the 24/7 Ops Starter Kit, alongside automation packs built from the same playbook.
Top comments (0)