The problem
Most small-business automations have no monitoring. A broken Zapier step, an expired login or a renamed column means the data just stops, and nobody notices until a report is wrong or a customer complains days later.
Why it works
Staleness is a rule, not a judgement: "this feed should update every 24 hours and the newest row is 3 days old". Code checks it. A model is optional, only to write the alert in plain words. Alerts go to a person once per problem, not every 4 hours.
Step by step
- List each feed: where it lives, how often it should update, the date column to check, and who owns it.
- Every 4 hours, read the newest timestamp and row count of each feed.
- Flag: newest row older than allowed, row count zero or far below normal, required column missing.
- Send one alert per new problem with the owner, the feed and how long it's been stale.
- Send a 'recovered' note when it's fixed. Keep a log.
Pitfalls to avoid
- Alerting every run trains people to ignore it: alert on change only.
- Weekends and holidays: set expected gaps per feed.
- Timezones: store and compare in one zone.
- Don't let the watchdog 'fix' feeds; restarting the wrong thing can duplicate data.
Free · Kickstart prompt
Get a rough first version today
Paste this into Claude, ChatGPT or Grok with your own data. It gets you a useful first pass, not the tested setup.
Help me set up a simple watchdog for the data feeds my business depends on. Here is my list of feeds (name, where it lives, how often it should update, owner): [PASTE LIST] And here is the latest status I can see for each (newest timestamp, row count today, row count normally): [PASTE STATUS] 1. Tell me which feeds are stale or broken right now and why, in a table: feed, problem, stale since, owner, suggested first check. 2. For each feed, write the exact rule I should monitor (e.g. "newest row older than 26 hours on weekdays"). 3. Write the alert message template and the "recovered" template, plain and short. Don't assume a feed is fine if I gave no data for it; mark it "unknown".
✓ Prompt tested 14 Oct 2026: we ran it with gpt-5-mini on a synthetic sample and checked the output by hand.
This one is a free idea, not a paid Blueprint. If you want us to build and test the full version, tell us below.
Should we build this next?
One tap moves this up our build queue. Add an email only if you want a note when the tested Blueprint is ready.