Why the same five
These are not the most impressive things I could build. They are the ones that keep being worth building, because they all address the same structural problem: a system that fails silently.
A ranking drop, a lead nobody answered, a seasonal complaint that never becomes a ticket, a feed that stopped syncing — none of these announce themselves. They are absorbed by the people closest to them and only surface at the moment they cost something.
The five
1. Search Console anomaly watchdog. Daily pull, per-page and per-query baselines, alert on real deviation. The failure it catches: a specific page losing a specific query, visible for weeks before anyone looked at the report.
2. Inbound lead triage router. Score, tag, route, and draft the reply on arrival. The failure it catches: an enquiry sitting unread in a shared inbox for two days.
3. Review-mining pipeline. Cluster complaints across every platform by the underlying problem, not the words used. The failure it catches: the theme that never became a support ticket because the customer simply left.
4. Content brief engine. SERP and sub-question extraction into a sectioned brief. The failure it catches: an hour of reconnaissance per article, which is the hour that never gets spent on whether the article should exist.
5. Feed integrity checker. Field-level comparison between the store and the channel. The failure it catches: an out-of-stock item still marked available, generating orders you cannot fulfil.
The shared shape
Every one of these has the same three parts:
- A baseline. Not a global threshold — a per-thing baseline, because “traffic is down” is noise and “this page lost this query” is a signal.
- A diff. The output is a change list, not a dashboard. Something is different from last time; here it is.
- An owner. The alert goes to a named person with the detail already attached, so the first question (“which page?”) is answered before anyone opens a tab.
What they have in common as builds
They are all small. None of them took more than half a day to build the second time. The setup time in the library — fifteen to thirty minutes — is the honest number for someone following the documentation with credentials in hand.
And they all replace judgement-free work. Deciding what to do about the alert is the job. Noticing it was never a job worth a human hour.
Where to start
Whichever one addresses a failure you have already experienced. If you have been surprised by a ranking drop, build the watchdog. If you have found a lead in your inbox that was three days old, build the router. The value of an automation is not in its sophistication; it is in whether it is watching something you currently cannot see.