Integrations Fail Silently

A lead sync we inherited had been broken for weeks. Ads to a spreadsheet to a notification channel to the email tool, and it worked perfectly until somebody swapped the ad form. Nothing errored. Nothing alerted. Leads simply stopped arriving, and the first signal was a salesperson asking why it had gone quiet.

Why do integrations fail without erroring?

Because most integrations are triggered rather than polled. A form submits, a webhook fires, a row appears, the chain runs. Change the form and the trigger points at something that no longer exists, so the chain is never invoked.

Nothing throws an exception. There is no failed run to appear in a log, because there was no run. From the monitoring tool’s point of view the integration is healthy and idle.

What breaks, and whether anything tells you:

Change	Made by	Errors?	How it surfaces without monitoring
Form rebuilt, new ID	Marketing	No	Sales notices the silence, weeks later
Field renamed	Anyone	Sometimes	Records arrive with blank fields
Tool migrated	IT	No	Webhook posts into nothing
API key rotated	IT or security	Yes	Failed runs in a log nobody reads
Send quota hit	Volume growth	Sometimes	Partial delivery, silent truncation

What actually breaks these chains?

An ID changing underneath. A form rebuilt rather than edited, so it gets a new identifier. A field renamed, so the mapping no longer matches. A tool migrated, so the webhook URL is stale. An API key rotated on a schedule nobody told you about.

All four are routine maintenance by somebody who has no idea an integration depends on them.

How do you monitor for silence?

Set a floor and alert when volume falls under it for longer than the normal quiet period. If the form usually produces leads every weekday, then two consecutive weekdays at zero should page somebody.

This is the only check that catches the failure mode, because the failure looks exactly like a slow week until it does not.

Alert on absence, not on failure. A broken pipe and a quiet week look identical until you have a baseline.

How often should you test one?

After any change to a form, field, tool or key, and on a schedule regardless.

A monthly synthetic submission that runs the whole chain end to end and confirms arrival at the destination is the only test that proves the pipe works during a genuinely quiet period. Without it you cannot distinguish a broken integration from a slow month, which is the exact ambiguity that let this one run broken for weeks.

What should you document?

For each integration: what triggers it, which specific IDs or field names it depends on, where the data lands, who owns each end, and what normal volume looks like.

That last item is what makes an alert possible. Without a baseline, nobody can tell the difference between a broken pipe and a quiet Tuesday.

The neighbouring failures worth auditing at the same time are checking the ceiling of every tool in the chain and verifying what is actually live.