Silence detection is a monitoring approach that models the events a business expects to occur and raises an alert when they don't. Conventional systems are event-driven. Silence detection is absence-driven, the invoice that never arrived, the approval nobody gave, the customer who stopped replying three weeks ago.

Why absence is harder than presence

Detecting an event is trivial: the event carries its own signal. Detecting a non-event requires you to already know what was supposed to happen, and when. Silence detection needs three things: an expectation model derived from historical patterns, a timing distribution specific to each counterparty, and a threshold for meaningfulness.

What it catches that nothing else does

A supplier invoice that never arrived. An approval that was requested and never given. A client who stopped replying mid-engagement. A recurring report that skipped a month. A renewal conversation that never started. Each of these is individually small. In aggregate they are where margin quietly leaks out of a business.

How an expectation model is built

You learn it from history rather than declare it up front. Given past activity mail, tickets, invoices, document activity; you derive sequences that normally occur together. Once known, an incomplete instance becomes detectable. Three response options in increasing order of usefulness: alert (tell someone), contextualise (tell someone with the thread and comparable cases), or draft the next move (surface the absence alongside a prepared follow-up ready to send).

Where to start

Start narrowly: receivables (expected payments ranked by amount and delay unusualness), sales conversations (deals where the counterparty has gone quiet relative to their own historical response pattern), and approvals (requests sent to a named person with no reply and no downstream activity). Run in observation mode first for four to six weeks before acting on findings.

What it isn't

It isn't a reminder system. Reminders only cover what somebody already remembered to worry about. It isn't SLA monitoring. SLA monitoring watches items inside a system with a state machine. Silence detection watches for items that never entered a system at all.