Baseline deviation

Baseline deviation is the metric that separates teams who get pulled into every false alarm from teams who get listened to when they actually escalate. It answers a different question than the absolute reading on the dashboard. The absolute reading tells you what a number is. Baseline deviation tells you whether that number is unusual for this specific entity.

Think of your commute. If it normally takes 25 minutes and today it takes 30, you shrug. If it takes 55, something happened. Two standard deviations is the monitoring equivalent of "my commute just doubled and I need to find out why."

This is the input I want in the first thirty minutes of any escalation conversation. Where is the entity's signal now, and how far is that from where the entity sits on an average week? Two standard deviations (how much the number normally bounces around day to day) below baseline gets a response — roughly, the reading is twice as far from normal as your typical day-to-day variation, which for most entities means something genuinely unusual is happening and not just noise. Within one standard deviation does not get a response, even if the absolute number looks ugly. The full reasoning, including why this is one of the load-bearing decisions in the first ninety minutes of a crisis, is in the first 90 minutes of a crisis response — and the reason absolute thresholds keep producing wrong answers in reputation work is the subject of why sentiment analysis fails during a reputation crisis.

The clarification that gets lost in vendor pitches: baseline deviation is not a feature you turn on. It is a number you have to maintain. The teams that respond well already had it on file before the crisis. The teams that do not are reverse-engineering it during the only ninety minutes they cannot afford to spend on math.