Crisis Velocity: Measuring Against Your Own Normal

Crisis velocity is how fast mentions of a brand are accelerating compared with that brand’s own normal for the same channel and time of week. Measured against a personal baseline instead of a fixed count, it flags real events early and ignores routine busy days.

Ask most monitoring tools what counts as a crisis and you will eventually find a number sitting in a settings page. Fifty mentions in an hour. A hundred. Two hundred. Whatever it is, it was almost certainly not chosen for the brand it is now protecting.

The number is not the problem in itself. The problem is that a single fixed count cannot describe an event across brands of different sizes, in different categories, on different days of the week, which means it will always be simultaneously too sensitive for one brand and too blunt for another.

Why fixed thresholds cannot work

Consider two brands with the same threshold of a hundred mentions per hour.

The first is a national consumer brand that generates several thousand mentions on an ordinary day. For it, a hundred an hour is a slow Sunday. The threshold will either be crossed perpetually, in which case it is ignored, or it will be raised so high that only a catastrophe trips it, at which point the alert is telling the team something they already learned from the news.

The second is a specialist business supplier with eleven mentions in a normal week. A hundred an hour will never happen, not during a serious crisis, not ever. This brand’s genuine emergency might be nine mentions in an afternoon, including two from trade press, and the system is numerically incapable of noticing.

Both brands bought the same product and both are unprotected, in opposite directions, for the same reason.

Velocity is the rate, not the level

The useful reframe is to stop asking how many and start asking how fast this is changing compared with how fast it usually changes.

Volume that jumps once and then holds steady is nearly always a single item performing well. One advert, one popular post, one article in a large outlet. It produces a high number and a flat curve afterwards.

Volume that compounds looks different. Each interval produces more than the one before, because the mechanism is not one item being seen but many items being created in response to each other, which is what a story spreading actually is. The absolute numbers at hour one may be small. The shape is the signal.

This is why velocity is worth measuring properly rather than approximating with a count. A brand can reach a very high number harmlessly, and it can be in genuine trouble at a number that would not trip any threshold, and the two are distinguishable by acceleration long before they are distinguishable by size.

What a baseline actually needs to account for

Normal is a more complicated object than most alert configurations assume.

A long enough window. Long enough to contain the brand’s own cycles. A baseline built on a few weeks will confidently flag the first busy season it encounters as an anomaly, because it has never seen one.

Day and hour structure. Almost every brand’s mention volume has a weekly rhythm and a daily one. Treating a Monday morning and a Saturday night as the same expected level produces alerts that fire on ordinary weekday activity and miss genuine weekend events, which is the wrong way round given that weekends are when most teams are least staffed.

Channel separation. Press and social do not share a scale. Six press articles is a lot. Six social posts is nothing. Averaging them into one baseline produces a number that describes neither, and alerts on the noisy channel while being unable to see movement in the significant one.

Subject separation. Your brand, your chief executive, your flagship product and your category each need their own normal. Category volume rising because a competitor is having a bad week is a genuinely useful signal, and it is a completely different event from your own brand volume rising. A single combined baseline cannot tell you which one is happening.

An absolute floor. Percentage-based deviation is unstable at low volumes. Two mentions becoming eight is a 300 percent increase and almost never means anything. Any deviation system needs a minimum count below which it simply does not fire, and for sparse brands that floor does most of the work of preventing false alarms.

Severity should escalate on evidence

Once something is outside normal, the question is how serious, and the temptation is to answer it with more volume. That produces the wrong ladder, because ten thousand additional mentions of a harmless thing is still a harmless thing.

A more defensible ladder escalates when new kinds of evidence appear rather than when the same evidence accumulates.

The first rung is deviation alone. Something is outside this brand’s normal pattern. Worth a human look within the working day, nothing more.

The second rung adds a second independent signal. Sentiment has moved against its own baseline, or the composition of who is talking has changed, or a second channel has started moving. Two unrelated things being unusual at once is considerably more meaningful than one thing being very unusual.

The third rung requires something qualitatively new: a journalist in the conversation, a named victim, a regulator, an identifiable second-day mechanism. This is the rung that justifies interrupting someone’s evening, and keeping it rare is what makes it work when it fires.

The design principle is that the top of the ladder should be reachable only by evidence that a person would independently agree is serious, which keeps the alert honest and keeps the team responsive to it.

Rebuilding the baseline after a step change

Baselines go stale, and the failure is quiet. Anything that permanently changes a brand’s normal volume invalidates the history behind it: entering a new market, a major acquisition, a sustained campaign that lifts coverage for months, or a previous crisis that left ongoing coverage at a higher level than before.

Continuing to measure against a pre-change baseline produces one of two outcomes. Either everything looks like an anomaly, and the alerts get muted, or the new higher floor gets absorbed into the average and the brand becomes correspondingly harder to alarm, which is worse because it is invisible.

The practical habit is to review the baseline whenever a deliberate step change happens, and to notice when the alert rate itself changes character. An alerting system that suddenly goes quiet for a month has often not had a quiet month.

The traps worth checking

The baseline period was unusually calm. Most tools are configured during onboarding, which tends to be a deliberately quiet stretch. A baseline built there will read ordinary activity as elevated for months.

Search-discovery data is being counted as volume. Some sources can establish that a conversation exists without honestly measuring its size, because what is being counted reflects how a crawler indexed something rather than how many people posted. Feeding that into a velocity calculation produces spikes that correspond to nothing in the world.

Only one language is baselined. For a brand operating across language markets, a baseline built on English coverage will be blind to the market where the story usually starts. The velocity calculation is only as good as the coverage underneath it.

Nobody checks the false negative rate. Teams notice false alarms because they are annoying. Nobody notices the alert that should have fired, because there is no notification for silence. The only way to find these is to look back at incidents that did happen and ask whether the system saw them coming.

When there is no history to build on

New brands, new products and newly monitored subjects have no baseline, and pretending otherwise produces confident nonsense for the first few months.

The honest approach during that period is to run on an absolute floor and on composition rather than on deviation. Alert on any press mention at all, because there are few enough to read. Alert on any mention by a journalist or a named account of significance. Accept that volume-based detection is not yet available and will not be for a while, and let the baseline build in the background until it has actually seen enough to be worth using.

This is less satisfying than a dashboard with a working alert on day one, and it is considerably better than an alert that fires on a threshold nobody had grounds to choose. The general question of what separates a genuine event from noise is covered in crisis detection, and the continuous version of this calculation is what the crisis velocity clock exists to run.

Common questions

How do you set a crisis alert threshold?

Derive it from the brand's own history rather than picking a number. Establish what normal volume looks like for that specific subject and channel, accounting for day of week and season, then alert on deviation from that pattern rather than on an absolute count. A single fixed number cannot work across brands of different sizes.

What is crisis velocity?

Velocity is the rate at which mention volume is changing rather than the level it has reached. A crisis has a characteristic acceleration, where each interval produces more than the last, which distinguishes it from a single piece of content that performs well once and then plateaus at a high number.

How much history do you need to build a baseline?

Enough to have observed the brand's own cycles, which usually means several months rather than several weeks. A baseline built on a short window mostly describes that window, and will misread the first seasonal peak or campaign period as an anomaly.

Why do small brands get false alarms from percentage-based alerts?

Because percentage change is unstable at low volumes. A brand going from two mentions to eight has risen 300 percent while nothing meaningful has happened. Any deviation-based system needs an absolute floor underneath it, below which percentage movements are ignored.

When should you rebuild a crisis baseline?

After any step change that permanently alters normal volume: entering a new market, a major acquisition, a large sustained campaign, or a previous crisis that left coverage elevated. Continuing to measure against a pre-change baseline produces either constant alerts or permanent blindness.