Industry
Part of How to tell whether sharing or recommendation carried a viral spike
A viral spike has four phases: what onset, peak, decay, and residue tell you
A viral spike in four phases: what onset, peak, decay, and residue can and cannot tell you, when to stop counting, and how to measure the residue.
A spike is not one event; it is four phases with different causes, different information content, and different correct responses. Treating them as one thing called "going viral" is why people react at the wrong moment.
This page separates the phases, says what each can tell you, and gives the point at which continuing to watch stops paying.
What to take away
- The onset carries almost no information. It looks identical for episodes that end at a hundred and episodes that end at a million.
- The decay is the informative phase, because its rate tells you whether anything durable happened.
- The residue, whatever remains once the spike is over, is the only part that matters in three months, and it is almost never measured.
Four phases
| Phase | Duration | What is happening | What it tells you |
|---|---|---|---|
| Onset | Hours | A cluster is being reached for the first time | Almost nothing; every episode looks the same here |
| Peak | Hours to days | Reachable supply is being exhausted | The size of the cluster, not the quality of the item |
| Decay | Days to weeks | Saturation, plus competition from newer items | The rate here is the useful measurement |
| Residue | Permanent | Whatever stayed: audience, references, findability | The only phase with a lasting effect |
You want to act at onset and peak, when numbers move and pressure peaks. That is when you know least.
Four phases of a viral trend
- Onsetfirst cluster reached, tells almost nothing
- Peakreachable supply exhausted, size not quality
- Decaysaturation and competition, the useful rate
- Residuepermanent, the only lasting effect
Decisions in those hours rest on a few hours of a process whose shape is not yet determined. The arithmetic that governs the shape says an S curve's early part does not distinguish between outcomes.
Reading the decay
Once the peak is past, the shape of the fall carries real information, and it is cheap to read.
Three shapes of decay
Fast clean fall
- What it means
- attention dispersed
- Residue
- small step up
- Action
- common case
Slow fall
- What it means
- still being found
- Residue
- long tail
- Action
- investigate use
Undershoot
- What it means
- audience changed
- Residue
- blended rates
- Action
- read a cohort
A fast, clean fall to a level slightly above the old baseline means the episode was a burst of attention that has now dispersed. The residue is the small step up. This is the common case.
A slow fall means the item is still being found, usually because it has become discoverable through search or has been picked up in a second cluster. That is worth investigating, because it suggests the item has a use rather than a moment.
A fall that undershoots the old baseline is worth taking seriously. It usually means the audience changed composition, so the same absolute activity is now spread over a larger and less interested group, which is the mismatch problem that makes a post-spike rate unreadable. The separation that fixes it is reading a cohort rather than a total.
Daily numbers during and after a spike are dominated by day-of-week effects and the spike's own tail, so smooth the series first.
A short moving average is enough and is the standard first step for any time series of this kind. Fitting anything more elaborate to a few weeks of post-spike data is fitting noise.
When to stop counting
Three stopping rules, any of which is sufficient.
Stop when the series returns to within its own normal variation. Not to the old average, which it may never reach: to the range it used to move around in. Once you are inside that range, further daily readings are telling you about ordinary variation.
Stop when the decision the numbers were feeding has been made. Most of the watching that happens after a spike feeds no decision at all, which makes it a cost with no return.
Stop when the audience composition has settled. Until then the rates are a blend of two populations and mean nothing as a total.
Measuring the residue
The residue is what the whole episode was worth, and it is four questions asked once, a month or two later.
How many people are still here, and are they responding to ordinary work? A follower who never engages again is not residue.
Is the item still being found on its own? A piece that keeps surfacing has become an asset with a long tail; one that stopped is a memory.
Did anything durable get built during the window, such as a channel to your audience that does not depend on the surface? That is the one asset that survives a change of platform, which is why it is worth more than the follower count.
Has your baseline moved, and by how much? This is the number people quote afterward and almost nobody measures properly, because it requires the patience to wait until the series is readable again.
Only the last of these needs arithmetic. The other three are answered by looking, which is why they get skipped in favor of a dashboard.
Where the timeline misleads
Two traps are worth naming. The first is treating a second spike as a continuation. It is usually a different cluster with its own unsaturated supply, and the two should be measured separately for the same reason two cohorts should.
The second is comparing your spike's shape to a published one. Published shapes are drawn from survivors and are usually rescaled to look alike, so the comparison is uninformative before it starts. The general failure is set out in why comparing hits to hits teaches nothing, and the honest alternative is comparing your episode to your own ordinary weeks.
Common questions
How long is the whole cycle?
There is no general figure, and the ones in circulation come from samples of published cases. What is reliable is the ordering: onset before peak, saturation before decay, residue after. Time your own episodes and use your own numbers.
Can I extend the peak?
Occasionally, by reaching a cluster the item has not saturated, which usually means a different surface or a different framing. It is not a lever you can pull on demand, and the effort is often better spent on the residue, which is what remains afterward.
Should I publish more during a spike?
Usually not the same kind of thing, and this is the most common expensive mistake. The arriving audience came for one item; feeding them more of it deepens the mismatch, and the pull to do it is the incentive described in the pressure that forms after a hit.
What if there is no residue at all?
That is a common and honest outcome, and it is worth recording plainly rather than explaining away. It means the episode was attention that passed through, which is information about the surface and the audience rather than a verdict on the work.
