"How do we make something go viral" is the wrong question, and not for the moralising reason people usually give.
It's the wrong question because virality is a property of distribution, and distribution is the part you don't control. A platform decided to show something to a lot of people. You can influence that decision at the margins. You cannot cause it, and the things that caused it last quarter are frequently not the things that cause it this quarter.
There's a better question available, and it's answerable: which of our posts beat what we'd normally expect, and what did they have in common?
What an outlier is
An outlier is a post that substantially outperformed the baseline for the account that published it.
The "for the account" part is doing all the work. A video with 50,000 views is a disaster on a channel that normally does half a million, and a landmark on one that normally does two thousand. Only the second is telling you something you can act on.
This is why "viral" is analytically useless as a category. It's an absolute threshold applied to accounts of wildly different sizes, which means it selects almost entirely for how big you already were.
Why outliers are the good evidence
An outlier is the closest thing to a free natural experiment you get in content.
Think about what was held constant. Same channel. Same audience. Same subscriber base. Roughly the same posting cadence, the same brand, the same production quality, the same era of the algorithm. Nearly every variable that normally confounds a comparison was fixed.
Something about that specific post was different. And because everything else was constant, the difference is unusually likely to be causal rather than coincidental.
Compare that to the usual alternative: benchmarking yourself against a competitor. There, everything differs: audience, size, history, budget, luck. You can't isolate anything. Your own outliers are a far cleaner instrument than anyone else's averages.
Finding them
Four steps. You can do this in a spreadsheet in an afternoon.
1. Pick one platform and one format
Never mix. A Reel and a twelve-minute YouTube video are measured differently by different systems for different behaviours. A comparison across them produces a number, not information. Same for cross-platform averages: they're always misleading and usually flattering.
2. Establish the baseline
Take your last twenty to thirty posts in that platform-and-format and find the median, not the mean.
Use the median deliberately. One runaway post drags an average high enough that everything else looks like underperformance, which is exactly backwards, since the runaway is the thing you're trying to isolate. The median is robust to it.
If you have fewer than about fifteen posts, you don't have a baseline yet. Say so and wait. A benchmark built on six posts will produce confident nonsense.
3. Compute the multiple
For each post: its views divided by the baseline median. A post at the median scores 1.0. A post that did three times the norm scores 3.0.
Where you set the outlier threshold depends on how noisy your channel is, and you should set it by looking at your own distribution rather than adopting someone else's rule of thumb. On a stable channel, 2× is genuinely unusual. On a volatile one, 2× happens most months and means nothing.
4. Do the same for engagement
Repeat with engagement per view rather than raw views.
This second pass is where the useful surprises live, because the two lists don't overlap as much as you'd expect. A post can be a view outlier and an engagement dud. That's distribution, and it tells you about your title and thumbnail. A post can be an engagement outlier with ordinary views. That's content, and it's usually the more valuable finding, because it means the piece landed hard with the people who saw it and simply never got shown to more of them.
Reading them
Now the part that isn't mechanical.
Line up your outliers and look for what they share. Structure of the opening. Subject matter. Who's on camera. Length. Whether they answer a question or tell a story. Whether they were about the product or about the audience's problem.
Then (and this is the step everyone skips) look at your worst performers and check whether they share it too.
If your top five all open with a question, that's interesting. If your bottom five also all open with a question, you've found a habit, not a signal. A pattern only counts as a content signal if it's present in the wins and absent from the losses.
The traps
One outlier is an anecdote. A single post that did 5× tells you almost nothing, because content performance has a long tail and occasional freak results are the expected behaviour of the system, not a message from it. You need the pattern to repeat.
Timing masquerades as content. A post published the week your category was in the news isn't a format insight. Check the calendar before you conclude anything.
Sample size beats cleverness. Thirty posts and a boring conclusion beats eight posts and a brilliant one. Every time.
Copying yourself has a half-life. Outlier analysis tells you what worked. Audiences habituate; the fourth version of the format that worked in March will not perform like the first. Treat the finding as a hypothesis with an expiry date, not a formula.
Where this shows up in Acumin
The Snapshot ranks your top performers by format from real public view counts, and Outlier Studio is the surface built specifically for this analysis: your own posts, scored against your own baseline rather than against the internet.
The reason it's built that way rather than as a "what's trending" feed is the whole argument of this article. Trending tells you what someone else's audience did. Your outliers tell you what yours did.
How to use this tomorrow
Export your last thirty posts on one platform. Median views. Divide. Sort.
Look at the top three and the bottom three, side by side, and write one sentence about what separates them. Then go and check whether that sentence also holds for posts four through eight.
If it does, you've got something to test. If it doesn't, you've saved yourself a quarter of building on a coincidence, which is the more common and more valuable outcome.
Related: What your evidence is actually worth explains why public view counts can support this analysis but not a creative-quality judgement. How to research your competitors on YouTube applies the same method to accounts that aren't yours.