A YouTube Example of ANF: When “Typical Performance” Becomes a Normal Functioning Band

Sometimes the best examples of an idea appear in places where you do not expect them.

Recently, while looking at analytics for one of the videos on my YouTube channel @iTrubin, I noticed something that immediately reminded me of work I started more than twenty years ago.

YouTube now shows a “Typical performance” band around a video’s actual performance.

In the example below, the blue line shows the actual number of engaged views, while the gray area shows the range YouTube considers typical.

That immediately looked familiar to me.

This idea goes back to my SETDS work

In 2002, I published my first paper on SETDS — Statistical Exception and Trend Detection System at CMG’2002. That paper received the Best Paper Award.

The central idea was simple: a raw number by itself is often not very meaningful. What matters is whether that number is inside or outside the expected range of behavior.

This is how I approached IT performance data at the time.

Instead of asking:

“Is this value high or low?”

the more useful question was:

“Is this value normal for this system, at this time, under these conditions?”

That idea later became part of what I now describe more broadly as the Area of Normal Functioning — ANF.

And then I saw YouTube doing something very similar

I started my YouTube channel back in 2009. At that time, this kind of “typical performance” band was not part of YouTube Analytics.

Today it is.

YouTube compares the current behavior of a video with a range based on historical performance and presents that expected range visually.

For example, in the case shown here:

• Engaged views: 44
• Typical range: 10–50

and:

• Views: 99
• Typical range: 20–100

The raw values are useful, of course. But the ranges are much more informative.

Forty-four engaged views by itself tells me almost nothing.

Forty-four engaged views inside a typical range of 10–50 tells me that the video is performing toward the upper end of what is normal for this channel and this type of content.

That is a much richer interpretation.

Collective of Customers / Viewers as an ANF example. YouTube Analytics compares actual video performance with a “Typical performance” range. The example is based on analytics from my YouTube channel, @iTrubin.

Why I see this as an ANF example

What makes this particularly interesting to me is that the expected range is produced not by one viewer, but by the behavior of a collective.

A large number of viewers interact with content over time. Their combined behavior creates a statistical pattern.

That pattern becomes the “normal” or expected range.

So in this case:

the collective of viewers creates the Area of Normal Functioning, and the individual video is evaluated against it.

This is exactly the kind of extension of ANF that I find interesting.

Originally, I used similar ideas for computer systems and performance data.

Now the same way of thinking can be applied to:

• customers,
• viewers,
• users,
• athletes,
• teams,
• organizations,
• and many other dynamic systems.

One important clarification

I am not suggesting that YouTube copied SETDS or used my work directly.

I did not patent SETDS, and I have no evidence of any connection.

What I find much more interesting is that similar analytical ideas appeared independently in commercial tools years later.

Over time, I have seen other products introduce concepts that looked familiar to me as well, and I have written about some of them on this blog.

For me, that is actually encouraging.

When the same basic concept keeps reappearing independently in different fields, it may be a sign that the underlying principle is useful and general.

From anomaly detection to normal functioning

My original SETDS work was mostly about finding exceptions, trends, and changes.

But over time, I became increasingly interested in the opposite side of the same problem:

What does normal behavior look like?

And more importantly:

How wide is the normal range, how does it change over time, and what happens when the system begins to move outside it?

That is one reason I now use the broader ANF concept.

The YouTube “Typical performance” band is a simple and very practical illustration of that idea.

It tells the creator:

“Here is where your content normally operates. Now let us see where this particular video falls relative to that area.”

That is ANF thinking in a very accessible form.

Final thought

The concept of an Area of Normal Functioning is not new in my work. I first introduced ОНФ/ANF decades ago in my PhD research on robotic grasping and passive adaptation.

Later, SETDS brought a related idea into statistical analysis of dynamic computer systems: instead of judging a measurement in isolation, define the expected range of behavior and detect meaningful departures from it.

What interests me now is how these lines of work come back together.

Seeing YouTube use a “Typical performance” band is interesting not because I believe there is any direct connection to my earlier work, but because it demonstrates how naturally the same principle appears in another domain.

Whether we are looking at a robotic grasp, a computer system, an athlete, a customer population, or millions of viewers, the underlying question is remarkably similar:

What is the Area of Normal Functioning for this system — and when does its behavior meaningfully move beyond it?

That is the broader idea I am now trying to develop with ANF.