Which number answers which question

Most reporting measures the wrong stage

A piece passes through stages, and each stage has its own number. Reporting one number for the whole thing tells you nothing you can act on.

• Impressions: how many times it was shown. This measures distribution, not your work. It is the number most often reported and the least informative.
• Click-through, or the first three seconds retained: whether the promise attracted the right person. This measures the cover, the title and the opening frame.
• Retention through the middle: whether the structure held. This measures the piece itself.
• Completion: whether the ending was worth arriving at.
• Saves and shares: whether it had value beyond the moment. A share is somebody spending their own reputation on you, which is why it is the most expensive action a viewer can take and the most meaningful.
• Comments: ambiguous by construction. Agreement, confusion and hostility all produce comments and none of them look different in the count.
• The outcome in the brief: enquiries, bookings, sign-ups, the thing you wrote down before you started. This is the only one that answers the question the work existed for.

Match the number to the stage. If the opening is the problem, completion will not tell you; if the ending is the problem, impressions certainly will not.

Reading a retention curve

The retention graph is the most useful diagnostic any platform gives away, and its shape is readable.

• A cliff in the first few seconds means the promise and the piece do not match: either the cover attracted the wrong people, or the opening did not answer what is this and is it for me. A very steep cliff with a high click-through is the signature of an inflated promise.
• A steady gentle decline is normal and healthy. People leave continuously; you cannot fix this and should not try.
• A sharp dip at a specific timestamp is the useful one. Something happened at 0:34, and you can go and watch 0:34. It is usually a slow passage, a moment where the piece restates something, a graphic nobody could read, or a claim that lost people’s trust.
• A rise means rewatching. Either a loop, or something people went back for: a number on screen, a detail, a joke. A deliberate rise is a strong signal; an accidental one usually means something was unreadable the first time.

This one curve, looked at honestly, is worth more than every other number on the page, because it points at a place in the file rather than at a mood.

Your sample is almost certainly too small

Here is the part that saves people from a great deal of wasted effort.

A video with forty views has a completion rate that is nearly meaningless. The difference between 38 per cent and 45 per cent on samples that size is ordinary variation; you can watch the same piece behave differently on two days for reasons that have nothing to do with the piece. Acting on it (changing your style, abandoning a format) is responding to noise.

What a small operator can do reliably:

1. Look at patterns across many pieces, not at one. Twenty pieces where the ones with a face in the cover consistently outperform is a signal. One that did well is an anecdote.
2. Look at the curve rather than the totals. A dip at a specific second is diagnostic even with few viewers, because you go and look at the thing rather than believing the percentage.
3. Prefer the outcome number, which is usually small but real. Three enquiries is three enquiries.

One number, chosen in advance

The brief already asked what success looks like. The discipline at this end is to hold to it: pick the one number before publishing and write it down, along with what result would make you change what you do.

Choosing afterwards guarantees you will find something that went up, because with a dozen metrics moving at once, something always did. That is not evidence. It is the same mistake as choosing the window that flatters a growth figure, arrived at from the other direction.

What these numbers cannot see

The honest limitation is about what is not counted, and it is large.

A piece watched in full by one person who then commissioned you registers as one view. A piece forwarded in a private message (where a great deal of actual sharing happens) is usually invisible to any dashboard you can see. A piece that changed somebody’s mind and prompted no action is unmeasured by construction. Meanwhile a piece with excellent numbers can have produced nothing at all.

So the numbers are a diagnostic instrument for specific questions, not a verdict on the work. They are very good at telling you where in a piece something went wrong. They are poor at telling you whether the piece was worth making, and treating them as though they answered the second question is how people end up with a body of work that performs well and says nothing.