What to track, why attribution lies, and the three figures that say whether the content paid for itself.
The three figures that count
They are enough to decide, and all the others are diagnosis.
The cost per purchase attributed to the video, compared with your other creatives over the same period and the same budget.
The conversion rate of the page the video sends to, before and after it went live.
And the volume of direct contacts, measured as described above.
A rule that avoids mistakes: always compare at equal budget and over the same duration. A video receiving three times more money mechanically gets better figures.
Why attribution lies
Three reasons, and they stack there more than elsewhere.
Part of the audience refuses tracking, and those people buy anyway: they show up as direct traffic, with no origin.
Sharing often happens privately, in a conversation, which severs any link with the original video.
And the physical shop is still very present: a video seen in the morning can produce a purchase at the counter in the afternoon, which nothing will ever connect to the campaign.
Practical consequence: stop looking for precision, look for direction. An imperfect but constant measurement beats a sophisticated model that changes every quarter.
The cost nobody counts
Four items, invisible in the quote and very real in the result.
Internal coordination time: searching, exchanges, sending the product, approval.
The cost of unplanned revisions.
The media budget needed for the video to exist: a good creative with no distribution proves nothing.
And the opportunity cost of delay: a video delivered three weeks late misses the window it was ordered for.
Add those four items to the price paid to the creator and you get the real cost of your content, often double the invoiced amount.
How to know whether a specific video works
Three conditions, without which the test says nothing.
An identical budget for each variant, on the same audience, in the same week.
One single difference between versions: the hook, or the format, or the person, never all three at once.
And a long enough run: two weeks minimum, because a variant can lead for three days then collapse.
What you observe, and when
| Indicator | What it tells you | When |
|---|---|---|
| Three-second view | Whether the hook holds | From 1,000 views |
| Full view | Whether the edit holds | At 5,000 views |
| Click to the product page | Whether the promise is clear | After two days |
| Purchase | Whether the product follows | After a week |
What distorts the reading
Comparing two videos on different budgets, or drawing a conclusion from a weekend. A test is read over seven full days, with the same budget on each side.
What you do with each weak segment
A weak hook is replaced, a weak middle is shortened, a weak ending is rewritten. One part does not condemn the whole video.
What to note every time
Five lines in a table, filled in on the day.
The publication date and the stop date.
The budget committed.
The hook used, in one sentence.
The result on the three figures at the start.
And a free observation: what you noticed, in the comments or in the calls.
After ten lines, that table will tell you what no tool will: which type of video works on your product, with which profile, in which period. It is that accumulation, not an automatic report, that lowers acquisition cost from one year to the next.
