Why Pulse
Three things a careful buyer usually has to dig for: how the measurement works, how it differs from what you already run, and where it stops being reliable. They are on one page because you should not have to ask.
1
Command reproduces every figure
from our own repository, not a spreadsheet
12/12
Runs agree on the top repair
across every seed and weighting tested
0
Conversion statistics on this site
because none has been measured on merchant traffic
The third number is the one worth reading. If a figure cannot be reproduced, it is not printed, and that rule is why there is no uplift claim anywhere here.
The method
Start with a shelf of products and a set of questions shoppers are trying to answer. For each attribute, we compute how much recording it everywhere would improve a shopper's ability to tell those products apart, and subtract what it already contributes today. The difference is the gap, measured in bits.
Bits are the right unit because the question is genuinely about information: how much uncertainty does knowing this field remove? A field recorded on every product with the same value removes none. A field recorded on nothing removes none today but could remove a great deal once populated. That is why absence alone is a poor guide and why a completeness percentage cannot answer this.
The second half is demand. An attribute only matters to the extent that shoppers are trying to decide something it bears on. We score each attribute against the stated requirements for that category, so a field that nobody's decision depends on scores near zero however empty it is.
How we compare
Most tools that prioritise catalogue work rank by engagement. That is reasonable and it has one structural blind spot, which happens to be the expensive one.
| Â | Engagement-ranked tooling | Pulse |
|---|---|---|
| Ranks by | What shoppers engaged with: filters used, clicks, channel performance | How much each attribute would change the decisions shoppers are trying to make |
| A field recorded on zero products | Generates no engagement, so it never rises | Can rank first. Screen size did, on the shelf we measured |
| Unit | Counts and percentages | Bits of decision information |
| Needs traffic | Yes, to learn from | No. It is a property of the catalogue |
| Tells you the cost of the fix | Rarely | Yes: extractable from your text, mappable from a feed, or genuinely absent |
| Publishes its limits | Seldom | On this page, in the next section |
This is not a claim that engagement ranking is wrong. If you already run one, keep it. They read the same catalogue fields we do, which is why the audit helps them too.
What we cannot claim yet
We have run no live experiment on merchant traffic, so we have no conversion or revenue number and will not offer one. What we will do is show you the experiment design, with a held-back control group, and report the result including when it is inconclusive.
On the shelf we measured, screen size came first in 12 of 12 runs across every seed, weighting and imputation scope tested. Positions two to four moved between runs. We publish the whole ranking but we only stand behind the top of it.
Our simulation of a completed catalogue can only use values the shelf already records. Where an attribute's recorded values are unrepresentative of its real range, that simulation is wrong in a way we can name but not yet correct. It needs a value distribution drawn from outside the catalogue being repaired.
The method is category-agnostic in principle. In practice we have measured a small number of shelves. Whether the same effects hold in yours is exactly what the audit finds out, and it is a fair reason to start with one category rather than a contract.