Amazon's 'up to 71% more product coverage': count your own SKUs

Amazon's page claims its AI creative tools reach "up to 71% more product coverage and up to 15% sales growth," and the only useful way to treat "coverage" is to measure it yourself: the share of your SKUs that have a usable video. A Sume bulk run gives you a count to put against that.

Both numbers are Amazon's, they are "up to" figures, and the page does not say what the baseline or the sample was.

What does the page actually say?

Under AI Creative Tools within Amazon Ads Agent, the unBoxed page says the tools generate display ads and TV-quality video from product assets and quotes the two figures. It lists the tools as live in the conversational experience, with no rollout date. It gives no definition of "coverage," no sample size and no time window.

Without a definition, the number can mean products with any video, products with a new video or ad impressions per product. Do not compare it to yours until you pick one.

"Up to" figures describe a best case, so they say little about a typical account. A fair reading of both numbers is that something happened for some advertisers under some conditions Amazon did not publish on that page.

How do I define coverage for my catalog?

Pick the simplest definition: SKUs with at least one video that meets your channel's specs, divided by active SKUs. Write the number down before you start. Then decide the denominator: all SKUs, or only those with enough inventory to advertise.

How do I generate for the gaps and count honestly?

A Sume bulk run queues up to 100 Format runs with a concurrency window of 1 to 16. The docs are blunt about the trap: completed means every item is terminal, not that every item succeeded, so read counts.failed. A bad item fails the create with a 400 before any queue exists, and a reused Idempotency-Key returns the old queue, so mint a fresh key per batch.

Use the failed count to subtract. If 100 SKUs were queued and 7 failed, your coverage rose by 93 SKUs at most, and you still need to confirm each file.

curl -sS https://api.sume.com/v1/format-run-queues/$QUEUE_ID \
  -H "Authorization: Bearer $SUME_API_KEY"

What would be a fair comparison with Amazon's number?

There is no fair comparison, because the definitions differ. What you can do is compare your own before and after, and compare cost per covered SKU across tools. Sume bills each accepted job on its own, so you can divide spend by new covered SKUs. Do the same for any other tool and keep sales out of it unless you ran a holdout.

A sales claim like "up to 15% growth" needs a control group. Without one you are reading seasonality, promotions and ranking changes along with the video.

A holdout is simple in principle: pick comparable products, give video to some and not the others, and compare over the same weeks. It costs you the video on the holdout group, which is the price of knowing.

What would I do before the holiday season?

Fill coverage gaps for your best sellers first, since a missing video there costs the most. Then repeat the count. If you need the per-SKU recipe for Meta, this post covers it.

Whatever the result, record it next to the date and the model names you used. Generators change, and a number from September will not describe a clip made in December.