Skip to main content
Amazon’s numbers for an ad keep changing for about a week after the click: orders are credited days later, and spend is restated, sometimes downward. A day read too early looks worse than it was, so every number we read about ad results either waits for the day to settle or corrects for its age.

​
What counts as an ad’s order

Amazon credits an order to an ad when the shopper clicked the ad and then bought within a fixed attribution window: the number of days after the click during which a purchase still counts. Amazon credits the order to the last ad clicked (last touch).
  • Windows. Amazon reports purchases and sales in 1, 7, 14 and 30 day windows. Its default is about 7 days for Sponsored Products and 14 days for Sponsored Brands and Sponsored Display. Sponsored Brands publishes no 7-day figure at all, only 14 days, so an account that runs Sponsored Brands mixes two windows in any total.
  • Other products count too. Amazon’s default 7-day figure also credits purchases of the advertiser’s other products bought after the click, not only the product in the ad. On one account this blended figure read 9% fewer purchases than the 30-day figure for the same days, and a return on ad spend computed from the advertised product’s own sales undershot all 14 figures of a hand-checked audit.
  • One order, one ad. The last ad clicked gets the whole order. Amazon Marketing Cloud is the one place where we can see the full path of ads a shopper saw and clicked (see below).

​
Orders arrive late

A click’s order can be credited days after the click, so a recent day always shows fewer orders than it will end with.
  • Purchases reach 99% of their final 7-day count 5 to 6 days after the click, on every placement.
  • A day read at age 2 understates its settled 7-day purchases by 8 to 10.5%. Reports frozen at about 48 hours understated settled purchases by 12 to 14%, and sales by 11 to 14%.
  • Clicks and spend barely move after the day ends (at most 1%), so a young day’s cost is right while its orders and sales are low. Its conversion rate and return on ad spend therefore read low.
This is why we let each day’s numbers settle, and correct for orders that arrive late, before judging results.
Never compare a young day’s conversion rate or return on ad spend with a settled day’s. The young day will look worse even when nothing changed.

​
Spend is restated, sometimes below zero

Amazon restates its hourly traffic after the fact. Over 35 days on our accounts, the corrections netted to about −1% of spend. The net is the right number, so we sum corrections rather than drop them. Before an account subscribes to Amazon’s hourly stream, Amazon can still send scattered corrections for earlier days, and they are negative. On one account the stream summed to a small negative total in a month with real spend. So a day with no data in the stream means we have no data, never that the account spent nothing.

​
Ratios over a period

Every ratio we show over a period (a week, a month, a group of products) is computed from the period’s totals: total ad spend ÷ total ad sales for ACOS, total ad sales ÷ total ad spend for ROAS, total spend ÷ total clicks for CPC. We never average daily or per-product ratios, which would let a quiet day or a small product weigh as much as a busy one.
  • Spend with no sales. Spend over zero ad sales shows as an infinite ACOS, and spend over zero total sales as an infinite TACOS, not as a blank: those are the products wasting the most money.
  • TACOS mixes two feeds. Its spend comes from the ads data and its total sales from the Selling Partner API, which arrive on different schedules. Our totals divide sums over only the days that have both, so a day still missing its sales does not inflate TACOS.
  • Change against the previous period is (current − previous) ÷ previous for amounts. For a metric that is already a percentage, such as TACOS, we show the change in percentage points, so a move from 10% to 12% reads +2.0 pt, not +20%.

​
The data we read

​
Accounts and registrations

The accounts a customer connects, and the registrations Laurence holds as an Amazon partner:

​
What we have measured