How Amazon Calculates New-to-Brand, and Its Blind Spots

The calculation, as Amazon defines it

Amazon's own advertising guidance states plainly that new-to-brand metrics identify "customers who purchase your brand or product for the first time on Amazon during the last year." Under the hood, every ad-attributed order is checked against that customer's Amazon purchase history for the brand; if nothing shows up in the trailing 12 months, the order is flagged new-to-brand. It's evaluated per order, at the moment the order completes, against a rolling window rather than a fixed calendar period.

That's the whole mechanism. There's no separate model, no machine-learning layer weighing probability — it's a lookback rule against a purchase ledger Amazon already holds because the transaction happened on its own marketplace. That simplicity is also exactly why the blind spots below exist: a rule this literal can only see what's inside the ledger it's checking.

Blind spot one: the window is shorter than a lot of real repeat cycles

Twelve months is Amazon's default, but it's genuinely too short for durable goods, seasonal categories, and slow-consumable products where a real customer might reasonably return in month 14 or 16. Amazon has addressed this at the infrastructure level rather than changing the default: inside Amazon Marketing Cloud, the Amazon Retail Purchases dataset extends the purchase-history lookback used for new-to-brand and lifetime-value calculations from 13 months out to 5 years. It's a paid feature — available at a published rate, with a free trial period, though audience-building use cases are free — and it exists specifically because the standard window undercounts genuine repeat behaviour in these categories. If your category has a real repeat cycle longer than a year, the standard NTB number is quietly counting some returning customers as brand-new, and the 5-year dataset is the documented fix Amazon itself built for that gap.

Blind spot two: it can't see your own site at all

The metric only checks Amazon's own transaction history. A customer who has bought from your Shopify store loyally for three years, and buys your Amazon listing for the very first time this month, registers as 100% new-to-brand — because from Amazon's side, they are new, even though they're one of your best existing customers by any other definition. This isn't a bug; Amazon Attribution and the sponsored-ads reporting stack were never built to see off-Amazon purchase history, and there's no cross-referencing mechanism that would let them. If a meaningful share of your Amazon buyers are converted DTC loyalists rather than genuinely new people, your reported new-to-brand rate is overstating real customer growth by exactly that share, and no setting inside Amazon Ads will correct for it — the correction has to come from your own CRM.

Blind spot three: it's an order count, not a person count

Because the rule resets on a rolling 12-month basis rather than checking a lifetime record, the same individual can register as new-to-brand more than once — once now, and again in a future year after their prior purchase ages out of the window. There's no native "unique new customer" metric that corrects for this; new-to-brand orders and new-to-brand sales are both order-level counts. For most fast-repeat categories the practical effect is small, because genuine repeat buyers return well inside 12 months and never re-qualify. For slower categories it compounds with blind spot one — the same structural gap, showing up twice.

The common mistake, including ours

The mistake is treating new-to-brand as a precise, audited number rather than a rule-based proxy with known edges. We've quoted a client's new-to-brand percentage in a growth summary without noting that a meaningful share of their Amazon buyer base was also a long-standing DTC customer list we'd separately confirmed existed — the number wasn't wrong by Amazon's definition, but presented alone it implied more net-new customer growth than the account was actually producing. Once we cross-referenced the two customer lists, the real net-new figure was meaningfully lower, and that's the number that belonged in the growth conversation.

How the three blind spots interact

These three limitations aren't independent — they compound in a specific, predictable direction for a brand that sells DTC in a slow-repeat category. The window is too short for the category (blind spot one), so genuine returning customers register as new; the metric can't see the DTC purchase history that would otherwise flag them as existing customers (blind spot two); and because it's counted at order level, a single loyal customer can trip the false-new flag more than once across different years (blind spot three). None of the three is large on its own for most brands. Together, in the wrong category, they can push a reported new-to-brand rate meaningfully above the real net-new customer growth rate — which is exactly the gap worth checking before the number goes into a forecast.

What to do when the blind spots actually matter to your business

Three checks, roughly in order of effort. First, if you sell DTC as well, cross-reference your Amazon buyer list against your CRM's email or name match where legally and practically possible, even a rough estimate of overlap tells you how much to discount the headline NTB number. Second, if your category has a repeat cycle longer than 12 months, evaluate whether the AMC extended-lookback dataset is worth the monthly cost against the size of your ad spend — for a mid-size DSP account, resolving even a few points of NTB miscount can be worth more than the dataset's price. Third, stop reporting the blended NTB number without a one-line caveat about lookback window and off-Amazon visibility whenever it's feeding a decision bigger than a weekly trend check — a board deck, a CAC model, a channel-mix decision.