Same publisher, same window: ~11.7% conversion on Amazon vs ~1.3% on the brand's own paid traffic

A brand running both a marketplace listing and its own store is running two businesses with two cost structures and usually one profit and loss statement. The published figures make the difference stark. Amazon Ads converts at 11.7% with a $13.98 cost per acquisition and a 3.08 return on ad spend. Paid traffic to brands' own sites converts at 1.3% with a $23.20 cost per acquisition and a median order value of $61.22. The comparison is worth making because the publisher and the twelve-month window are the same on both sides, which is rarer than it sounds and is the only reason the two numbers can sit in the same sentence.

Six times the conversion rate at roughly half the acquisition cost — and the cost per thousand impressions on that channel rose 39% in a year.

1. 01

   The gap is intent, and intent has a price attached

   A shopper searching on a marketplace has already decided to buy something in the category and is choosing between listings. A click on a brand's own site is frequently the first contact with the brand. The marketplace is not converting better because the page is better, it is converting better because the decision was made before the impression was served. That advantage is bought, not earned, and the price of it is rising: cost per thousand impressions up 39% year on year.

2. 02

   Return on ad spend spreads three-fold across categories while acquisition cost barely moves

   Automotive returns 3.27 and health and wellness 1.55, a 2.1x spread. Cost per acquisition across the same categories runs $49 to $37.85, a 1.28x spread. The variable doing the work is order value — $116.34 in automotive against $60.12 in pets — not media efficiency. Automotive achieves the highest return in the set on the lowest conversion rate in the set, at 1.59%.

3. 03

   The category ranking flips depending on which number you optimise

   Food and beverage has the highest conversion rate of the measured categories at 2.60% and one of the lowest returns at 2.03, because its order value is $63. A brand told to improve conversion and a brand told to improve return on ad spend will build two different businesses from the same starting point.

4. 04

   Flows are 5.3% of sends and about 41% of email revenue

   Automations run a 4.66% click rate and a 1.49% conversion rate against 0.74% and 0.08% for campaigns, and $3.41 revenue per email against $0.155 — a 22-fold difference per send. Revenue per recipient on flows runs roughly eighteen times the campaign figure. The split is the clearest allocation signal in the whole ecommerce benchmark set.

5. 05

   The number everyone quotes for email's share of store revenue does not have a source

   The circulating claim that email drives 30 to 40% of total store revenue traces back to a single agency's own portfolio of about fifty brands, not to an independent study. There is no credible primary source for it. We do not use it, we do not model against it, and a plan whose case rests on it is resting on one agency's client list.

Amazon Ads conversion **~11.7%** vs off-Amazon ecommerce **~1.3%** (same-publisher comparison; updated from the older 11.7% / 1.3% pairing). Amazon avg CPC ~$1.22, ACoS ~30.7%.

The model puts margin before channel. Phase one builds a contribution view per channel: landed cost, marketplace take rate, fulfilment, media and returns, so the marketplace and the owned store are compared on what is left rather than on conversion rate. Phase two allocates media against that view instead of against blended return on ad spend, which systematically favours whichever channel happens to carry the higher order value. Phase three builds the lifecycle layer, on the basis that the owned store is where a repeat relationship is possible and the marketplace is where it is structurally not. Reporting separates first orders from repeat orders throughout, because a blended acquisition cost in a two-channel business describes neither channel.

1. 1

   Weeks 1–3

   Margin model

   Landed cost, take rate, fulfilment, returns and media assembled into one contribution view per channel

   Owner: OmniFlow + client finance

2. 2

   Weeks 2–4

   Channel baseline

   Conversion, acquisition cost and order value measured per channel on the brand's own data

   Owner: OmniFlow

3. 3

   Weeks 4–8

   Reallocation

   Media rebalanced against contribution, not against blended return on ad spend

   Owner: OmniFlow

4. 4

   Weeks 6–12

   Lifecycle layer

   Flows built and reported separately from campaigns, revenue per recipient tracked per flow

   Owner: OmniFlow

5. 5

   Month 6

   Review

   Contribution per channel, first-order versus repeat-order acquisition cost, flow share of email revenue

   Owner: OmniFlow

The model reports contribution per channel and never blends the two. At published rates a marketplace order converts roughly six times more often at roughly 60% of the acquisition cost, and arrives carrying a take rate the owned store does not pay and a customer relationship the brand does not own. Those are two different trades and the arithmetic decides them separately per category, because the published conversion gap is widest exactly where order value is lowest. Every benchmark here is replaced by the brand's own figures once the margin model exists, which is usually inside three weeks.

Modelled. Inputs: Triple Whale, Amazon Ads Benchmarks 2026 (2,800+ brands); Triple Whale, Ecommerce Benchmarks 2026 (53,000+ brands, August 2025 – July 2026); Klaviyo, 2026 Email Marketing Benchmarks (183,000+ customers); Omnisend, 2026 Ecommerce Marketing Report (20B+ campaign emails, 27,000+ brands). Modelled outputs are not a forecast or a guarantee of results. Cited: Ad Badger *Amazon Advertising Stats 2026*. Figures refreshed 2026 against Ad Badger 2026.

Every figure below is published research, not a client result. They are the inputs to the arithmetic above, listed so it can be checked rather than taken on trust. The bracketed number points to the full citation at the end of this page.

These are the surfaces this engagement is run and measured from, shown with representative figures built around the benchmarks cited on this page. Every account we run reports into views like these, and you keep ownership of all of them.

Each figure on this page, the system it is read from, and the definition and window it is measured over.

Not applicable — this is a modelled engagement. Its weakest input is that both sides of the headline comparison come from a self-selected panel: brands that installed one analytics tool, which skews toward a particular size and a particular direct-to-consumer shape. Same publisher and same window remove the definitional problem, not the sampling one, and the Amazon and own-site populations are not necessarily the same brands. The margin model exists to make the comparison on the brand's own numbers before any budget moves.

A modelled capability, not a client account. Figures illustrate what the model produces and are labelled as modelled wherever they appear. See all our work

Full citations for everything cited on this page, with the sample and period each study covers, so you can go and read the original.