Amazon Conversion Rate, and What Good Actually Looks Like

Amazon calls it unit session percentage, which is units ordered divided by sessions, and it is in the business reports rather than anywhere obvious. It is the single most useful number on a listing because everything else feeds into it: the images, the title, the price, the reviews, the availability and the traffic quality all end up expressed here.

It is also the number most often quoted with a benchmark attached, and the benchmarks are close to meaningless.

There is no useful universal figure for a good Amazon conversion rate. It varies several times over between categories, and within one listing by where the traffic comes from: a page fed by shoppers searching the brand's name converts far better than one fed by broad browsing. Amazon reports it in the business reports as unit session percentage, units ordered divided by sessions, for each child ASIN rather than the variation family. The comparison worth making is the listing against itself, this month against last, and a sudden drop usually has a cause outside the listing: a competitor's price moving, advertising starting to bring shoppers who were never going to buy, or a few days out of stock. The sections below cover what moves the rate, roughly in order.

So the useful habit is a monthly note per listing: the rate, the traffic behind it and anything that changed. A rate read against that note says something; a rate read against an industry average says almost nothing.

A figure like ten to fifteen per cent circulates as the Amazon average and it is not wrong so much as unusable. Conversion varies by category by a factor of several: a low cost consumable somebody buys without thinking converts at a rate a considered thousand dollar purchase never will, and neither is doing anything wrong.

It varies by traffic source too, which is the part that gets missed. A listing whose traffic is mostly branded search converts far better than one whose traffic is mostly broad category browsing, and that is a fact about who is arriving rather than about the page they arrive at. Turning on a broad campaign will reliably push the rate down while increasing total sales.

So the comparison worth making is the listing against itself: this month against last, before a change against after. That is the only version of the number where a movement means something specific.

The main image does more than everything else combined, because it is the only element that has to work at thumbnail size in a grid of competitors. Most of the decision to click happens there, and a large share of the decision to buy is made before the shopper reads a word.

Price relative to the alternatives on the same screen is second, and it is relative rather than absolute: a price that was fine last quarter converts worse when a competitor moves, without anything on the listing changing. That is one reason a conversion rate can fall while nothing was touched, and it is why a repricing strategy is a conversion lever rather than only a margin one.

Reviews are third and behave as a threshold rather than a slope. The distance from zero reviews to a handful is enormous. The distance from four hundred to six hundred is close to nothing. Rating matters more than count once the count is credible, and a slip below four stars is felt immediately. What can be done when a bad one lands is its own subject: what you can do about a bad Amazon review.

Availability is fourth and is the one people forget to check. A listing that went out of stock for four days did not have a conversion problem, it had four days of sessions with nothing to buy, and the monthly average carries that for the rest of the month.

A single conversion rate for a listing is an average of several audiences, and the average hides where the change happened. Two splits show most of it.

App against browser. The business reports separate sessions on the Amazon mobile app from sessions in a browser. Most shoppers now arrive on a phone, where the main image is small, the bullets are below the fold and the first few seconds decide everything. A listing that converts well on a desktop and poorly in the app usually has a main image or a first screen that does not work at phone size.

By advertising placement. The placement report shows how ads convert at the top of search results, further down the results, and on other products' pages. Shoppers who click from the top of search have usually searched for something close to the product; those who click from another product's page are comparing. A falling rate that comes entirely from product page placements is a targeting decision, not a listing problem.

By search term. The search term report shows conversion for each term that brought paid traffic. Terms with many clicks and no orders are the ones dragging the rate down, and they say exactly which shoppers the listing is not for.

Branded and unbranded. Shoppers who searched the brand's name convert far better than those who did not. A month where branded searches fell, perhaps after an advertising campaign elsewhere ended, will show a lower rate with nothing wrong on the page.

Read the splits before changing anything on the listing. Often they show the listing is fine and the mix of shoppers changed.

A falling conversion rate is at least as likely to be a traffic story as a listing story, and separating the two is the first diagnostic rather than the last.

Advertising that broadened is the most common cause. A campaign that started picking up loosely related search terms brings sessions that were never going to buy, and the rate falls while the listing is untouched and total sales may even rise. The fix is in the search terms rather than on the page, and negative keywords are the cheapest work available: negative keywords, the cheapest work in Amazon PPC.

Ranking for the wrong thing does the same at a slower pace. A listing that has drifted into relevance for a term describing a different product will accumulate sessions from people who wanted that product, and no image or price change will convert them.

The way to tell them apart is the search terms report, which shows what people actually typed to arrive. Brand analytics shows the same thing from the other direction, for the terms where the brand is competing.

An illustrative case, to show the order of the checks. A listing that converted at 12 percent for months falls to 8 percent over three weeks. Sessions are steady. Nothing on the listing was changed by the brand.

Stock first. Was the product out of stock, or low enough to show a longer delivery date, on any of those days? Here it was not.

The Featured Offer. Did the brand's offer stay in the buying position? The business report's featured offer percentage held steady, so shoppers were seeing the brand's own offer.

Price against the screen. A competitor appearing beside the listing in search had cut its price by a fifth two weeks earlier. That alone could explain part of the fall.

The traffic. The search term report showed a broad match campaign had started picking up searches for a related but different product. Those sessions converted at under 2 percent and made up a quarter of the traffic.

The page. The main image on the live listing had changed. Another seller's contribution to the catalogue had replaced it with a weaker photograph, which the brand had not noticed because nobody had looked at the live page.

The fixes, in order of cost. Negative keywords for the wrong searches, the same day. The main image restored through Brand Registry, the same week. The price reviewed against the competitor's, with the margin in front of the person deciding. Within a fortnight the rate was back above 11 percent.

No single cause explained the drop, which is common. Checking in this order finds the cheap fixes before anybody rewrites a listing that was working.

Change one thing and wait. Conversion rate is noisy at low session counts, and a listing with a few hundred sessions a week needs a fortnight before a movement means anything. Changing the image and the price and the bullets in the same week produces a number nobody can attribute.

Start with the listings that have traffic and convert badly, not the ones that convert badly and have no traffic. A page with two thousand sessions a month at four per cent has more available upside than one with fifty sessions at one per cent, and the second is usually a traffic problem wearing a conversion problem's clothes.

Doing that systematically across a catalogue, rather than on whichever listing somebody happened to look at, is what Amazon account management is largely made of.

Most conversion changes are judged by looking at the rate before and after. That works for large effects and misleads for small ones, because the rate moves week to week for reasons nothing to do with the change.

Use Amazon's experiments where available. Brands enrolled in Brand Registry can test a main image, title, A+ content or bullets through Manage Your Experiments, which shows each version to a share of shoppers at the same time and reports which won. Running both versions together removes the season, the competitors and the advertising from the comparison. Testing a main image with Manage Your Experiments covers how.

Otherwise, a careful before and after. Change one thing, note the date, and compare at least two weeks after with two weeks before, the same days of the week, with nothing else changed in between. Note anything that did change: a price, a campaign, a competitor.

Enough sessions to mean something. A listing with a few hundred sessions a week needs several weeks before a difference of a point or two can be trusted. A listing with thousands can be read sooner.

Decide the measure first. Conversion, sales per session, or contribution per session. Choose before the test, so the result cannot be read in whichever way flatters the change.

Keep the losers on record. A change that did nothing, or made things worse, is worth as much to the next decision as one that worked, provided somebody wrote it down.

It is worth saying plainly, because optimising this number in isolation leads somewhere unhelpful. Conversion rate can always be improved by narrowing the traffic: turn off every campaign except branded search, and the rate rises while the business shrinks. Raise the price to the point where only the most committed buy, and it rises again.

The number that matters is profit, and conversion rate is one of the inputs to it rather than a substitute for it. A change that lifts conversion from six to seven per cent while halving sessions has cost money, and a change that drops it from six to five while tripling sessions has usually made it.

So the useful framing is contribution rather than rate: how much a listing earns after the cost of the goods, the fees and the advertising that brought the sessions. That is the figure a weekly review should open with, and it is the one that stops a team optimising a percentage into a smaller business.

It also changes which listings deserve attention. A page converting at three per cent on a product with sixty per cent margin is worth more work than one converting at twelve on a product that barely clears its fees, and the conversion figures alone would send you to the wrong one. What it actually costs to sell on Amazon is the other half of that arithmetic.