Case study
10Beasts case study: comparison table tests and 66.3% more ad fees on the winning pages
This 10Beasts case study follows a full CRO campaign on a well-known Amazon affiliate site, from the first audit to the last test. The first round of comparison table tests won on some pages and lost on others. Then a question about traffic forced a pivot, and the second round tested which products the tables recommended.
10Beasts case study at a glance
| Client | 10Beasts, an Amazon affiliate site reviewing tech products |
|---|---|
| Vertical | Affiliate |
| Goal | More Amazon ad fees (commissions) from the existing traffic |
| Pages worked on | The money pages that earned the most, tracked with a separate Amazon tracking ID each |
| Test 1 | New comparison table on 10 money pages, 21 December to 4 January, VWO split test |
| Test 2 | Tables rebuilt from Amazon sales data on 6 money pages, 20 March to 18 April |
| Headline metric | Ad fees on the winning pages (not clicks, not all pages) |
+66.3%
Ad fees, winning pages only, test 2, 20 March to 18 April
Source: 10Beasts, Amazon Associates report, split test on 6 money pages, 4 winners rolled out
+55.1%
Ad fees, winning pages only, test 1, 21 December to 4 January
Source: 10Beasts, Amazon Associates report, VWO split test on 10 money pages, winners rolled out
Who is 10Beasts, and how did this campaign start?
10Beasts is an Amazon affiliate site that reviews tech products and earns a commission, which Amazon calls ad fees, when a visitor it sends to Amazon buys. It became well known among affiliate SEOs after Glen Allsopp featured it on Gaps. By the time Convertica took it on as a client in December 2018, it was getting hundreds of thousands of visitors a month.
The campaign began on a stage. In November 2018, Kurt Philip used 10Beasts as an example in a talk on CRO at an SEO conference, to show that even famous authority sites are not always laid out for conversions and are not always the ones to copy. Someone who worked for 10Beasts was in the room. A month later 10Beasts was a client.
This is the whole campaign, including the parts that did not go to plan. A site with this much traffic should have been a quick win: the more traffic, the faster a split test reaches an answer. It took about four months instead of the one to two that Convertica expected, and the detour taught more than the wins did.
What was holding the money pages back?
The 10Beasts money pages had two problems at the start of the campaign: nobody knew exactly which pages earned the most, and the comparison tables on those pages showed no product images and did little to help a buyer choose. A third problem, the products the tables recommended, only surfaced halfway through.
1. Guessing which pages earn
Many affiliate site owners use one Amazon tracking ID for every page, then guess which pages bring in the revenue. That guess is often wrong, and it is a poor place to start a CRO campaign, because the first pages you optimize should be the ones that already earn. 10Beasts had about 50 pages by then. Convertica picked the top 10 by revenue: the 80/20 rule, applied to money pages.
2. A comparison table that listed specs, not reasons to buy
The original table was a text grid of product names, weights, sensors and button counts, with a "Check Price" button on every row. It had no images. When people cannot touch a product, a picture does part of the work a store shelf would do, and the table gave a buyer no hint which product suited them.
How do you find which affiliate pages make the most money?
To find which affiliate pages make the most money, give each money page its own Amazon Associates tracking ID, use that ID in every Amazon link on that page, and wait a month or two, depending on your traffic. The Associates reports then show earnings per tracking ID, so you can rank your pages by revenue instead of guessing.
- Sign in to your Amazon Associates account and open the tracking ID manager (in 2019 it sat under your account name, as "Manage Your Tracking IDs").
- Add a new tracking ID for each money page.
- Change the Amazon links on each page to use that page's tracking ID.
- Wait one to two months, depending on traffic, then rank the pages by earnings and start with the top earners.
Amazon's interface has changed since this campaign, but the tracking ID approach still works the same way. The same IDs later made both split tests measurable in revenue, not only in clicks.
What was the hypothesis for the first test?
The hypothesis behind the first 10Beasts test was that a comparison table built to help each kind of buyer choose would send visitors to Amazon already decided, so more of their clicks would turn into orders. The table's job is to answer the buyer's questions before Amazon does: which product is for me, what does it do, and why this one.
On affiliate sites Convertica almost always tests the comparison table first, because that is where the buying decision happens. The new table had four ingredients:
- Product images. A picture creates the connection to the product that a text grid cannot.
- A label for each kind of buyer. A ribbon on each image, such as "Affordable", "Good for sport" or "Modern with lots of features", so budget buyers and feature hunters each find their product fast.
- A clear call to action that stands out. "Check on Amazon" on a bold button. Other texts worth testing include "Check price on Amazon" and "Buy now at Amazon", as long as Amazon's rules allow them.
- Features and benefits. Enough detail to educate the buyer, without sending them into the full review.
What changed in the first comparison table?
The first 10Beasts variation turned the text grid into a visual table: an image for each product with a buyer label in the corner, the product name, two or three key details, and a red "Check on Amazon" button. It looked cleaner, was easier to scan, and told each kind of buyer where to look.
What happened in the first split test?
The first 10Beasts split test ran in VWO on the 10 top-earning money pages from 21 December and reached statistical significance on 4 January. The new table raised revenue on six of the ten pages and lowered it on four. On one page, ordered revenue rose 677%.
Stop on that for a moment, because it is the lesson every CRO team relearns: the same hypothesis can win on some pages and lose on others, on the same site, in the same weeks. If that happens within one site, a winning table copied from someone else's site is a guess on yours. Test it.
Rolling out the new table only on the pages where it won gave these results for those pages:
+55.1%
Ad fees, winning pages only, Test 1, 21 December to 4 January
Source: 10Beasts, Amazon Associates report, VWO split test on 10 money pages, winners rolled out
+30%
Orders, winning pages only, Test 1, 21 December to 4 January
Source: 10Beasts, Amazon Associates report, VWO split test on 10 money pages, winners rolled out
Look at the clicks. On the winning pages they rose only 6.8%, while orders rose 30% and ad fees 55.1%. The new table barely changed how many people clicked through to Amazon. It changed who clicked: visitors who already knew which product they wanted. That is the point of a comparison table. It is not only there to raise click-through rate. Its job is to educate the buyer so the click arrives at Amazon ready to buy, inside Amazon's referral window. Commission is clicks × conversion rate × order value × commission rate: the affiliate commission calculator lets you change each one and see the commission on your own numbers.
Why did traffic double while revenue stayed flat?
Before the second test could run, 10Beasts asked a fair question. Google Analytics showed organic users up 92.12% for 1 to 19 January 2019 against the same days of 2018, roughly double. Revenue had not risen anywhere near as much. Why not?
A split test cannot answer that question. It compares the original page and a new page at the same time, on the same traffic. A comparison between this year and last year mixes in everything else that changed: the season, which products were in demand, the traffic mix, even how used to the page returning visitors had become. The second test was cancelled, and the team went digging through a year of Amazon earnings and Analytics data instead.
The answer was one page. The smartwatches page had been 1.24% of the site's traffic in 2018 and was 44.47% of it in 2019, and according to the Amazon earnings it converted at only 1 to 2%. Looking closer at that page:
- Nobody bought the product the page recommended as the best. Not a single sale.
- One recommended product linked to an Amazon listing sold only by third-party sellers.
- Another recommended product sold well but had a high return rate.
- The first two products in the table got many clicks and no sales.
Several other pages had the same kind of problem. The lesson: you can build the best table in the world, and one that has won on many sites, but if the products in it are wrong, fix that first. Which products you recommend can matter as much to your earnings as how you present them.
How did the campaign pivot?
The 10Beasts campaign pivoted from table design to the products themselves. Before testing layout again, the team set three priorities for each money page:
- Update the list of recommended products and remove the outdated ones.
- Reorder the products so premium or popular items sit at the top.
- Add features and benefits to the table to educate the buyer further.
Deciding which products to add, drop and move came from a source many affiliate site owners never open: their own Amazon reports.
How do you use Amazon reports to choose products for a comparison table?
To use Amazon reports for a comparison table, put a unique tracking ID on each top money page, collect a month or two of data, then download the earnings report for that ID and group the products sold. The report shows what visitors from that page actually bought, which is often not what the page recommends.
Tag your top pages
Give each of your top 10 money pages its own tracking ID and collect data for one to two months, depending on traffic.
Download the earnings report
Download the Amazon report data for that tracking ID and open the fee earnings tab (named "Fee-Earnings" in 2019).
Sort by name or ASIN
Sort by product name if your products come in variations such as color or size, so the variations sit together; otherwise sort by ASIN.
Group the same products
Group the variations of each product so you can see which products, not which colors, sold the most.
| If the report shows | Test this | Watch out for |
|---|---|---|
| Your second or third product outsells your first | A split test that only reorders the products in the table | One good week is not a pattern; use a month or more of data |
| Premium products, which pay the most ad fees, sell the most | A table with the premium product first, against the current table | Do not move a premium product up only because it pays more. Do it when the sales data backs it |
| A product you do not review sells well | Add it to the table, with a new review section or a "Read review" link to a separate review page | If you worry about changing the page's content for SEO, use the link |
| A product gets a lot of returns | Take it off the list | Judge returns over a few months of data, not one bad week |
What changed in the second test?
The second 10Beasts test ran from 20 March to 18 April on six money pages, with tables rebuilt from the Amazon report insights: updated products, the strongest sellers first, a "Top pick" label, and features and benefits for each product. Each version had its own tracking ID, so the results came straight from Amazon's sales data.
Four of the six pages won. Rolling out the new tables on those four pages gave these results for them:
+66.3%
Ad fees, winning pages only, Test 2, 20 March to 18 April
Source: 10Beasts, Amazon Associates report, split test on 6 money pages, 4 winners rolled out
+61.4%
Revenue, winning pages only, Test 2, 20 March to 18 April
Source: 10Beasts, Amazon Associates report, split test on 6 money pages, 4 winners rolled out
+23.4%
Orders, winning pages only, Test 2, 20 March to 18 April
Source: 10Beasts, Amazon Associates report, split test on 6 money pages, 4 winners rolled out
Clicks fell slightly, yet orders, revenue and ad fees rose. Across all six pages, losers included, the report shows ad fees up 43.7%. The winners-only figures are what the site earned after keeping the new table only where it won, which is how a test program should be run, but they are not what the change did everywhere.
Why did it work?
Both 10Beasts tests most likely worked for the same reason: the table did more of the buyer's thinking before the click. In test 1 that meant images, buyer labels and benefits. In test 2 it meant recommending the products people were actually buying, in the order they bought them. In both tests clicks barely moved while orders and ad fees rose, which is what you would expect if the change sent better-prepared buyers to Amazon rather than more of them.
The case study cannot say which single element mattered most, because each test changed several things in the table at once. It also covers one site in 2018 and 2019. The products, the tables and Amazon's pages have all changed since. As Kurt Philip later replied to a reader who noticed the site's newer star-rating tables, different pages often produce totally different results for the same table layout, so the answer is always to test.
What can you apply to your own affiliate site?
You can apply the 10Beasts method to any affiliate site with money pages: find the pages that earn, make the comparison table do the selling, check that the products in it are the ones people buy, and split test every change against the current page.
- Track every money page separately. One Amazon tracking ID per page, so you know which pages earn and can measure tests in revenue.
- Start with the 20% of pages that earn the most. Changes there pay back fastest.
- Test the comparison table first. Images, a label for each kind of buyer, a clear call to action, features and benefits.
- Judge tests on orders and ad fees, not clicks. Both tests here barely moved clicks and still raised earnings.
- Expect some pages to lose. Roll out winners page by page, and keep the original where the new version lost.
- Read your Amazon reports. Reorder, add and remove products based on what visitors buy, and drop products with high return rates.
- Do not judge a change by comparing this year with last year. Split test it, so both versions see the same traffic at the same time.
Add one check that did not exist when 10Beasts ran these tests: can an AI assistant read your comparison table? AI search answers to "best" questions pull from pages they can parse. A table built in real HTML, with product names, the criteria you judged on and a clear top pick in text, is easier for an assistant to read and cite than the same table drawn as an image. AI agent readiness is one of the eight checks in Convertica's free CRO audit.
For the testing setup Convertica used on affiliate sites, with three pages and a tracking ID for each, read why Amazon affiliate split tests fail. For another comparison table test, read the comparison table case study, and for an earlier affiliate campaign built on the same 80/20 approach, read the affiliate case study for Robert's site. To plan your own tests, the A/B test sample size calculator shows how much traffic a test needs, and the guide to affiliate marketing conversion rates explains how to measure yours.
10Beasts case study FAQ
What did the 10Beasts case study test?
It tested the comparison tables on 10Beasts' Amazon affiliate money pages. Test 1 replaced a text-only table with one that showed product images, labels for different kinds of buyer, features and benefits, and a clear call to action. Test 2 changed which products the tables recommended and in what order, based on the site's Amazon sales reports.
How much did ad fees increase in the 10Beasts case study?
Ad fees on the winning pages rose 55.1% in test 1 (21 December to 4 January) and 66.3% in test 2 (20 March to 18 April). Both figures count only the pages where the new table won and was rolled out. Across all six pages in test 2, losers included, the report shows ad fees up 43.7%.
Why did the same table win on some pages and lose on others?
Because each page sells different products to different buyers. In test 1 the new table raised revenue on six of ten pages and lowered it on four, on the same site, in the same weeks. A layout that wins on one page is a hypothesis for the next page, not a rule. Test it there too.
Why did traffic double but revenue stay flat?
Most of the new traffic went to one page, the smartwatches page, which converted poorly. Its share of the site's traffic rose from 1.24% to 44.47%, but buyers did not purchase its top recommendation, one product was sold only by third-party sellers and another had a high return rate.
Can you compare this year's revenue with last year's to judge a change?
Not reliably. A year-over-year comparison mixes your change with seasonality, product demand, traffic mix and how long visitors have seen the page. A split test runs the original and the new version at the same time on the same traffic, so the difference between them can be put down to the change.
How do I find which affiliate pages make the most money?
Give each money page its own Amazon Associates tracking ID and use it in that page's links. After a month or two, depending on traffic, the Associates reports show orders and earnings per tracking ID, so you know which pages earn instead of guessing. Start CRO work on the top earners.
What should an affiliate comparison table include?
Product images, a label that tells each kind of buyer which product suits them, the key features and benefits, and one clear call to action per product that Amazon's rules allow. Then use your sales reports to decide which products to list and in what order, and test the result.
Can Convertica run this kind of campaign on my affiliate site?
Convertica starts with the free CRO audit, an app that checks your page for people and for AI agents while you watch. After it, if you want the fixes built for you, the one page rebuild is $497 and the full site rebuild, up to 10 pages, is $1,997: Convertica's team builds every fix the audit found.