Split Testing Offers and Landing Pages: A/B Testing Without Fooling Yourself

The same audience reacts differently to different offers, pre-landers and even button colors. Neither a beginner nor a seasoned media buyer can guess in advance what will work. That is why every campaign sooner or later comes to split testing: traffic is divided between variants, and numbers make the call instead of intuition.

What is split testing in affiliate marketing

A split test (also called an A/B test) shows several variants at the same time on the same traffic and then compares the results. The key words are «at the same time». If you run offer A for a week and offer B the following week, you are comparing two different weeks, not offers: a different auction, different creatives, a different audience mood.

Affiliates split test three levels:

1. Offer: two networks with the same product, two products in one vertical, or the same geo version of an offer in different networks.
2. Pre-lander: a quiz against an article, a review against a «success story», short copy against long.
3. Landing page: different lead forms, prices on the page, trust blocks.

Creatives are usually tested on the ad platform side, which has its own tools for that. Everything that happens after the click is the tracker's job. If you have not yet looked into why you need one, start with affiliate tracker.

What to compare: CR, EPC or ROI

The most common split testing mistake is comparing the wrong metric. Each test level has its own target metric.

Why EPC and not conversion rate? Say offer A converts at 4% with a $20 payout, and offer B at 3% with a $30 payout. A wins on CR, but per hundred clicks A brings $80 and B brings $90. If B's approval rate is half as high, the picture flips again. All these metrics are covered in affiliate marketing metrics: ROI, EPC, CR.

Tip. Judge offers by approved money, not by leads. A lead on hold is a promise, not income. How statuses affect the numbers is explained in conversion statuses: hold, approved, trash.

How weighted rotation works

Technically, a split test is rotation: each click is sent to one of the variants. There are two main ways to split traffic.

Weighted rotation

Each variant gets a share in percent: 50/50, 70/30, 34/33/33. For every click a variant is picked at random with that probability. On large volume the shares converge to the target; on small volume they drift. If 20 clicks came in during an hour, the split may well be 13 to 7, and that is not a tracker bug but the nature of randomness.

Unequal weights are useful when one variant is proven and the other is an experiment. Give the newcomer 20–30% of traffic: if it fails, you lose little.

Priority

All traffic goes to the first enabled variant, and the rest stay in reserve. This is not a split in the strict sense but insurance: if the main offer is stopped or switched off, traffic keeps converting on the backup.

Visitor pinning

If someone returns from the ad a second time, it is better to show them the same variant as the first time. Otherwise they see a different price or product, get confused, and your statistics get mixed up. Trackers handle this with visitor pinning for a set period: anyone returning within the window lands on the same pre-lander and the same offer.

How long to wait for split test results

This is the key question, and it is where most budget burns. Concluding «variant B is better» after 30 clicks and two leads is a coin toss.

Here is how to reason in practice:

• Count conversions, not clicks. A hundred clicks without a single conversion in either variant tells you nothing. Look at the number of target actions in each variant.
• The smaller the gap, the bigger the sample. A twofold gap shows up quickly. A 10% difference needs an order of magnitude more data.
• Capture a full cycle. At least several full days, ideally a week: weekday and weekend traffic behave differently.
• Do not peek and decide. If you look at the chart every two hours and stop the test at the first lead, you will systematically pick random winners.

Statistics uses the notion of confidence level for this: how likely it is that the difference is not random. The standard is 95%. That means, all else being equal, you will be wrong no more than once in twenty.

Tip. Decide in advance how much traffic you are willing to spend on the test, and do not stop it earlier. If the budget runs out without confidence, the honest conclusion is «no visible difference», not «whichever is ahead right now wins».

Split testing pre-landers: what to watch separately

With a pre-lander and a landing page, the test has two stages: the visitor first sees the pre-lander, then clicks the button and goes to the offer. So a pre-lander has two metrics:

1. Pre-lander CTR: the share of visitors who made it to the offer.
2. Final conversion and EPC: how much money the traffic that went through this pre-lander brought.

The typical trap: an aggressive clickbait pre-lander gets a 40% CTR, a calm article gets 15%. But the people who read the article and still clicked through are already «warmed up», and they convert several times better. In money terms, the article may come out ahead.

One more subtlety: when testing a pre-lander, the offer behind it must stay the same, and when testing an offer, the pre-lander must stay the same. Otherwise you change two variables at once and will not know what worked.

Clean traffic is a condition for a fair test

A split test compares how people react. If bots are mixed into the traffic, they are spread across variants at random and dilute the statistics. One variant may «lose» simply because it caught a burst of data center bots that will never buy anything.

So test on traffic that has already been filtered: filter first, rotate second. How filtering works is explained in how to filter bots out of ad traffic. And for conversions to be tied to variants at all, you need a working postback from the affiliate network; without it you are only comparing clicks.

How to run a split test in ArtisanClo

In ArtisanClo, a split is configured inside one flow, on the «Destinations» step. The flow must run in «Tracker» or «Cloaking + Tracker» mode so that money is counted on the traffic that passes.

Offers and weights

1. Add a second offer (and more) to the flow. Each has its own name for reports, display mode and URL.
2. As soon as there are two or more offers, each gets a Weight field from 1 to 100%. The weights must add up to exactly 100%; the Auto weights button splits them evenly.
3. To stop an offer from getting traffic for a while, switch it off with the toggle instead of setting its weight to 0.

The ad link does not change with the number of offers: the offer is chosen for each visit after all filter checks. That means you can add a variant to a live campaign without editing the ad.

Besides weighted offer rotation there is Priority mode, where all traffic goes to the first enabled offer and the rest are kept in reserve. And visitor pinning, from 0 to 720 hours: a returning visitor gets the same pre-lander and the same offer.

Pre-landers

Pre-landers are added on the same step; with two or more, they also get a weight. A pre-lander leads either «By weight among offers» or to a specific offer. The pre-lander button must point to the {click_url} macro:

https://your-prelander.com/?go={click_url}

If the page has several buttons for different offers, append /2, /3 to the URL to open the second and third offers.

Branches instead of a blind split

Sometimes a test shows that offer A is better on Android and B on iOS. Then the split turns into branches: different visitors are sent to different offers by condition (country, device, OS, language, link parameter, tag). Branches are checked top to bottom, and the first match wins.

Where to see the results

• In Reports, the «Money by slice» table breaks down by offer and pre-lander: clicks, conversions, CR, revenue, spend, profit, ROI. You can also turn on the «To offer», «LP CTR», «On hold», EPC and CPA columns.
• In Flow diagnostics, on the «Auto-tuning» tab, there is an Offer split test block. A verdict on the best offer appears only at 95% confidence or higher and with at least 100 visits per offer: exactly the protection against early conclusions described above.

The full list is on the ArtisanClo features page, and which plans include branches, pre-landers and priority is shown on the pricing page.

Common split testing mistakes

• Testing on different traffic. Variant A ran in one campaign, B in another. Those are two different experiments.
• Changing conditions mid-test. If you changed the creative, bid or geo, reset your conclusions.
• Comparing leads with different approval rates. Wait until the network approves at least part of the leads.
• Too many variants on a small budget. Five offers at 20% each are five small samples, none of which will prove anything.
• The never-ending test. Once you have confidence, switch off the loser: every click on the worse variant is profit left on the table.

Bottom line

A split test is not «throw in two offers and see»; it is discipline: one variable at a time, the same bot-free traffic, comparison by money rather than leads, and patience until statistical confidence. Weighted rotation, visitor pinning and an offer report give you the tools, but the call on whether you have enough data is still yours. Once you have a winner, the next step is to work out how much it really earns; see how to calculate ROI.