A/B Testing With Tracked Links: Split Traffic, Measure Revenue

At the median 4.2% click-to-paid rate on the TrackRev platform, a link-layer A/B test needs several thousand clicks per variant just to gather a few hundred conversions — which is exactly why measuring these tests on clicks instead of revenue is so tempting and so misleading.

A/B testing with tracked links means splitting one link’s traffic across two or more destinations, then measuring which variant produced more revenue per visitor, not which got more clicks.

The link is the experiment: it rotates visitors between destinations and remembers which one each visitor saw, so the downstream charge can be credited to the right variant — the same revenue lens that shows which channel produces revenue, not just clicks.

This article covers how link-layer testing works, why revenue is the only honest metric for it, and — most importantly — the sample-size discipline that separates a real result from a coin toss.

The 2026 link-tracking benchmarks give the conversion baselines you will use to size a test before you run it.

Why this matters for your revenue

The reason to test at the link layer is that the link is often the cheapest place to change the thing that matters.

You may not control the destination page’s code, or be able to ship a page-level experiment quickly, but you can always point a link at two different destinations and split traffic between them.

That makes the link a fast, low-friction experiment surface for exactly the decisions that move revenue — which offer, which landing page, which pricing framing wins.

But the same accessibility makes link testing dangerous when done carelessly.

Because clicks are abundant and conversions are scarce, it is tempting to declare a winner on click-through rate after a day — and click-through rate is almost useless here, because the variant that gets more clicks routinely earns less money.

A punchier headline can win the click and lose the sale; a page that sets honest expectations can lose the click and win the customer.

Optimise on clicks and you will systematically pick the variant that is better at being interesting rather than the one that is better at converting.

Worse, calling a revenue result too early — on a handful of conversions — means acting on noise, and noise will happily point you at the wrong variant with total confidence.

Getting link testing right is therefore two disciplines at once: measuring the right outcome, and waiting for enough of it. Both cost nothing but patience, and both are routinely skipped.

What A/B testing at the link layer is

Link-layer A/B testing uses the tracked link itself as the randomiser. One link, two or more destinations, traffic split between them, and each visitor’s assignment remembered so their eventual conversion is credited correctly.

Rotating destinations behind one link

The mechanism is a link that resolves to different destinations for different visitors, splitting traffic by whatever ratio you set — usually evenly.

When a visitor clicks, the redirect assigns them to a variant, records the assignment against their visitor ID, and sends them on.

Every subsequent event from that visitor — session, signup, charge — is tied back to the variant they were assigned, so the revenue comparison is clean. The visitor experiences an ordinary link; the split happens invisibly at the redirect.

Why the link layer, not the page

Page-level A/B tools change what a single page shows; link-level testing changes which page the visitor reaches.

The link layer wins when the variants are different destinations (two landing pages, a pricing page versus a case study, two entirely different offers) or when you cannot easily deploy a page experiment.

It is also the only option when you do not control the destination — an affiliate sending traffic to two different merchant pages, for instance.

The two approaches are complementary: use page tools to test elements on a page, and link tests to choose between pages.

When you do not control the destination page

Link-layer testing is uniquely suited to cases where editing the destination is off the table.

If you are driving traffic to a partner’s page, an app store, or a marketing site owned by another team, you cannot install a page-level experiment — but you can still split your link between two destination URLs and measure which earns more.

This makes link testing the practical experiment layer for affiliates, partners, and anyone whose conversions happen on pages they do not own.

Splitting across more than two destinations

Nothing restricts a link test to two variants — you can split across three or more destinations (an A/B/n test), and sometimes you should.

The catch is arithmetic: every extra variant divides your traffic further, so each one accumulates conversions more slowly and the whole test takes proportionally longer to reach a trustworthy sample.

With scarce conversions, three variants need roughly half again the total volume of two to give each the same confidence.

The practical guidance is to test two strong contenders rather than five weak ones — a shorter, better-powered test beats a sprawling one you can never conclude.

Measuring to revenue, not clicks

The metric choice is the whole game. Link tests fail most often not because the mechanism is wrong but because the wrong number is used to pick the winner.

Why clicks and CTR mislead here

Click-through rate measures curiosity, not value. The variant that wins on clicks is the one that was better at getting tapped, which is a function of the copy, the promise, and the placement — none of which guarantees the visitor buys.

A test judged on CTR reliably promotes the most clickable variant, and the most clickable variant is frequently the one that overpromises and under-converts.

Using clicks to pick a revenue winner is not a small approximation; it is measuring a different thing and hoping it correlates.

Revenue per visitor as the metric

The honest metric is revenue per visitor (or per click): total attributed revenue for a variant divided by how many visitors it received.

It folds click-through, conversion rate, and order value into one number that answers the only question that matters — which variant made more money per person sent to it.

Because it is denominated in revenue, it is also far harder to fool with vanity engagement. A variant can win on clicks and still lose on revenue per visitor, and when they disagree, revenue per visitor is right.

Click-to-paid versus raw conversion

Even within conversion metrics, be precise about which conversion you mean. Click-to-signup and click-to-paid can point at different winners: a variant might drive more free signups but fewer paying customers, or vice versa.

For a revenue decision, click-to-paid — a click that becomes a paid charge within the attribution window — is the metric that counts.

Signups are a leading indicator worth watching, but do not let a signup-rate win override a paid-rate loss; the money is in the paid rate.

Significance and sample size

This is the section most link-testing guides skip, and it is the one that decides whether your result means anything. The honest truth is that most link tests never reach the volume needed for a confident answer.

Why small samples lie

With conversions this scarce, small samples are dominated by chance.

If a variant has had a few dozen visitors and two conversions, a single extra sale would swing its measured rate enormously — the number is mostly noise, and noise looks exactly like signal until more data arrives.

This is why a test that appears to have a clear winner after a day so often reverses after a week: the early lead was random.

The scarcer your conversions, the more visitors each variant needs before its measured rate stabilises, and at a 4.2% median click-to-paid rate, conversions are scarce.

The peeking problem

Checking a test repeatedly and stopping the moment it looks significant is a reliable way to fool yourself.

If you keep peeking and declare a winner the first time the numbers cross a threshold, you will call winners that are pure noise, because random fluctuation will eventually cross any threshold if you look often enough.

The discipline is to decide the sample size in advance and read the result once, at that size — not to watch it live and pounce. Continuous peeking turns a rigorous test into a slot machine.

How long a link test should run

Sample size, not calendar time, decides when a link test is done — but calendar time still matters for a separate reason.

Even after you have enough conversions, you should run a test across at least a full weekly cycle, because traffic and buyer behaviour differ by day of week and a test that spans only weekdays or only a weekend can bake a timing artefact into the result.

So the honest stopping rule has two conditions: enough conversions per variant to trust the difference, and enough elapsed time to cover the natural cycles of your audience.

Meet the sample floor first, then make sure you have not sampled a biased slice of the week.

What to measure: clicks versus revenue

The table contrasts the metrics you might use to call a link test and what each one actually tells you. Only the bottom rows should decide a revenue test.

Reading an example test

The table shows a realistic readout where clicks and revenue disagree — the common case. Variant B is more clickable; Variant A makes more money per visitor. A CTR-based call picks B and loses revenue.

When link A/B testing is not worth it

Testing is a tool with a cost — mostly time and traffic — and it is honest to say when it does not pay for itself.

Too little traffic to conclude

If a link will never receive enough clicks to accumulate a meaningful number of conversions per variant, an A/B test on it cannot produce a trustworthy answer, and running one just delays the decision while dressing a guess in the costume of data.

For low-traffic links, you are usually better off applying a strong prior — send traffic to the pricing page, use the branded link, follow the benchmarks — than splitting scarce clicks between variants you can never distinguish.

Test where you have volume; decide by principle where you do not.

How TrackRev supports link testing

TrackRev tests at the link layer and reads results on revenue, because the same platform owns the split, the click, and the charge.

One pipeline from split to charge

A TrackRev tracked link can rotate destinations and record each visitor’s assignment, and because the same first-party pixel and Stripe connection power revenue attribution, each variant’s revenue per visitor is computed from real charges rather than click estimates.

The split, the assignment, and the conversion live in one pipeline, so there is no join between an experiment tool and a revenue tool that could disagree about which variant a customer belonged to.

You read the winner on money, from the same data model that runs the rest of your tracking.

When NOT to use TrackRev

If your links do not carry enough traffic to reach a meaningful sample per variant, link A/B testing will not give you a trustworthy answer regardless of the tool, and you are better served by a strong default than a underpowered test.

If you need element-level, on-page experimentation — testing button colours and headlines within one page — a dedicated page-experiment platform is the right tool, and link testing chooses between pages rather than within them.

And if you do not bill through a supported provider, the revenue read that makes link testing honest will be incomplete. TrackRev fits teams with real link volume who want to choose between destinations on revenue.

A link test is only trustworthy when the split and the revenue come from one tool — measure the split in one system and the charge in another and you cannot be sure which variant a customer was assigned to.

The default stack splits exactly there: Bitly Growth at ~$35/month can rotate and count clicks, but the revenue lives in Rewardful Starter at ~$49/month or a separate analytics tool, $84+/month for two systems that never share a variant assignment.

TrackRev is $39/month for link tracking, revenue attribution, and the affiliate programme on one billing connection, so a link test’s split and its revenue per visitor are computed from the same data.

You can only read a link test on money when one tool owns the whole experiment.