A/B testing e-commerce: 7 high-impact ideas (2026)

Updated on September 23, 2026

Summary

• Before the 7 ideas: why not copy a recipe
• 1. The free delivery threshold
• 2. Emergency and stock signals
• 3. The number of checkout steps
• 4. Social proof near the buy button
• 5. The pop-up exit-intent, well-balanced
• 6. The specific mobile experience
• 7. The text and position of the purchase button
• How to prioritize these 7 ideas on your website
• Frequently Asked Questions
• Key takeaways

A 304% increase in conversions. This is the figure cited in almost every article on e-commerce A/B testing, attributed to a simple move of the call-to-action button below the fold, as documented by ContentVerve. The problem: this result comes from a specific website, with its own traffic and context. When replicated as is elsewhere, it often yields no results—sometimes even the opposite.

This guide offers 7 concrete test hypotheses for e-commerce, chosen for their potential real impact on an average French website — in the same spirit as the e-commerce case study catalog published by VWObut by emphasizing the hypothesis to be tested rather than the result to be copied. But it begins with a necessary warning, which most lists of the same kind omit.

Before the 7 ideas: why not copy a recipe

Here's what lists of "best A/B testing ideas" almost always fail to mention: the spectacular percentages they cite (+89% sales after changing a product title, +100% conversions after removing a navigation menu, +349% leads with a warmer email tone) each come from a single site, tested only once, in a specific context. These aren't universal laws of CRO. They're isolated case studies.

And this is no small detail. A change that worked on an American dietary supplement website is by no means guaranteed to work on a French clothing website. The correct way to interpret these examples is therefore not "I copy the change," but "I test the underlying hypothesis on my own traffic." It is this nuance, more than the list itself, that distinguishes a testing program that progresses from one that piles on changes without ever truly learning.

There is another, more technical reason to be wary of these isolated figures: a spectacular result on a small sample is statistically more likely to be a happy coincidence than a modest result observed on a large traffic sample. This phenomenon is documented in our article on the statistical significance The more extreme the claimed effect, the more it deserves to be verified before being taken at face value, not the other way around. That's why each of the 7 ideas below is presented with the hypothesis to be tested, not a result to copy and paste.

1. The free delivery threshold

Displaying a clear free shipping threshold on the shopping cart page remains one of the most direct ways to combat cart abandonment — a phenomenon that affects 70.22% of shopping carts on average, according to the 50 studies aggregated by the Baymard InstituteIn France, with an average basket measured at 62 euros by Fevad and half of the transactions under 30 euros, a poorly calibrated threshold may either never be reached (too high), or not encourage the addition of an extra item (too low).

Hypothesis to be tested: display a dynamic message such as "More than X€ for free delivery" as soon as the item is added to the cart, with two or three different thresholds to compare according to your actual average cart size — not a round number copied from a competitor.

A threshold set too high relative to your actual average order value discourages rather than encourages: if your average order value is around €45 and the free item threshold is set at €80, most visitors will consider the goal unattainable and abandon the idea of ​​adding an item rather than getting closer to it. A threshold slightly higher than the current average order value—in the range of 15% to 25% above—generally remains the most realistic starting point to test first.

2. Emergency and stock signals

A limited stock indicator ("Only 3 left in stock") or a countdown timer on a sale can accelerate the purchase decision—provided it's honest. A false sense of urgency, detected by a visitor who returns to the page the next day and still sees "Only 3 left in stock," damages trust more permanently than it generates short-term conversions.

Hypothesis to be tested: display a real stock signal (connected to your inventory) only below a critical threshold, rather than across the entire catalogue.

The guardrail metric to watch here is not only the conversion rate but also the product return rate: an overly urgent signal can lead to an impulsive purchase that is regretted a few days later, thus negating part of the apparent gain once returns are taken into account.

3. The number of checkout steps

Should you have a one-page or multi-step checkout process? This question comes up constantly, and the honest answer is that it depends on the product. A simple, inexpensive purchase often benefits from a short checkout process; a complex purchase (multiple options, delivery options to configure) can benefit from separate steps that gradually build trust.

Hypothesis to be tested: compare your current funnel to a version that groups (or conversely separates) the delivery and payment fields, tracking both the final conversion rate and the abandonment rate at each stage — it is often this latter data, more precise than the overall rate alone, that reveals where the user really drops out.

And that's not all: the number of fields required often matters more than the number of screens. A single-page checkout with twenty fields to fill out can be more discouraging than a three-screen checkout with only four fields each. Before deciding between "one page" and "multiple steps," first count the number of fields that are truly essential—most payment forms contain more than necessary.

4. Social proof near the buy button

A well-presented customer review can have a greater impact on the final decision than expected—a documented case at a sporting goods retailer cites a 69% increase in sales after adding a product expert review. Again, the exact figure doesn't translate directly; the principle, however, is worth testing: social proof works best when it appears at the precise moment of the decision, not just at the bottom of the page.

Hypothesis to be tested: Move a relevant customer review (number of stars, short verbatim) directly next to the add to cart button, rather than in a separate block further down on the product page.

5. The pop-up exit-intent, well-balanced

The pop-up that triggers when a visitor is about to leave the page can retain some of the traffic that is about to leave — but a poorly calibrated pop-up annoys more than it converts. Practitioners in the sector recommend a minimum delay of 8 to 15 seconds on the page and a scroll depth of 35% to 60% before triggering the pop-up, to avoid interrupting a visitor who has just arrived.

Hypothesis to be tested: compare a pop-up triggered only on exit intent (cursor leaving the window) to the absence of a pop-up, monitoring the overall bounce rate in guardrail metric — an overly intrusive pop-up can drive away visitors who would have returned later without it.

A technical detail also matters here: a pop-up that appears with a noticeable delay, or that causes a visual jump effect on the page, degrades the experience before its content is even read. Our page on the Ultra-fast diffusion explains how to avoid this kind of technical friction on dynamic elements such as pop-ups.

6. The specific mobile experience

This is a point too often treated as a mere sub-case of the desktop, when it deserves its own hypotheses. Smartphones now represent 69% of usage among French online shoppers according to Fevad, up 7 points in one year, and 55% of shoppers are multi-screen users — a context that also has a direct implication for tracking, since the CNIL's recommendations of January 2026 on multi-device consent.

Hypothesis to be tested: do not settle for a "responsive" test modeled on the desktop, but specifically test elements specific to mobile — size of clickable areas, position of the purchase button at the end of the page vs fixed at the bottom of the screen, number of fields visible without scrolling on the payment form.

If you're new to this and some terms (segment, sample, engagement rate) are still unclear, our CRO glossary They simply define them before embarking on a mobile-specific test.

7. The text and position of the purchase button

The text on a call-to-action button influences the decision more than you might think—one button color tested in a documented case generated 21% more clicks. But color alone never explains everything: it's often the combination of visual contrast, precise call-to-action text, and placement within the field of vision that makes the difference, not any single element in isolation.

Hypothesis to be tested: rather than testing only one color, compare a generic button text ("Validate") to a benefit-oriented text ("Receive my package tomorrow") — a change that relates to intention rather than aesthetics, which is often more decisive.

How to prioritize these 7 ideas on your website

Seven ideas doesn't mean seven tests to launch simultaneously. Prioritize according to two simple criteria: the available traffic on the page in question (no point in testing the checkout if you only have a few dozen orders per month) and the scale of the current problem (if your cart abandonment rate far exceeds the industry average, start with delivery and checkout before social proof).

Document each hypothesis tested in a shared backlog, along with the result—successful, unsuccessful, or inconclusive—to avoid retesting the same idea twice six months apart. It's this disciplined tracking, more than the number of available ideas, that makes a testing program successful over time. Our guide on goals and tracking details how to structure this backlog in order of priority.

Frequently Asked Questions

Which e-commerce A/B test is the most profitable to launch first?

The free delivery threshold displayed on the shopping cart page, because it directly impacts shopping cart abandonment while being simple to implement, without page redesign.

Why not simply copy the results of case studies found online?

A result like "+304% conversion" almost always comes from a specific website, with its own traffic, audience, and context. Replicating this change as is on another site may have no effect, or even harm conversions. What is transferred is the hypothesis being tested, not the reported percentage.

Should we test the checkout as a single page or in multiple steps?

There's no one-size-fits-all answer: it depends on the number of fields required and the type of product. A short funnel often encourages simple, impulse purchases, while a multi-step funnel can provide reassurance for complex or expensive purchases. Only testing with your own traffic will definitively answer this question.

How long does it take to test an e-commerce idea with high seasonal traffic?

Ideally, cover a full cycle including a weekend and avoid concluding during an atypical promotional period (sales, Black Friday), which distorts usual purchasing behavior.

Do exit-intent pop-ups actually work?

This depends heavily on the timing and the proposed reduction threshold. Best practices recommend a minimum delay of 8 to 15 seconds on the page and a scroll depth of 35% to 60% before triggering, to avoid interrupting a visitor who has just arrived.

Key takeaways

The seven ideas presented here cover the most documented levers of e-commerce A/B testing: delivery, urgency, checkout, social proof, exit intent, mobile, and call to action. But the real lesson goes beyond the list itself: no percentage found in an article, even a sourced one, guarantees an identical result on your site. What transfers from one case to another is the question asked—not the answer obtained elsewhere.

Prioritize based on your actual traffic, document each result, and be especially wary of figures that are too dramatic to be representative. A test that progresses methodically, hypothesis by hypothesis, builds a more sustainable advantage than a series of changes copied from the industry's most viral cases.

To launch these tests without technical complexity, our visual editor allows you to modify these elements directly on your site, with a results tracking which clearly distinguishes between real gain and statistical noise. If you manage multiple stores or a large catalog, theCRO autopilot It can also automatically prioritize these hypotheses based on the available traffic on each page. Find all our resources dedicated to...A/B testing for e-commerceAnd if you're currently comparing several platforms, our user feedback compared to Kameleoon details how each one handles these specific online sales use cases — accessible from our entry-level offers.