Useful ecommerce A/B testing ideas start with a specific moment: a visitor lands on the wrong product, cannot find the filter they need, or gets interrupted just as they are ready to buy.
Those are worth investigating. Whether they deserve a test depends on what your evidence shows.
Here are seven places I’d look, with the visitor problem and evidence that would make each relevant. Most can be explored using existing analytics, customer feedback, page behavior, and tools. You don’t need a new platform to ask a better question.
| Test idea | Possible visitor problem | Evidence to check |
|---|---|---|
| Offer a relevant alternative on the product page | The landing product isn’t right, and there’s no obvious next step | Entry sources, exits, product revisits, onward paths |
| Surface a useful filter | Visitors cannot narrow the range around what matters | Filter use, search terms, product attributes, usability findings |
| Put promotions in an on-demand drawer | Persistent offers compete with shopping | Placement, overlap, promo interactions, shopping paths |
| Ask one useful question | Visitors need direction the page doesn’t provide | Gift-related feedback, browsing behavior, existing curated destinations |
| Help returning visitors resume | Previously considered products are hard to recover | Return visits, repeated searches, product revisits |
| Clarify a purchase objection | An unanswered question prevents a confident decision | Support themes, research, delivery and payment questions |
| Change cross-sell timing | An extra decision interrupts purchase progress | Cart progression, removals, errors, previous upsell results |
Start with the moment that needs investigating.
Three possible visitor problems. Three different questions worth asking.
The useful filter is buried.
Can visitors narrow the range around what matters to them?
This product isn’t quite right.
Is there a useful next step when the landing product doesn’t fit?
One more decision before checkout.
Does the extra offer help the purchase or interrupt it?
Seven ecommerce A/B testing ideas worth investigating
1. Give product-page entrants a relevant alternative
Evidence to check: Look at visitors entering directly on product detail pages. Do some leave without exploring, while research or browsing paths suggest the product doesn’t quite fit?
Control → treatment: Compare the current PDP with a compact route to similar items or a relevant category.
Hypothesis: For direct PDP entrants considering alternatives, an easier onward path should improve revenue per visitor by helping them find a better fit.
Measure: RPV across the eligible audience. Use onward browsing, add-to-cart, purchase conversion, and AOV to explain the result.
Leave it alone when: Visitors already find suitable products and move forward. More recommendations can create another decision instead of solving one.
The different jobs a product page needs to do help explain why entry context matters here.
2. Make a high-value filter easier to find
Evidence to check: Combine filter behavior with search terms, customer questions, and usability findings. Which attributes actually help visitors narrow the range: material, fit, compatibility, use case?
Control → treatment: Compare the current filter layout with one useful attribute made easier to notice and use.
Hypothesis: For visitors shopping a category where that attribute matters, surfacing it should improve RPV by reducing the effort needed to find suitable products.
Measure: RPV, with filter use, product views, add-to-cart, conversion, and AOV as diagnostics.
Leave it alone when: Filters are already easy to find, or the catalog data is unreliable. Don’t automatically put price first just because everyone understands it.
3. Reduce persistent promotional clutter
Evidence to check: Map banners, carousels, offer messages, and overlays. Where do they overlap? Do interactions lead toward useful shopping, or into repeated detours?
Control → treatment: Compare persistent promotional placements with an accessible, clearly labeled offers drawer. Keep the offers available.
Hypothesis: For the defined shopping audience, reducing persistent promo exposure should improve RPV by making product discovery easier to continue.
Measure: RPV, conversion, AOV, shopping progression, and promo interactions. Include discount and margin guardrails. Examine relevant audiences defined before exposure.
Leave it alone when: Evidence suggests the offer is central to the visit and supports progress. A deal-seeking visitor may need a different experience from someone comparing full-price products.
4. Ask one useful question at a decision point
Evidence to check: Look for an unanswered need you can do something about. Gift-related feedback and an existing curated gift destination are a useful starting point.
Control → treatment: Compare the current page with a compact gift/self-purchase question that leads to a relevant destination.
Hypothesis: For eligible gift-shopping visits, a useful question and onward route should improve RPV by helping visitors find products that suit the occasion.
Measure: RPV across everyone assigned to the test, then prompt interaction, onward browsing, add-to-cart, conversion, and AOV.
Leave it alone when: You have nowhere useful to send the visitor, or the page already has competing interruptions. Asking a question creates an obligation to do something useful with the answer.
5. Help returning visitors resume shopping
Evidence to check: Do returning visitors repeatedly search for or revisit products they considered earlier? Look for difficulty recovering those items, not just evidence that people return.
Control → treatment: Compare the current return-entry experience with recently viewed products at a relevant location, where browsing history is available and permitted.
Hypothesis: For returning visitors with usable product history, easier access to considered items should improve RPV by reducing the effort required to resume shopping.
Measure: RPV, with product recovery, module interaction, add-to-cart, conversion, and AOV as diagnostics.
Leave it alone when: Most eligible visits lack useful history, or the module would displace something more relevant.
6. Clarify a real purchase objection
Evidence to check: Use research and support themes to identify a recurring question about delivery, sizing, returns, payment, or product suitability. Don’t assume hesitation means someone needs urgency.
Control → treatment: Compare the current experience with concise, accurate information near the decision it supports.
Hypothesis: For visitors facing the identified objection, answering it at the relevant moment should improve purchase completion and RPV by reducing uncertainty.
Measure: For a revenue question, use RPV as primary; purchase conversion, checkout progression, support contacts, and relevant return reasons are diagnostics.
Leave it alone when: The answer is already clear or isn’t a meaningful blocker. Another reassurance badge can become another thing to ignore.
7. Change cross-sell timing or presentation
Evidence to check: Review what happens around the existing upsell: checkout progression, offer acceptance, removals, errors, and revenue. A higher attachment rate doesn’t settle the question.
Control → treatment: Compare the current placement with a less intrusive inline offer or a later placement that still lets visitors make an informed choice.
Hypothesis: For visitors eligible for the offer, reducing interruption should improve RPV by protecting purchase progress while retaining useful cross-sell opportunities.
Measure: RPV, conversion, AOV, attachment, removals, and errors. Check margin where product or service mix changes.
Leave it alone when: The offer already fits the decision and there’s little evidence of disruption.
Three examples from my experimentation work
These are reported results from specific experiences, not expected lifts for the ideas above.
Fewer promo clicks, better business performance
Persistent promotions occupied prominent space across the site. We moved them into an on-demand offers drawer.
The promotion drawer case study reports nearly 60% fewer promo clicks alongside improvements in overall conversion, AOV, and RPV. Deal seekers underperformed.
That creates a useful next question about audience needs. It doesn’t establish that removing promotions always increases revenue, or prove exactly why the overall result improved.
A gift question with somewhere useful to go
Returning holiday visitors were landing on PDPs with low intent, while the route to an existing gift guide was easy to miss. A compact question offered a curated path.
The holiday PDP prompt case study reports increases in AOV, add-to-cart, and RPV, without numerical lifts.
Visitors choosing Yes or No were self-selected groups. Their behavior can inform another hypothesis, but comparing those groups doesn’t isolate the prompt’s treatment effect.
More service-plan carts, less revenue per visitor
A lightbox prompted a service-plan decision immediately after Add to Cart.
The post-cart pop-up case study reports about 60% more service-plan carts while RPV fell 2.28%. Removal and error rates also rose. Sample size and statistical significance aren’t provided in the public write-up.
The offer got more attention. Revenue moved the other way.
Service-plan carts
+~60%
Increased after the post-add-to-cart prompt.
Revenue per visitor
−2.28%
Decreased in the same reported experiment.
A better attachment metric doesn’t automatically mean a better business outcome.
The local metric improved while the revenue metric moved the other way. That’s the tradeoff to investigate, rather than concluding that every modal fails.
How to choose the next ecommerce test
Find the problem before choosing the treatment
Triangulate analytics with research, feedback, recordings, and previous results. A funnel drop-off identifies a place to investigate. It doesn’t prove friction or explain the cause.
Write down the visitor problem in plain language. “People cannot compare available sizes” is more useful than “the PDP needs optimization.”
Define who qualifies before exposure
Use entry source, category, returning status, or behavior recorded before treatment to define eligibility. Then randomize within that audience.
“Visitors who clicked the new module” is an analysis group, not a fair control comparison. Clicking may reflect motivation the visitor already had.
Separate fixes from experiments
If a checkout control is broken, fix it. You don’t need a winning variant to justify a functioning button.
Test when you’re uncertain about the response or tradeoff. Will a clearer offer help purchases, or distract from the product?
Compare opportunities using evidence and effort
Consider affected traffic, plausible business impact, evidence strength, and delivery effort.
Illustrative example, not actual site data or a forecast:
| Factor | Surface a useful category filter | Add a new cross-sell carousel |
|---|---|---|
| Affected traffic | A well-trafficked category | A smaller eligible cart audience |
| Evidence | Research and repeated searches show difficulty narrowing | Stakeholders want more product exposure |
| Plausible business impact | Help visitors find suitable products | Increase basket value, with a purchase-completion tradeoff |
| Delivery effort | Existing filter and reliable attributes | New component and recommendation rules |
I’d investigate the filter first. It has a clearer visitor problem, broader reach, and less delivery effort. The carousel may deserve attention later, but a stakeholder request isn’t enough evidence to put it first.
Write the hypothesis and decision rule
Use this pattern:
For [eligible audience], changing [experience] should improve [business outcome] because [evidence-backed visitor problem].
Before launch, specify the primary metric, meaningful effect, sample needs, measurement window, guardrails, and how you’ll handle uncertainty. Decide what would justify shipping, revising, or stopping the treatment.
Measure the business outcome
For a revenue-focused experiment, my usual primary outcome is revenue per visitor. Conversion rate, AOV, add-to-cart, and interaction help explain what moved.
Calculate outcomes consistently across everyone randomized into the eligible audience, including non-clickers. Keep visitor counting, revenue definitions, and measurement windows consistent between groups.
More CTA clicks don’t establish more revenue. Higher clicker-only revenue doesn’t establish a winning treatment. Neither does a raw difference between two visitor segments.
RPV isn’t profit. Check margin and discount costs where relevant, and remember that revenue alone doesn’t establish a better visitor experience or long-term value. My guide to revenue per visitor versus conversion rate covers those tradeoffs in more detail.
Plan duration and sample requirements for the experiment’s traffic, outcome variability, meaningful effect, and purchase cycle. There isn’t one visitor threshold or number of days that makes every ecommerce test reliable.
Have plenty of ideas but no clear first move?
The Ecommerce Growth Opportunity Sprint brings your existing analytics, research, feedback, session behavior, and previous results together into 5–10 prioritized opportunities, with a recommended next action for each.
The fixed investment is $4,500. Typical turnaround is 7–10 business days once access and materials are available, focused on one primary journey or objective. You get a 60-minute strategy readout and one follow-up clarification round.
The standard Sprint covers diagnosis and recommendations. Experiment builds, production design and development, new primary research, and detailed implementation roadmaps are outside that scope.