In short
Learning goal: by the end of this lesson you can pick a product page test that addresses a real reason shoppers hesitate, judge it on revenue rather than add-to-cart, and tell when a multivariate test is worth the traffic.
Who this is for: Shopify brands doing roughly $1M to $10M a year whose best sellers get thousands of product page views a day.
Where the decision is made
In lesson 1's example funnel, 62 of every 100 sessions viewed a product and only 14 reached the cart. For most brands, the product page is where the largest share of shoppers decide not to buy. That makes it the step with the most to gain, and also the step with the most noise: lots of ideas, most of them about how the page looks rather than what the shopper needs to know.
There is a lot of room. Baymard Institute's product page research reports that "only 49% of e-commerce sites have an overall 'decent' or 'good' UX performance for their product pages" (Baymard, read 12 October 2026). The useful question is not "what could look better?" but "what makes a shopper who wants this product hesitate?"
What to measure at this step
| Metric | Definition | Use it as |
|---|---|---|
| Revenue or profit per visitor | Revenue (or gross profit) from visitors who saw the tested product pages, divided by those visitors | The primary goal |
| Add-to-cart rate | Share of those visitors who add any item | A diagnostic |
| Variant selection | Share who choose a size, color or pack | A diagnostic |
| Reviews engagement | Share who open or scroll to reviews | A diagnostic |
| Refund rate | Refunds on orders from those visitors | A guardrail: does the page oversell? |
Test ideas, each with its mechanism
Price presentation. Mechanism: the shopper judges value, not just price. Per-unit prices for packs, a subscription saving shown as an amount, or a genuine "was" price can change how the same price reads. Testing the price itself is a separate kind of test, with its own rules (see price tests before Black Friday).
Delivery information. Mechanism: uncertainty about when it arrives, and what it costs, stops a decision. A delivery estimate and the shipping cost (or the free-shipping threshold) near the add-to-cart button answer the question before the shopper goes looking.
Reviews. Mechanism: social proof for a first-time buyer. Moving the star rating near the title, or showing the most helpful review for the most common doubt (fit, taste, durability), helps the shopper who is unsure. It does little for returning customers.
Imagery. Mechanism: the shopper can't touch the product. Images that show scale, texture or the product in use answer questions that a white-background shot can't. This one often helps most on mobile, where the first image is most of the screen.
Variant selection. Mechanism: choice friction. A size guide next to the size selector, or a "most popular" label on one pack, reduces the work of choosing.
Write each test as a hypothesis with its mechanism: "Showing the delivery date next to add to cart will raise revenue per visitor, because shoppers who need it by a date can decide without leaving the page." A test that loses still teaches you whether that doubt mattered.
Why add-to-cart-only winners mislead
Add-to-cart is the step right after the product page, so it moves first and it moves most. That makes it tempting to call a winner on it. It is not the same as buying.
Illustrative counts, not data from a real store. Both results computed with Benchmyrk's results engine, 14 days, two versions.
A product page test with 20,000 visitors per version. The new version adds a "most popular" badge to the three-pack.
| Goal | Control | New version | Lift | 95% interval | Verdict |
|---|---|---|---|---|---|
| Add to cart (diagnostic) | 1,600 (8.0%) | 1,760 (8.8%) | +10.0% | +3.1% to +17.4% | Winner |
| Orders per visitor (primary) | 600 (3.0%) | 612 (3.1%) | +2.0% | −8.7% to +14.0% | Inconclusive |
More shoppers added to cart. Almost all the extra adds dropped out before buying: they were shoppers comparing packs, not shoppers deciding to buy. On the primary goal, the result could be anything from a 9% loss to a 14% gain. This test is inconclusive. If the page now nudges shoppers to a pack that they then abandon in the cart, that is worth knowing, and it's a hypothesis for the next test, not a win.
When multivariate testing fits
A multivariate test (MVT) changes two or three parts of a page at once (say, the headline and the image) and shows every combination. It tells you the effect of each part and whether the parts interact. It also needs more traffic.
Illustrative inputs. Computed with Benchmyrk's MVT sample-size formula (two-sided, 95% confidence, 80% power, even split).
Inputs: 5,000 product page visitors a day, a 3.0% conversion rate, and a 10% relative change to detect.
| Design | Visitors per version | Total visitors | Days |
|---|---|---|---|
| A/B test, one change | 53,211 | 106,422 | 22 |
| MVT, 2 parts × 2 versions (4 combinations), effect of each part | 35,487 | 141,948 | 29 |
| Same MVT, to check whether the parts interact | 135,401 | 541,604 | 109 |
The effect of each part costs about a week more than one A/B test, and you learn about two changes. Detecting an interaction between them takes nearly four months at this traffic. So an MVT is a good fit when you have two independent ideas for the same page, and a poor one when you're mainly hoping the combination will do something special.
Common problems
- Testing a design, not a doubt. "A cleaner layout" has no mechanism. Find the doubt first.
- One template, many products. A change to the product template reaches every product. Results are an average: a size guide helps clothing and does nothing for candles. Target the products the doubt applies to.
- Seasonal products in November. A gift product's page behaves differently in BFCM week. Read results from a normal period.
- Changing the price display in a way checkout doesn't match. A per-unit price or a "was" price must be the same at checkout.
Exercise (10 minutes, on your own store)
- Open your three best-selling product pages on a phone.
- For each, write down the first doubt you'd have as a new customer: delivery, fit, value, quality or trust.
- Find where on the page that doubt is answered. Count the scrolls or taps it takes.
- Pick the doubt that takes longest to answer on the page with the most traffic. Write a hypothesis with its mechanism.
- Check that page's daily visitors and conversion rate in the sample-size calculator for a 10% change.
You don't need Benchmyrk for any of this.
Do it in Benchmyrk
Check yourself
- Add-to-cart rose 10% and orders per visitor is inconclusive. What's the verdict? Inconclusive. Add-to-cart is a diagnostic. The primary goal, set before launch, decides.
- When is an MVT a better choice than two A/B tests? When you have two independent ideas for the same page and enough traffic for the effect of each part. Detecting how parts interact needs several times more traffic.
- Why write the mechanism into the hypothesis? So a losing test still teaches you something: whether that doubt actually stopped shoppers.