September 28, 2026
With a method that matches the constraint, because you cannot split visitors on price the way you split them on a headline. Showing two visitors different prices for the same product creates fairness problems and data problems at once. The workable methods are geo-based splits, time-based before-and-after designs, and price-presentation tests that change how the price is framed rather than the number itself. Pick the method first; the tooling follows.
A standard A/B test shows half your visitors version A and half version B. Do that with price and you have two problems. The first is fairness: two shoppers seeing different prices for the same product will eventually compare notes, and the fallout costs more than the test is worth. The second is contamination: price-sensitive shoppers talk, screenshot, and share, so your clean split is not clean for long.
Price also moves slowly through the business. Changing a headline affects one page; changing a price affects margin, inventory velocity, ad performance, and customer expectations. A price test needs a longer runway and a wider set of guardrails than a page test, because the blast radius is bigger.
That does not mean pricing is untestable. It means the unit of the test cannot be the visitor. The workable units are geography, time, and presentation.
Split by region: one set of markets sees the test price, a matched set keeps the control price. Geo splits sidestep the fairness problem because shoppers in different regions expect different prices anyway, thanks to currency, taxes, and shipping. Use Shopify Markets to set the regional prices and compare matched regions with similar traffic and conversion baselines.
Matching matters more than the split mechanism. Pick regions with comparable volume and seasonality, run the test for full weekly cycles, and watch for spillover: cross-border shoppers and VPN users blur the lines. Geo tests are directional rather than precise, but they are the closest thing to a clean price experiment most stores can run.
Set a kill switch before you start. Define the revenue or margin floor that ends the test early, and make sure someone is watching it. A price test that runs unattended through a holiday weekend can do real damage.
Time-based designs compare a period at the test price against a comparable period at the control price. They are simple to run and honest about their weakness: seasonality, campaigns, and day-of-week effects all contaminate the comparison. Use them only with matched comparison windows and a skeptical eye, and never run one across a major sale.
Presentation tests are the underused option. Instead of changing the number, test how the price is framed: charm pricing versus round numbers, showing per-unit prices, strikethrough anchoring, or bundling that changes the effective price without touching the list price. These split cleanly by visitor, need no geo machinery, and often move conversion as much as a price change would.
Start with presentation. It is cheaper, faster, and reversible in minutes. Move to geo or time-based tests only when presentation tests stop finding wins and the pricing question is still open.
Watch margin, not just conversion. A price cut that lifts conversion 10% while cutting margin 15% is a loss wearing a win's clothes. Track revenue per visitor and gross margin per visitor as primary metrics, with conversion as a secondary signal.
Protect your regulars. Email subscribers and repeat buyers notice price changes first and forgive them least. Consider excluding your loyalty segment from price tests, or at least monitoring their behavior separately. The lifetime value math on alienating repeat buyers is brutal.
Document everything and time-box the test. Price tests have a way of becoming permanent by inertia: the test price just stays because nobody turned it off. Decide the end date, the decision rule, and the rollback plan before the first visitor sees the test price.
Alternate the price for set periods, or use a testing tool that splits traffic between two prices for the same product, while holding everything else constant. Period switching is simpler and suits low traffic, but must account for weekday and seasonality effects. Traffic splitting is cleaner statistically and harder to set up without affecting discounts and reporting.
Units per order, conversion rate, margin per visitor, returns, and discount dependence during and after the test. A higher price that wins revenue while collapsing units and margin quality may not win profit. Decide the primary measure, usually margin per visitor or per session, before the test starts, and write it in the test record.
Through at least two full weekly cycles and enough orders to see past noise, longer at higher price points where purchases are rarer. Avoid launching or ending across a sale event that the other arm did not experience equally. Price results also decay as competitors move, so date the result and revisit it rather than treating it as permanent truth.