What A/B price testing actually measures
A/B price testing (often just called price testing) splits your traffic and serves different price points to statistically similar groups. You then compare how each price performs on the metric that pays your bills.
The mistake almost everyone makes is choosing that metric wrong. Most people declare the higher-converting price the winner and move on.
That instinct is backwards. A price test is not a conversion test — it's a profit test, and the two usually point in opposite directions.
Why conversion rate is the wrong scoreboard
Lower prices convert better. That's not opinion; it shows up clearly in real store data.
Across 21 Shopify stores, ones with an average order value under sixty dollars posted a median conversion rate of 4.63%, while stores above two hundred dollars converted at a median of just 0.95%, according to DTC Pages' 2026 benchmarks. The overall median across the set was 2.81%.
So if you test a low price against a high price and crown "most orders" the winner, you'll pick the low price almost every time — and often leave money on the table. Conversion rate tells you how many people bought. It says nothing about whether those orders made you money.
The number that decides the winner: contribution margin per visitor
Contribution margin is what's left of an order after variable costs — cost of goods, shipping, payment and transaction fees, pick-and-pack — but before ad spend. Multiply that by orders, divide by visitors, and you get profit per visitor. That's the scoreboard.
Say you sell a candle. Every unit costs you twenty dollars all-in (product, shipping, fees), and you're testing a forty-dollar price against a forty-eight-dollar price.
At $40, your margin is $40 − $20 = $20 per order. Say 40 of every 1,000 visitors buy — that's 40 × $20 = $800 of margin per 1,000 visitors, or $0.80 per visitor.
At $48, your margin is $48 − $20 = $28 per order. The higher price scares off some buyers, so say only 32 of every 1,000 buy — that's 32 × $28 = $896 per 1,000 visitors, or about $0.90 per visitor.
The $48 price converted lower (32 orders versus 40) and still won by $96 per thousand visitors. If you'd scored the test on conversion rate, you'd have shipped the worse price. This is the entire game: optimize contribution margin per session, not order count.
Price is an ad-efficiency lever, not just a revenue lever
Here's the part most guides skip. The price you land on quietly resets how hard your ads have to work.
Break-even ROAS — the return on ad spend where revenue exactly covers the goods plus the ad — is simply one divided by your contribution margin. At the $40 price with $20 of variable cost, margin is 50%, so break-even ROAS = 1 ÷ 0.50 = 2.0x. At the $48 price, margin is $28 ÷ $48 = 58%, so break-even ROAS = 1 ÷ 0.58 ≈ 1.71x.
Read that again. By raising the price, you dropped the ROAS your ads must clear from 2.0x down to about 1.7x — without touching your ad account at all.
Now imagine a Meta campaign running at 1.85x ROAS. At the $40 price it loses money (below the 2.0x break-even); at the $48 price the same campaign turns a profit (above 1.71x). A winning price test can rescue a channel your ad manager was about to kill.
This is why pricing and paid ads belong in the same conversation. If you're working through how to scale ad spend without torching profit, remember that lowering your break-even ROAS through price and margin buys you room to scale further down the diminishing-returns curve before your marginal dollar goes underwater. Wringing more out of each click through better ad creative and click-through rate helps too — but price sets the ceiling both live under.
How to run a clean A/B price test
Getting the price right is worthless if the test itself is rigged. A few rules keep you honest.
Split traffic fairly, not by time. Running $40 this week and $48 next week isn't a test — it's two different weeks with different demand, ads, and seasonality. Show both prices in the same window to randomly assigned visitors.
Score on profit, then check significance. Wait for enough orders per variant to trust the result; a handful of sales can't tell you anything. On lower-volume stores, test fewer prices for longer rather than many prices at once.
Test one thing at a time. If you change the price and the offer and the product photo together, you can't attribute the result to any of them. Isolate the price.
Mind charm pricing separately. Ending prices in nine is nearly universal — up to 65% of all prices end in the number nine, per Convert's review of 500 ecommerce pricing tests. Whether $39 beats $40 for your buyers is its own small test; don't bundle it into a big price jump.
Price testing mistakes that quietly lose money
Chasing conversion rate. Covered above, but it's the number-one killer, so it's worth repeating: more orders at a thinner margin can net you less profit than fewer orders at a fatter one.
Ignoring the fee drag on high prices. Payment processors take a percentage, so a higher price carries a slightly higher fee in absolute dollars. Fold that into your variable cost before you compute margin, or you'll overstate the winner.
Testing price when the real lever is AOV. Sometimes the profitable move isn't a higher unit price at all — it's a bigger basket. Raising average order value across your store through bundles or thresholds lifts margin per order the same way a price increase does, usually with less conversion drag. A one-click post-purchase upsell adds margin at zero extra acquisition cost, because the customer has already converted.
Guessing at your true per-order cost. Every calculation above depends on knowing your real variable cost per order — the exact COGS, shipping, and fee figure. If that number is a guess, your "winner" is a guess too.
Where PodVector fits
Every worked example on this page needs one input you probably don't have cleanly: true per-order profit. PodVector connects your Shopify, Meta Ads, Google Ads, Printify, Printful, and Stripe data and computes that true per-order profit — the contribution-margin number a price test lives or dies on.
Victor, PodVector's AI operator, reads that live data, spots where your margin and break-even ROAS actually sit, and proposes moves — taking Shopify-side actions with your approval. He does not touch your ad account; he reads ad data and hands you the decision. If you'd rather have AI help find the profit levers than reconcile spreadsheets, start with PodVector and let the per-order math run itself.
FAQs
How long should an A/B price test run?
Run it for at least one full purchase cycle for your product, and long enough to collect enough orders per variant to trust the difference. Two weeks is a common floor, but volume matters more than the calendar — a slow-moving product needs longer, a high-traffic store can conclude sooner.
Should I test more than two prices at once?
Only if you have the traffic. Each price point splits your orders further, and thin samples produce noise you'll mistake for signal. Low-volume stores are better off testing two prices well than four prices badly.
Is it risky to show different prices to different shoppers?
Standard A/B price testing shows randomly assigned visitors different prices for a limited window, which is common practice. Keep it clean by honoring whatever price a customer saw at checkout and not varying price by anything that looks like a personal characteristic. When in doubt, test at the collection or store level rather than per individual.
What if the higher price converts lower — is it still worth it?
Often, yes. As the candle example showed, a price that converts lower can win on profit per visitor because each order carries more margin. Do the arithmetic: orders × margin per order ÷ visitors. Whichever price produces the bigger number wins, full stop.
Can't I just look at conversion rate?
No — that's the trap this whole article exists to warn against. Conversion rate ignores margin, and lower prices almost always convert better while making you less money. Score price tests on contribution margin per session, and treat conversion rate as a diagnostic, never as the verdict.