When landed costs jump, most merchants add a percentage to every price in the catalogue. It is quick, easy to explain in a meeting, and it answers the wrong question. The decision that matters is which products can carry a higher price and which will lose more in volume than they gain in margin. A blanket increase cannot tell you that, because there is no version of the month left that excludes the increase.
Costs rarely move evenly across a catalogue. A heavy, bulky item absorbs freight and duty differently from a light, high-value one. Return rates differ by category, payment fees differ by basket size, and supplier increases arrive product by product. Raise everything by 10%, and you overcharge the products where costs barely moved while still losing money on the ones now underwater. Neither error shows up in the sales report.
Step 1: Build a replacement-cost model before you touch a price
Start with the landed cost per unit: purchase cost, freight, duties, fulfilment, payment fees and an allowance for returns. That gives you the range you have to work inside.
The floor is replacement cost rather than historical cost. The unit you sell today has to be bought again at today’s price, so anything still carrying an old cost basis in your system can look profitable while it loses money on every reorder. The ceiling is what the market will pay, which is a demand question and not a cost question. Your model cannot answer that one. A test can.
Keep the model in a spreadsheet you can re-run. Duty rates, surcharges and freight quotes change on dated schedules, and a rate that is provisional this quarter may be gone the next.
Step 2: Pick one product group under margin pressure
Pick a group where the cost increase actually landed, the margin is now below where you want it, and there is enough traffic to finish a test in weeks rather than quarters.
Leave out anything where the price is contractual. Wholesale and B2B accounts, subscriptions and marketplace listings behave differently from a direct-to-consumer page, and mixing them into one test gives you a result you cannot act on.
Step 3: Run both prices at the same time, on randomly assigned visitors
The most common mistake is a before-and-after comparison: old price in March, new price in April, then a look at how conversion moved. That design measures the market rather than the price. Ad spend changes, seasonality shifts, a competitor runs a promotion, payday cycles move the traffic mix, and one product going out of stock changes what people buy instead.
A controlled test runs both prices at the same time and assigns visitors to one or the other at random, so both groups see the same ads, the same demand and the same competition. Keep everything else fixed while it runs: the landing page, the imagery, the shipping terms. Change two things at once and you will not know which one moved the number.
Step 4: Read the result as contribution per visitor, not conversion rate
Conversion falls when a price goes up. That is expected. The number that tells you whether the increase worked is contribution per visitor: contribution per order multiplied by the conversion rate.
A worked example, on a product with a replacement cost of $47:
- Control, price $79: $32 of contribution per order, 3.4% of visitors buy. That is $1.09 of contribution per visitor.
- Variant, price $87: $40 of contribution per order, 3.2% of visitors buy. That is $1.28 of contribution per visitor.
Conversion fell 6% against the control. Contribution per visitor rose 17.6%. A dashboard that tracks conversion rate alone calls that variant a failure and reverts a change that made money. Contribution per visitor also answers the original question, which is how much of the cost increase this group of customers will absorb.
Two things to watch. Contribution per visitor measures the test window, so it is the right metric for a price decision and the wrong one for judging customer value over time. And the difference has to clear the noise before you act on it, which means sizing the test up front against your baseline conversion rate and the smallest change worth shipping. These tests do not win often: in a meta-analysis of 1,001 A/B tests, about 33.5% produced a statistically significant positive result, with a mean lift of 15.9% among the winners (Analytics-Toolkit). One win in three is a good return on the exercise. Reading a two-point conversion gap in a two-day test as an answer is not.
Step 5: Make the tested price real everywhere it appears
A price test breaks at the edges far more often than it breaks on the product page. The test price has to be applied consistently on the product page, the cart, the checkout, the order confirmation, refunds, your analytics and the product feed.
Part of this is now a policy question rather than a technical one. Google Merchant Center crawls your landing page and compares the price in the HTML and structured data against your feed, and it checks the checkout as well. A mismatch can get products disapproved, and repeated mismatches can escalate (Google Merchant Center Help). Two things follow for anyone testing prices:
- Prices loaded by JavaScript after the page loads are a known source of crawl mismatches, so what the crawler sees has to match what the shopper sees.
- Google’s rules also state that prices must not change on the landing page based on location, cookies, device or browser. A visitor-level price test on traffic that comes through Shopping feeds sits awkwardly next to that requirement, so decide before you start how paid shopping traffic will be handled during the window.
Price testing still works. It just makes the stack review part of the setup, which is why a price test is a project with an owner rather than a checkbox in an app.
Step 6: Set sample size, duration and decision rules in advance
Set these before the test starts and do not change them while it runs:
- The primary metric: contribution per visitor, with conversion rate and revenue per visitor as supporting numbers.
- Sample size, calculated from your baseline and the smallest effect worth acting on, for each variant.
- A duration of whole weeks, long enough to cover a full pay cycle. Two weeks is a floor, not a target.
- Decision rules: what result means roll out, what means leave the price where it is, and what means the test was too weak to decide.
- A freeze on every other change to those pages, prices, promotions and shipping terms for the duration.
The rule that matters most is the one about stopping. Peeking at a running test and calling it the moment the numbers look good turns a measurement into a coin toss.
Step 7: Watch the customer relationship, not only the number
Contribution per visitor tells you what the increase earned, not what it cost. The costs show up later:
- Complaints and contact volume, including in reviews and on social, where price frustration usually appears first.
- Returns and refunds for the tested group.
- Repeat purchase rate, the number most likely to fall after an increase that customers accepted once.
- The tone of support tickets, an early warning when a price feels unfair rather than merely higher.
Keep subscriptions and B2B contracts out of the test, and be careful with products people buy on habit, where an increase gets noticed every month rather than once.
Step 8: Roll out with an owner and a review date
Write the decision down: which products, the old and new price, the date, the metric and what it showed. Give one person ownership of the roll-out and set a review date.
Then treat repricing as routine work, because cost shocks keep arriving. In a survey of 500 US small and mid-sized businesses in April 2026, 96% said tariffs had hurt their shipping, sourcing or supply chain in the past year, and around half said their supply chain disruption protocol had never been tested (Ship4wd, reported by Digital Commerce 360). Every merchant needs a repeatable answer to “how much of this increase will customers accept?” A controlled test answers it with evidence rather than instinct.
To wrap up
- Do not raise every price by the same percentage. Costs move per product, and so does the demand response.
- Model replacement cost first: it sets the floor, and only a test finds the ceiling.
- Test one product group at a time, both prices at once, on randomly assigned visitors.
- Judge the result on contribution per visitor. A lower conversion rate can still mean more money, as the 17.6% example shows.
- Apply the price from product page to checkout to feed, watch repeat buyers, and roll out with an owner and a review date.




