A/B testing for pricing: What works, what fails, and why

Pricing moves revenue faster than almost any other change you can test, which is why getting it wrong is expensive. A price test can win on the dashboard and still lose money for the business, and most teams don't catch it until after they've rolled out the winning variant. The trick is knowing which metric to watch during the test, and which one to check before you roll anything out.

Key takeaways

• A pricing test can lift sales and total revenue and still lose money once real costs are factored in, so check profit before rolling anything out.
• Track revenue per visitor while the test runs, since it shows the combined effect of price on how many people buy and how much they spend.
• Pricing tests usually need around 1,000 visitors per price to produce a trustworthy result.
• The elements worth testing are the price itself, discounts, bundles, and how often customers are billed for subscriptions.

What is A/B testing for pricing?

A/B testing for pricing is a controlled experiment that shows different price points, discounts, or billing structures to different visitor groups. It measures the impact on business outcomes like conversion, average order value, and ultimately revenue per visitor.

What this could look like

• E-commerce brands: Often test bundle pricing, offering a discount when customers buy an assortment of products together in a single order, since shoppers respond well to a well-priced bundle.
• Subscription or SaaS products: Commonly test monthly versus annual billing, or the price and feature mix across different plan tiers.
• Single-product or DTC brands: Usually test a straightforward price point, since there's only one product to price and less complexity to isolate.

What you need in place before you start testing price

Before launching a pricing test, make sure you have the following in place:

1. Enough traffic to trust the results: Pricing tests usually need a bigger sample size than copy or design tests, since it takes longer to see a clear effect on revenue per visitor. As a rough guide, we recommend waiting for around 1,000 visitors per variation before trusting the results. On low-traffic pages, that can mean weeks of testing, or a different method altogether.
2. A single, isolated variable: Change the price and the product image at the same time, and a drop in conversion won't tell you which one caused it, as one pricing experimentation guide points out. Keep everything else on the page exactly the same.
3. Know your real costs upfront: Most A/B testing software tracks conversion rate or revenue, not profit. It doesn't know your shipping costs or unit costs, so you'll need that data ready to check against the test results before rolling anything out.

» Want a tool that keeps your price tests clean? Check out the best A/B testing tools

How to run a pricing test from hypothesis to rollout

The process for testing price follows the same basic shape as any other A/B test, from running the test to analyzing the results, with a few pricing-specific decisions layered in along the way.

Step 1: Choose your metric before you start

Before running any pricing experiment, the team needs to agree on what "winning" means.

Revenue per visitor blends conversion rate and average order value into one number, so a lift in one metric can't quietly mask a drop in the other.

» Learn how to build programs that scale with conversion rate optimization

Step 2: Define what's actually being tested

Is it a single product price, a discount structure, or a broader pricing page? That choice depends heavily on the type of site. A single-product page and a large catalog call for different approaches. A catalog with dozens of SKUs can spread traffic too thin for any one price test to reach significance.

Step 3: Form a specific hypothesis

Rather than testing a price change just to see what happens, start with a clear, testable statement, for example: "Lowering the price by 10% will increase revenue per visitor by making the product more competitive against similar options."

A specific hypothesis makes it much easier to judge afterward whether the test actually answered the question it set out to ask, rather than just producing a number without context.

Step 4: Segment and split your traffic

Split traffic evenly between the current price (the control) and the new price (the variant) using a proper A/B split test, and randomize the split so each group fairly represents your overall audience.

If certain segments, like new versus returning visitors, are known to behave very differently around price, it's worth checking whether the test needs to run separately for each group rather than blending them together into one result.

Step 5: Run the test to statistical significance

Let the test run until it reaches the sample size needed for reliable results, rather than calling it early based on an early lead. This is also when you should double-check profit, not just revenue.

Step 6: Analyze the results before calling a winner

Once a test reaches significance, check these in addition to your usual conversion rate and average order value numbers:

• Revenue per visitor between variants: Compare this first, since it combines the change in conversion and order value into one number.
• Profit per order, not just revenue per order: Layer in unit and shipping costs before rolling out, since the testing tool won't factor these in on its own.

Which pricing elements are worth testing?

Not every pricing lever produces a useful result, and testing the wrong one can waste a test cycle on something that was never the real problem. In our experience, four leverage points consistently produce a measurable uplift:

• Price point: Testing the actual number matters most when you don't yet know how sensitive customers are to price. That sensitivity varies a lot from one product to another, so what works for one item won't necessarily work for another.
• Discount framing: The same discount can feel bigger or smaller depending on how it's presented. Framing it as a percentage off, a dollar amount off, or "get one free" can change conversion, even when the actual saving is exactly the same.
• Bundling: Bundling multiple products together at a discount can lift average order value, but the discount size matters. One analysis found that a 20% bundle discount still left customers preferring individual items, while a 45% discount was enough to shift preference toward the bundle.
• Billing frequency and tier structure: For subscription and recurring products, testing monthly versus annual billing, or the number and pricing of tiers, often moves retention and total revenue more than the headline price does.

5 common mistakes that ruin pricing tests

Five mistakes come up again and again in pricing experiments, and most of them are invisible until after the test has already ended:

1. Testing price in a vacuum: Ignoring external factors like seasonality, competitor moves, or broader demand shifts can make a price test look conclusive when outside forces actually drove the result.
2. Ending the test too early: Calling a winner before reaching the sample size needed for statistical confidence (often around 1,000 visitors per variation) risks acting on noise rather than a real signal.
3. Over-complicating the test: Changing the price alongside other page elements at the same time makes it impossible to know which change actually drove the result.
4. Optimizing for conversion rate instead of revenue: Conversion rate still matters, but it shouldn't be the deciding factor. A price cut can boost conversion while quietly reducing average order value and total revenue per visitor, the opposite of the intended outcome.
5. Overlooking profitability and customer quality: A test can look like a clear revenue win while margin erodes underneath it, since standard testing tools don't factor in shipping costs, unit costs, or the quality of the customers a lower price attracts.

» Want to avoid these mistakes altogether? See the most common A/B testing mistakes

The bottom line on pricing tests

Always weigh a pricing test against total profit. A price change can win on conversion, win on revenue, and still lose money once real costs are factored in, so profit is the number that should decide whether a pricing test actually worked.

This is also where a managed approach can help, particularly for sites with many product pages, where some get plenty of traffic and others don't get enough to test on their own. CROforce structures and runs pricing tests across single products and larger catalogs, and knows how to combine results across low-traffic pages into one reliable read rather than testing each in isolation.

» Ready to see how a managed pricing test works in practice? Book a demo with CROforce