# How to test prices with low traffic (fewer than 100 sales a month)

Source: https://leverprice.com/blog/price-testing-low-traffic
Updated: 2026-11-17

> Test prices far apart. With 3% of visitors buying, telling $29 from $32 takes about 215,000 visitors per price; $29 against $49 takes about 8,500. Test the plan that earns most, pool every page that shows the price, treat a planned tie as a win for the higher price, and below 30 sales a month use surveys and fake doors.

At 100 sales a month, a test of $29 against $32 can need more than ten years to give a reliable answer, while $29 against $49 on the same traffic needs about five months (the worked example is below). With low traffic, small questions are out of reach and big ones are not.

The work is choosing questions a few hundred sales can answer, then getting the most out of every visitor who sees a price. Here is the math, the strategies that follow from it, and what to do at each sales level.

## Why does a bigger price gap need fewer visitors?

Because a bigger gap usually produces a bigger difference in how many people buy, and the visitors you need fall with the square of that difference. The [standard two-proportion formula](https://select-statistics.co.uk/calculators/sample-size-calculator-two-proportions/) gives the visitors needed per price:

```text title="Visitors needed per price"
n = (z₁ + z₂)² × [p₁(1 − p₁) + p₂(1 − p₂)] ÷ (p₁ − p₂)²

p₁, p₂   share of visitors who buy at each price
z₁ = 1.96 for 95% confidence, z₂ = 0.84 for 80% power, so (z₁ + z₂)² ≈ 7.85
```

Example: 3.0% of pricing-page visitors buy at $29. Assume one demand curve gives 2.86% at $32, 2.59% at $39 and 2.31% at $49, which is what a constant [price elasticity](https://leverprice.com/glossary/price-elasticity) of −0.5 produces: buyers are fairly insensitive to price. At 100 sales a month, the pricing page gets about 3,300 visitors a month.

| Test | Buy at the higher price | Revenue per visitor | Visitors per price | Sales in the test | Time at 100 sales a month |
|---|---|---|---|---|---|
| $29 vs $32 | 2.86% | $0.87 vs $0.91 | 214,900 | 12,600 | about 11 years |
| $29 vs $39 | 2.59% | $0.87 vs $1.01 | 25,000 | 1,400 | about 15 months |
| $29 vs $49 | 2.31% | $0.87 vs $1.13 | 8,500 | 450 | about 5 months |

Widening the gap from 10% to 69% cuts the visitors needed about 25-fold. The formula answers "did the price change how many people buy?", while the question you care about is [revenue per visitor](https://leverprice.com/glossary/revenue-per-visitor). The revenue version, `n = 7.85 × [P₁²p₁(1 − p₁) + P₂²p₂(1 − p₂)] ÷ (P₂p₂ − P₁p₁)²` with P as the price, gives 215,400 and 9,100 visitors per price for the first and last rows, so the conclusion holds.

*Figure: Line chart: the visitors needed per price fall steeply as the difference you want to detect grows, for products where 2% and 5% of visitors buy. Visitors needed per price at 95% confidence and 80% power. With 2% of visitors buying, telling a 20% difference apart takes about 21,000 visitors per price.*

A shortcut makes this easier to plan in sales, which you know, rather than visitors. When few visitors buy, p(1 − p) is close to p, and the formula reduces to:

```text title="Sales needed across both prices"
sales ≈ 7.85 × (2 − d)² ÷ d²

d   relative drop in the share of visitors who buy at the higher price
```

A 10% drop needs about 2,830 sales across both prices, 20% about 640, 30% about 250 and 50% about 70. The conversion rate drops out of the shortcut, which is why the decision table at the end is written in sales. To run your own numbers, use the [price test sample size calculator](https://leverprice.com/tools/price-test-sample-size-calculator).

## Which price should you test first?

The one that carries most of your revenue. Every sale in a small test is scarce, so spend it where the answer is worth the most, often the annual plan or the top tier. Example: if annual Pro brings in 70% of revenue, a price that earns 20% more per visitor there lifts total revenue by 14%. The same gain on a plan that brings in 10% lifts it by 2%.

Measure revenue per visitor across all plans, including the ones you didn't change. A higher monthly price can push some buyers to annual, and a higher top tier can push some down a tier. Only the total tells you whether the change earned money.

## How do you get more visitors into a small test?

Pool them. Use one assignment everywhere the price appears (pricing page, homepage, upgrade prompts, checkout) so every visitor who sees a price counts, and a visitor who meets the price on three pages counts once, with one price.

If you sell several plans, test one change across the whole price list, such as every plan 40% higher, instead of one plan at a time. Every purchase on any plan then feeds the same comparison, and you judge it by total revenue per visitor.

Count visitors when they see a price, not when they land on the site. A blog reader who never opens the pricing page can't react to the price; counting them makes the test look slower, and if some of them buy through another path, they blur the difference you're trying to measure.

## When is "no clear winner" a good result?

When you planned for it, because a tie favors the higher price. Set the stopping point before you start, for example 450 sales for $29 against $49. If you reach it and neither price is clearly ahead on revenue per visitor, the difference is probably smaller than the one your test was sized to catch.

Then take the higher price. Example: 100 customers at $29 bring in $2,900, and at $49 about the same revenue takes 59 customers. That is 41% fewer accounts to onboard and support, and 41 fewer fixed card fees (30¢ each on [Stripe's US pricing](https://stripe.com/us/pricing) as of November 2026).

You can also lower the bar. A price is easy to change back, so 95% confidence may be more than you need. Testing at 90% swaps 7.85 in the formula for (1.645 + 0.84)² ≈ 6.18, which is about a fifth fewer visitors; 80% confidence needs about 43% fewer. The cost is more false winners, which matters less when the old price is one change away.

Leverprice's automatic optimization uses the same idea. It calls a winner when one price is at least 95% likely to be best, or when the remaining prices are practically equivalent, meaning the expected revenue lost by picking one is under 0.5%.

## Can you test prices one after another instead of at the same time?

Yes, with care. A before/after test, such as four weeks at $29 followed by four at $49, needs no tooling to split traffic, which is why founders reach for it at very low volume. It needs as many sales as a split test for the same answer, and everything else that changed between the two periods lands in your result. If you can split traffic, do: both prices then live through the same weeks. If you can't, watch for the usual culprits:

- Seasonality, from weekday patterns and month-end budgets to summer, holidays and Black Friday.
- Launches and press. A Product Hunt launch, a newsletter mention or a Hacker News thread brings visitors who buy differently.
- Your own campaigns, because discounts, ads and emails change who arrives.
- Growth itself, since a product that is gaining traction can convert differently in month two.

You can shrink the damage:

1. Run whole weeks, and the same number of them for each price.
2. Keep launches, campaigns and holidays out of both periods. If one lands anyway, extend the test or drop that week.
3. Alternate: $29, $49, $29, $49 in one- or two-week blocks, so trends hit both prices.
4. Compare conversion by traffic source, so a burst of visitors from one forum doesn't decide your price.
5. Watch a number the price shouldn't move, such as free signups per visitor. If it shifts between periods, your traffic changed, and so will your result.
6. Change prices for new customers only, and honor the lower price for anyone who saw it.

## How can a survey tell you which prices to test?

A Van Westendorp survey gives you a direction and a range before you spend sales on a test. It asks four questions about your product: at what price it would seem so cheap you'd doubt the quality, a bargain, getting expensive, and too expensive to consider. The [Van Westendorp calculator](https://leverprice.com/tools/van-westendorp-calculator) turns the answers into curves and an acceptable price range.

Ask people who match your buyers, such as trial users, recent signups and newsletter readers. Treat the result as stated intent rather than behavior, since nobody pays to answer a survey. Use it to pick the two prices you'll test, ideally near the two ends of the acceptable range, where the difference in behavior is large enough to see.

## How do you test a new tier without building it?

With a fake door. Add the tier to your pricing page with a price, and send its button to a page that says the tier is coming and offers a waitlist. Count clicks per pricing-page visitor against the plans you already sell, and if traffic allows, show two prices for the new tier to two halves of your visitors.

Be honest about it. Label the tier "coming soon" or make the button say "Join the waitlist", never take payment, and remove it once you've learned what you needed. A plan that looks purchasable but isn't comes close to the bait advertising that EU and UK law ban, as covered in [is price testing legal](https://leverprice.com/blog/is-price-testing-legal). Clicks overstate purchases, so read the numbers against your existing plans' click rates, not as sales.

## What signals can you use besides a test?

The ones you already have. None is proof on its own, but together they tell you which way to move:

- How often price comes up in sales calls and demos, and at what stage.
- Cancellation and refund reasons that mention price.
- The plan mix, meaning the share of buyers who pick the top tier and how many choose annual.
- Discount requests, including who asks and how hard they push.
- Time from signup to purchase. Quick, unhesitating purchases on every tier can mean the price is low for the value.
- What prospects compare you with, and the prices they quote back to you.

## What should you do at each sales level?

Match the method to the number of paid orders you get in a month. The middle columns assume a three-month test of two prices.

| Paid orders a month | Sales in a 3-month test | Smallest drop in buyers you can detect | What to do |
|---|---|---|---|
| Under 30 | Under 90 | About 46% or more | Choose the price with a Van Westendorp survey, fake doors and customer conversations. If you test, run two prices far apart for several months and read the result as a direction, not a verdict. |
| 30 to 100 | 90 to 300 | About 28% to 46% | Two prices far apart, like $29 against $49, on the plan that earns most, with every pricing page pooled and a stopping point and tie rule set in advance. |
| 100 to 300 | 300 to 900 | About 17% to 28% | Two or three prices 30% to 50% apart, or the whole price list moved at once. A bandit becomes worth considering, since it shifts traffic toward the leader while the test runs. |
| 300 and up | 900 and up | Under 17% | Several prices and finer steps, with automatic optimization moving traffic toward the best price as it learns. |

The detectable drop is the relative change in the share of visitors who buy, at 95% confidence and 80% power, from sales ≈ 7.85 × (2 − d)² ÷ d².

Leverprice Free is built for the top of this table: it runs a manual split on one product for up to $500 a month in tested revenue, which is about 17 sales at $29. Pro adds automatic optimization, which moves traffic toward the price with the highest revenue per visitor and calls a winner as described above. Both are on the [pricing page](https://leverprice.com/pricing).

## Frequently asked questions

### Can you A/B test prices with 10 sales a month?

Not to a statistical answer in reasonable time: even a 50% drop in buyers takes about 70 sales to detect, seven months at that pace. Choose the price with a Van Westendorp survey and fake doors, and if you test, compare two prices far apart and treat the result as a direction.

### Is 90% confidence good enough for a price test?

Often, yes. A price is easy to change back, and 90% confidence needs about a fifth fewer visitors than 95%. In exchange, you accept more false winners.

### Should a low-traffic test count trials or paid orders?

Paid orders, net of refunds. Trial starts arrive faster but measure interest, and price can change who converts after the trial as well as who starts one.

### How long should a low-traffic price test run?

Until it reaches the sample you planned, in whole weeks. Set a maximum, such as three months, and treat a tie at that point as a vote for the higher price. More on this in [how long a price test should run](https://leverprice.com/blog/how-long-should-a-price-test-run).

### Does a multi-armed bandit help with low traffic?

It loses less revenue during the test by moving traffic toward the leading price, but it can't make a small sample more informative. At very low volume, a fixed split with a wide gap is simpler; the trade-offs are in [bandit vs A/B testing for prices](https://leverprice.com/blog/bandit-vs-ab-test-pricing).
