# We're A/B testing our own Black Friday price. Here's the exact setup.

Source: https://leverprice.com/blog/black-friday-price-test
Updated: 2026-11-17

> From November 24 to December 2, 2026, visitors to Leverprice's pricing page see one of three yearly Pro prices, $149, $129 or $99, instead of $199. Each visitor keeps one price, crawlers always see $199, and we judge the prices by revenue per visitor: $99 must sell 50.5% more per visitor than $149 to win.

From 00:00 UTC on Tuesday, November 24, to 08:00 UTC on Wednesday, December 2, people who open our [pricing page](https://leverprice.com/pricing) each get one of three prices for a year of Leverprice Pro: $149, $129 or $99. The list price is $199. Each visitor keeps the price they drew, and we will compare the three on revenue per visitor.

We run the test on Leverprice itself, with the same SDK our customers install. This post is the whole setup, published a week before the window opens, including the rule we will use to read the result. The second half is a playbook for running your own. The short version of the offer is on our [Black Friday page](https://leverprice.com/black-friday).

## What exactly are we testing?

Three discounts on one plan, yearly Pro:

| Yearly Pro price | Off the $199 list price | About per month | Saving vs. 12 months at $29 ($348) |
|---|---|---|---|
| $199 (list price, outside the window) | none | $17 | 43% |
| $149 | $50 (25%) | $12 | 57% |
| $129 | $70 (35%) | $11 | 63% |
| $99 | $100 (50%) | $8 | 72% |

Each discount covers the first year only. Renewal is at $199 for everyone. Monthly Pro at $29 and Lifetime at $299 are never discounted, and their checkouts don't accept discount codes at all.

Inside our own Leverprice project, yearly Pro is a product with a $199 standard price and a $99 to $199 range, and the experiment has three variants: $149, $129 and $99. Purchases reach it the way they would for any customer. Polar, which runs our billing, sends a signed webhook to our own Leverprice project, so the test only counts revenue Polar has confirmed.

There is no $199 arm. Everyone inside the window gets a discount, so this test compares discount depths and can't tell us whether discounting beats not discounting. A $199 arm would mean some visitors see no deal at all during a sale we announced publicly.

## How does each visitor get one of the three prices?

Our API asks our own SDK, and the site's edge worker writes the answer into the page.

1. A visitor requests the pricing page. Our site's Cloudflare Worker fetches the static HTML and, at the same time, asks our API for this visitor's yearly Pro price, with a 400 ms budget.
2. The API calls `@leverprice/sdk` with our own key. The SDK assigns the visitor to a variant and signs the assignment into a first-party cookie, `__pp_site`, set on our parent domain, so the marketing site and the app read the same assignment.
3. If the answer arrives in time, the Worker rewrites the price on the page, reveals a line such as "Black Friday: $129/year (was $199) until Dec 1", and passes the cookie on. The response is marked private and varies by cookie, so no shared cache can hand one visitor's price to another.
4. If the answer doesn't arrive within 400 ms, or arrives malformed, the visitor gets the static page with $199 already in it.

The Worker forwards a short allowlist to our API: the user agent, the language header, prefetch hints and our own two pricing cookies. A visitor's other cookies stay at the edge. When the API sets no cookie, as happens outside the window, the Worker shares one pricing response among anonymous visitors for 30 seconds instead of asking again for each of them.

The page says "until Dec 1" while the window closes on December 2 because 08:00 UTC on December 2 is midnight at the end of December 1 in California. The window opens at 00:00 UTC on November 24, which is the afternoon of November 23 there.

## How do we keep the page price equal to the checkout price?

With three locks. Each one, on its own, sends the visitor to $199.

1. The window is hard-coded: 00:00 UTC on November 24 to 08:00 UTC on December 2. Outside it, nobody is even enrolled, and everyone sees $199 whatever state the experiment is in. Nobody has to remember to stop it.
2. A price can appear only if we configured a matching Polar discount for it. A variant without a discount behind it never reaches the page.
3. That discount must be live in Polar right now. Its start and end dates are checked against Polar itself, cached for 5 minutes, and any doubt means $199.

The checkout runs the same logic with the same signed cookie and applies the matching Polar discount, so a discounted price on the page is the price Polar charges.

The fallback always errs toward the list price. A slow answer can hide a discount for one page view, but it can't show a discount the checkout won't honor. The end is hard as well: a checkout started after 08:00 UTC on December 2 is created at $199, even if the page was loaded a minute earlier, and Polar's checkout page shows that amount before anyone pays.

Crawlers, link previews and monitoring bots always see $199 and are never enrolled, so search results and shared links show the list price. We explain why that matters in [does A/B testing prices hurt SEO](https://leverprice.com/blog/does-ab-testing-prices-hurt-seo).

## Why test only the yearly plan?

Because a first-year discount on a yearly plan has a known, bounded cost. The buyer pays $149, $129 or $99 once and renews at $199, so the discount can't quietly turn into our permanent price.

A monthly discount has no good shape. One cheaper month barely changes the decision, and a permanently cheaper month changes our revenue forever. Monthly at $29 also stays the low-commitment way in: paid periods are non-refundable, so people who want to try Leverprice first can do it on Free or on monthly.

Lifetime is already a one-time offer: 50 seats at $299, with no renewal to recover a discount from. And one plan keeps the test readable. A single number changes on the page, and our traffic splits three ways instead of across every plan and price. For the general trade-offs between yearly and monthly pricing, see [annual vs monthly discounts](https://leverprice.com/blog/annual-vs-monthly-discount).

## Why judge by revenue per visitor instead of conversion rate?

Because a lower price almost always converts better, so conversion rate would crown $99 before the test even started. [Revenue per visitor](https://leverprice.com/glossary/revenue-per-visitor) multiplies the two:

revenue per visitor = conversion rate × price

A cheaper price wins only if its conversion rate grows by a bigger factor than its price shrinks. For two prices, the cheaper one needs a conversion lift greater than the expensive price divided by the cheap one, minus one:

| Comparison | Extra buyers per visitor the cheaper price needs |
|---|---|
| $129 vs $149 | 149 / 129 - 1 = 15.5% |
| $99 vs $129 | 129 / 99 - 1 = 30.3% |
| $99 vs $149 | 149 / 99 - 1 = 50.5% |

Illustration: 2,000 visitors see each price. $149 converts at 1.0%, so 20 sales bring $2,980, or $1.49 per visitor. $99 converts at 1.4%, so 28 sales bring $2,772, or $1.39 per visitor. The cheaper price sold 40% more and still earned less.

*Figure: Illustration: as the price rises, conversion falls steadily while revenue per visitor rises, peaks and then falls. Illustration: the price that sells the most is rarely the price that earns the most per visitor.*

The [price increase calculator](https://leverprice.com/tools/price-increase-calculator) handles cuts too: enter your list price and a sale price, and it shows the lift the sale price needs.

## What would make us pick $99 over $149?

$99 has to earn more per visitor than $149, which takes at least 50.5% more buyers per visitor, and the engine has to be sure of it. The sale ends on December 2 whatever the data says, so picking a winner here means choosing the depth we would run again.

Leverprice calls a winner when one price is at least 95% likely to be the best, or when the remaining prices are practically equivalent, meaning the expected relative loss of picking either is under 0.5%. Only visitors older than 30 minutes count, so a price that just received a burst of fresh traffic doesn't look worse while those visitors are still deciding.

Our rule, written down before the window opens: if $99 is the clear winner, $99 it is. If $99 and $149 come out practically equivalent, we lean to $149, because the same first-year revenue then comes with a smaller jump at renewal, 34% instead of 101%.

That renewal jump is the part an eight-day test can't see. A buyer's real value is the first-year price plus whatever their renewals bring, and the first renewals arrive in late 2027.

Illustration, with renewal rates nobody knows yet. Value per visitor = conversion × (first-year price + renewal rate × $199).

| Assumed renewal rate at $199 | Extra buyers per visitor $99 needs over $149 |
|---|---|
| Ignored (first year only) | 50.5% |
| 50% for both prices | (149 + 99.5) / (99 + 99.5) - 1 = 25.2% |
| 50% for $149 buyers, 30% for $99 buyers | (149 + 99.5) / (99 + 59.7) - 1 = 56.6% |

If every price renewed at the same rate, renewals would favor $99, because each extra buyer brings a full-price renewal. If $99 buyers renew less often, renewals can tip the result back toward $149. On December 2 we will know first-year revenue per visitor, and we will treat it as exactly that.

We plan to share what we learn after the window closes. If eight days of our traffic aren't enough to separate the prices, that will be the finding.

## Should you run a Black Friday discount at all?

Only if it brings in buyers who wouldn't have bought at full price, and enough of them to clear the break-even. A buyer who would have paid $199 in December and pays $129 in November is a $70 loss.

Who buys from you matters more than the date. If people buy your product the day they hit the problem, a November sale mostly reprices purchases that were coming anyway. If your buyers watch for deals and wait for them, the sale is when they show up.

A discount also sets a reference point. Someone who saw $99 remembers $99, so keep it tied to a date and to the first year, and say what renewal costs.

Each discount depth has a fixed break-even against your list price: extra buyers needed = 1 / (1 - discount) - 1.

| Discount from list price | Extra buyers per visitor needed to match list-price revenue |
|---|---|
| 20% | 25% |
| 25% | 33% |
| 35% | 54% |
| 50% | 100% |

## How deep should a Black Friday discount be?

As deep as earns the most per visitor, and the honest way to find that depth is to test it instead of guessing. Run two or three depths at once, keep each visitor on one of them, and compare revenue per visitor.

Make every arm a discount from your list price, as ours are, so nobody pays more than the list price because of your test. With thin traffic, test two depths instead of three, since every extra price splits the same visitors more ways.

If you'd rather not build the sticky assignment and the scoring yourself, that is what [Leverprice](https://leverprice.com/price-testing) does: your server asks for a price, each visitor keeps theirs, and traffic moves to the price that earns the most per visitor.

## How do you protect existing customers?

Discount new purchases only, and leave everyone's current price alone. Nobody's price should go up because of a sale.

Keep the discount to the first year so renewal costs the same for everyone. Leave commitment plans out: we don't discount Lifetime, where every one of the 50 seats costs $299 whoever buys it.

Decide before the sale what you will do for someone who paid full price last week. Extending their term by the difference, or offering the sale price at their next renewal, both cost less than an argument in a support thread. Customers who keep the price they signed up at are covered by your [grandfathering](https://leverprice.com/glossary/grandfathering) policy, so write one down if you don't have it.

## When should the sale run, and how should you announce it?

In 2026, Thanksgiving falls on November 26, Black Friday on November 27 and Cyber Monday on November 30. Our window opens at 00:00 UTC on Tuesday, November 24, and closes at the end of Tuesday, December 1, California time, so it covers the whole week around both days.

Put the end on the page in your buyers' terms, since a UTC timestamp means little to someone deciding whether to buy tonight. Show the regular price next to the deal ("was $199"), and say what renewal costs.

Make the end real. When a sale is announced to end, the price should go back at that moment. Ours is in code, so it does.

## A checklist for your Black Friday price test

1. Write down the break-even lift for each discount depth before launch.
2. Make every tested price a discount from your list price.
3. Keep each visitor on one price, and keep crawlers on the list price.
4. Make the checkout charge exactly what the page showed, and fall back to the list price on any doubt.
5. Hard-code the start and end in UTC, and show the end in your buyers' time zone.
6. Discount the first year only, and say what renewal costs.
7. Decide now what you will offer customers who paid full price last week.
8. Judge on revenue per visitor from verified purchases, with the decision rule written down before launch.

## Frequently asked questions

### Can I get a different price by clearing my cookies?

Possibly. A browser without our cookie counts as a new visitor and is assigned again, to one of the same three prices. All three are discounts from $199.

### Is it fair that two people see different prices?

Everyone in the window gets a discount from the same $199 list price, and nobody pays more than the list price because of the test. Assignment is random per visitor and doesn't depend on who you are.

### What happens if Leverprice itself goes down during the sale?

Visitors see $199. The page waits at most 400 ms for a price before serving the static list price, and the checkout applies a discount only when all three locks agree.

### Is a 50% Black Friday discount too deep for a software product?

It depends on the lift it brings. Against your list price, a 50% discount needs twice the buyers per visitor just to match the revenue you'd have made without it.

### How many prices should a small product test?

Two, if traffic is thin. With two prices each one gets half your traffic instead of a third, so it collects the visitors a decision needs sooner.
