The Friday restock test: a free gift or a discount

Two versions of the same email go out Friday, August 28, to the same type of customer, at the same send time. The only thing that changes between them is the offer.

• Version A gives a free gift of the customer's choice when they spend $39.
• Version B gives $15 off when they spend $39.

Everything else, the audience, the subject structure, the products featured, the send time, is held constant. So whatever difference shows up in the results can be credited to the offer mechanic itself, not to some other variable riding along with it.

Included: our lapsed reorder cohort, customers who are about a month past their last order. 1,187 profiles are in Version A and 1,181 profiles are in Version B, split roughly 50/50 by a seeded random assignment.

Excluded from both versions: six suppression segments, recent-offer recipients, active subscribers, unsubscribed contacts, bounced addresses, and our standing suppression cages.

One thing worth saying plainly: this audience is defined on purchase history, when someone last bought something, never on whether they've opened an email. That's what keeps the test honest.

Version A's subject and the top of the email say "a free item worth $9.99 to $17.49," not just "$17.49." That's deliberate, not a hedge.

Everyone who takes the gift gets exactly one item, at exactly that item's price: the chips at $9.99, the bread at $16.99, or the pizza crust at $17.49. The range is the true number for every single person who takes the gift, because it covers whichever one they actually pick.

Quoting only $17.49, the top of the range, would be accurate for roughly one customer in three and would make the gift look bigger than the discount across the board. On the mix of items people actually choose, the gift is worth slightly less than $15 on average, not more. Showing the honest range keeps the two versions of this email comparable on the merits, which is the whole point of running the test in the first place.

The main number we're watching is revenue per recipient, since it captures both whether someone buys and how much they spend.

Secondary reads: average order value, redemption rate, and how different customer segments respond to each mechanic.

This test doesn't stand alone. It pools with the Labor Day launch, which runs the same discount-vs-incentive split on the same three-SKU gift set. Together, the two tests carry more weight than either one would on its own.

Chips, multi-grain wheat bread, and pizza crust are three of the products people most often come back for. They travel well together in one box, so whichever one someone picks (or whichever the discount nudges them toward), it ships the same way with nothing extra. They are also a set we already know customers reorder often, so featuring them here is showing people what they are actually running low on, not just what we want to sell.

One important detail: the gift does not add itself to the cart. The customer picks one, adds it themselves, and the code brings that item's price to $0 at checkout.

Both versions end Sunday, August 30, 11:59pm Mountain. Same minimum, same three products, same deadline in both versions; the only real difference is whether the "thank you" shows up as a free item or as money off.

The test now has a second touch: a last-call email to everyone who has not purchased yet, sending Sunday morning, August 30. Each recipient gets the last call for the arm they were already in, gift-final to Version A recipients, dollar-final to Version B recipients, so no one crosses arms mid-test. Same offers, same minimum, same deadline: both close that night, 11:59pm Mountain.

Deadline reminders have historically matched or beaten their opening sends in this account, which is the reason a last call exists at all.