Est. reading time: 8 minutes
Most popups we open have been running the same offer since the store launched. Ten percent off the first order, with one line of supporting copy above the email field. The tests that have run against it were button colors, headline rewrites, and image placement, which optimizes the smallest variables on the form while the largest one goes untouched.
Ten percent survives because it performs acceptably and nobody has a reason to fight it. It also carries a cost structure that rarely gets examined. The discount scales with cart size, so the orders you most wanted cost the most to acquire, and the same percentage goes out whether the subscriber was ready to buy that afternoon or was three months away from it. The offer is the largest variable on the form and usually the least tested one.
Submit rate will hand you the biggest discount every time
The default read on an offer test is submit rate, and submit rate has a predictable relationship with discount depth. Run 10% against 20% and 20% wins. Run 20% against 25% and the pattern usually holds. Optimizing to that metric walks the offer down a ladder that ends at the largest discount the margin can absorb, and the test never shows the damage, because the damage happens after the submit.
The question is not which offer collects the most addresses. It is which offer produces subscribers who buy at a price the business can sustain. Klaviyo’s A/B test results support that read directly. Each variation reports unique views, unique submits, and submit rate alongside orders, average order value, and revenue from placed orders that followed a form submission within the lookback window. A variation can lose on submit rate, win on average order value, and still be the one to keep.
Two conditions make those numbers trustworthy. The test has to run long enough for the welcome flow to send and for a normal purchase cycle to close, which is a different length for a consumable than for a considered purchase. It also needs enough volume that a handful of large orders are not setting the direction. Six orders behind a variation will not tell you anything you can act on.
Offers that change the shape of the first order
Free shipping is the first alternative to run, because it removes a cost the shopper meets late in checkout and its cost to the brand is fixed per order, so it does not grow with cart size. It performs hardest for heavy or bulky products where delivery cost is a real objection. It gets weak on stores that already ship free above a threshold most carts clear, since the popup is then giving away something the site gives away anyway.
A dollar amount attached to a minimum does work a percentage cannot. “$15 off your first order over $75” sets a floor under the first purchase, protects margin on small carts, and hands the shopper a target to build toward. It works when unit prices sit comfortably under the threshold, so clearing it means adding a second item. It fails when a single unit already exceeds the minimum, at which point you have written a plain dollar discount with extra words.
Gift with purchase moves the cost from revenue to cost of goods. A sample, a travel size, or an accessory carries perceived value well above what it costs to include, and it puts a second product in the box, which matters for any brand whose repeat rate depends on trial across the range. The constraint is operational. If the warehouse cannot reliably attach the item, the offer generates support tickets and the second order never arrives.
The offers that do not discount at all
Early access works when the scarcity is real (restocks, drops, limited runs, seasonal capsules). If the brand never sells out of anything, the offer is a promise the list tests once and then ignores.
For considered purchases, information can beat money. A fit guide, a materials comparison, or a short quiz gives the visitor something they were already trying to work out, and the quiz has a second payoff. The answers land on the profile, so the welcome email can open with the product that matches the answer, and the segment stays useful long after the welcome series ends. Sweepstakes sit at the other end of the range. They collect volume quickly and collect indifferent subscribers just as quickly, so they belong in the test set only if you are willing to judge the resulting cohort on revenue and treat list size as secondary.
Choosing which alternative to run first is an economics question. If shipping is a visible line item that shoppers meet after they have committed, test free shipping. If the average cart sits just below a natural bundle point, test the dollar amount with a minimum. If margin is thin but the range is wide and trial drives repeat purchase, test the gift. If the purchase is high-consideration and infrequent, test information against money before you test discount depth.
Popup offer testing breaks in three predictable places
The first is variable creep. A test that changes the offer, the headline, and the image tells you a combination won and leaves you nothing to carry into the next test. Klaviyo’s own guidance is to move one element at a time, and the platform enforces some of that discipline for you, since form content cannot be edited once a test is live. Changing the copy means ending the test and starting over, so both variations need to be finished before you publish. Trigger timing and targeting are separate variables with their own test plan, and folding them into an offer test produces a result nobody can interpret.
The second is traffic weighting. By default Klaviyo splits traffic evenly and then shifts more of it toward the variation that is pulling ahead while the test is still running. That behavior grows the list faster, and it thins the sample on the trailing variation, which is the one you need order data from before you can compare revenue per subscriber. When the metric that matters sits downstream of the submit, set the weighting to manual and hold it even.
The third is delivery. Two offers on the form require two paths in the welcome flow, and Klaviyo’s documented setup is to attach a hidden profile property to each variation on submit, then use a conditional split so each subscriber receives the message carrying the offer they saw. Skipping that step is how a free shipping variation ends up sending a percentage code, which tests nothing at all. The welcome sequence behind the form deserves its own audit, since an offer performs only as well as the email that delivers it.
Two smaller mechanics are worth knowing before you publish. Visitors are cookied on first view, so a returning visitor keeps seeing the variation they were first served. And if the form is multi-step, the results break engagement out by step for each variation, which separates an offer problem from a form problem. A variation where people clear the first step and then abandon has a field or a step count issue, and the offer is not the thing to change.
What a winning offer does to the rest of the account
A code that wins a test becomes a standing feature of the site, and two things follow from that. Static codes travel. If the same string works for everyone, it reaches coupon extensions and deal sites, and the welcome discount quietly becomes the catalog price. Unique single-use codes keep the offer attached to the subscriber who earned it.
Then there is interaction with everything else running. During a sitewide promotion, a popup still offering the welcome discount either stacks or confuses, and both outcomes cost margin on orders you would have won anyway. Suppressing the form during promotional periods, or swapping in something that does not compound (early access to the sale, for instance), keeps the welcome offer pointed at people who are new.
Offers are worth retesting on a cadence, because the winner from two years ago won against a different catalog, different shipping costs, and a different mix of traffic. Pick the one alternative that matches your economics, run it against the current control at even weighting, and judge it on what the cohort spends over a full purchase cycle. If the offer economics or the flow routing could use a second set of eyes, get in touch.
