What to A/B Test in Bundle Offers, Ranked by Likely Impact

Test the things that change what the shopper pays and how much they buy before the things that change how the offer looks. In order: the tier points, the depth of the discount, which option is selected by default, the mechanic itself, the placement, and last the wording. Change one variable at a time, run each test for whole weeks, and pick the winner on profit per visitor. This guide explains each test, why it sits where it does in the ranking, and how to read the result.

What should you test first?

Start with the offer's economics, because they move both the customer's decision and your margin. Cosmetic tests can only move the first.

The ranking, from most to least likely to matter:

1. Tier points: which quantities or sizes the offer is built around.
2. Discount depth: how much each tier saves.
3. Default selection: which option is chosen when the page loads.
4. Mechanic: a bundle against a quantity break, or a gift against a discount.
5. Placement: where the offer sits on the page.
6. Wording: names, labels and the way the saving is phrased.

How do you test tier points and discount depth?

As two separate tests, tiers first.

Tier points. Compare structures such as one, two and three units against one, three and six. The question is where your customers' natural stopping point lies. A tier just above what people already buy pulls them up; a tier far above it is ignored. Hold the per-unit discount roughly constant across the two versions so that you are testing the quantities alone. How to choose the candidates is covered in setting quantity break tiers.

Discount depth. Once the tiers are settled, test how much the saving needs to be. Compare, for example, 10 percent against 15 percent on the middle tier. The deeper discount will nearly always sell more units. The test exists to find out whether it sells enough more to pay for itself, which is why the result has to be read in profit.

Does the default selection deserve its own test?

Yes, and it is the cheapest test on the list, because nothing about the price changes. Compare the page loading with one unit selected against the page loading with the middle tier selected.

Watch two things beyond profit. First, the conversion rate of the page: a preselected larger pack raises the first price the shopper sees, and some will leave. Second, returns and cancellations: if people are buying more than they meant to, the gain will reverse a few weeks later.

When should you test the mechanic itself?

After the first three, and only when you have a real doubt about the offer type. Mechanic tests compare different offers altogether:

• a quantity break against a fixed bundle with a companion product;
• a percentage discount against a free gift of similar cost to you;
• an add-on beside the button against a frequently-bought-together panel.

These are harder to read because several things change at once: the products in the order, the cost of goods, the picking work. Count all of it in the profit figure. The guide to Shopify bundle types sets out which mechanic suits which buying pattern, and that reasoning will give you the candidates.

What about placement and wording?

Test them last, and test differences big enough to matter.

Placement. Above the add-to-cart button against below it; on the product page against in the cart.

Wording. 'Save $12' against '20 percent off'; 'Most popular' against no label; a named bundle against a plain description. Wording tests tend to produce small differences, so they need more visitors than the others to give a clear answer.

Why only one variable at a time?

Because a test that changes two things cannot tell you which one worked. If version B has new tiers and a deeper discount and wins, you do not know whether you could have kept the old discount and the margin that went with it.

Keep everything else fixed during the test:

• no sitewide sale starting halfway through;
• the same traffic sources feeding both versions;
• both versions live in the same weeks, with visitors split at random between them.

How long should a test run?

For whole weeks, and for long enough that each version has a meaningful number of orders, not just visits.

• Whole weeks, so each version sees the same mix of weekdays and weekends.
• At least two of them, so one unusual week does not decide the result.
• A fixed end date set in advance. Stopping the moment one version edges ahead is how false winners are crowned. Early leads often reverse.
• Longer for consumables. A quantity break pulls future purchases forward, so part of its true cost appears as a later gap in reorders.

On Marka Bundles & Upsells, running experiments are included in the Growth and Scale plans, with a higher number of simultaneous experiments on Scale.

Why judge on profit per visitor?

Because it is one figure that accounts for conversion, order size and discount cost together.

Profit per visitor = total profit from a version's orders ÷ visitors who saw that version.

Say each version is shown to 1,000 visitors.

• Version A: 40 orders at $20 profit each. Total $800, or $0.80 per visitor.
• Version B: 36 orders at $24 profit each. Total $864, or about $0.86 per visitor.

Version B converts worse and still wins. Judged on conversion rate you would have chosen A.

Profit here means revenue after discounts, less cost of goods, shipping and payment fees. The full definitions are in our article on bundle metrics: attach rate, AOV and profit per order.

A short checklist

1. Test in order: tiers, discount depth, default, mechanic, placement, wording.
2. Change one variable per test.
3. Split visitors at random between versions running at the same time.
4. Run for whole weeks, at least two, to an end date set in advance.
5. Judge on profit per visitor, with every cost counted.
6. Check returns and reorders before declaring a default or quantity-break winner.

A short list of well-run tests on tiers and discount depth will teach you more than months of label changes. If you want the testing built into the offer tool, see the plans for Marka Bundles & Upsells.