Klaviyo A/B testing: campaign and flow method

Short answer. Write one hypothesis, change one meaningful variable, randomly split the same eligible audience, choose the primary metric and evaluation window before launch, and keep unsubscribe, complaint, and margin as guardrails. Treat Klaviyo's winner label as evidence for that test context, not a permanent rule for every audience.

A/B testing is useful when it changes a decision. It is not useful when variant A is a red button with a new subject line, different offer, later send time, and shorter email while variant B changes none of those things.

Klaviyo supports testing across campaign, flow, and form use cases. The exact options and winner metrics vary by asset, so define the business question before opening the builder.

The experiment brief

Complete this table first:

If the team cannot fill this table, it is not ready to test.

What to test first

Prioritize variables close to a customer decision:

1. Audience and lifecycle state.
2. Offer or value proposition.
3. Timing after a behavior.
4. Message count and sequence.
5. Product or category relevance.
6. CTA and creative hierarchy.
7. Subject line and preview text.

Subject lines are easy to test, but they optimize an imperfect open signal unless clicks or orders select the winner. Apple Mail Privacy Protection can generate machine opens, so an open-rate winner may not represent more human attention.

Choose the winning metric from the hypothesis

One practical constraint: the campaign builder only offers three winner metrics, open rate, click rate and placed order rate, the last one on eligible accounts only. Revenue or contribution per recipient is not in that selector, so pick the closest available metric for the automated winner and compute the commercial figure yourself from the results report.

The open-rate caveat is not only statistical on a French list: check what the CNIL changes for open tracking before building a protocol on that metric alone.

Do not choose the metric after seeing the charts. That turns normal noise into a story.

Statistical significance in Klaviyo

Klaviyo's current campaign significance documentation says a campaign result is labeled statistically significant when each variation has at least 50 recipients and the win probability reaches at least 90% for the selected winner metric.

That is a platform decision rule, not a universal sample-size calculator. Fifty recipients per variant is rarely enough to detect a small difference in a low-rate outcome such as orders. Required sample size depends on baseline rate, minimum effect worth detecting, number of variants, split, and desired error tolerance.

Use a pre-test sample-size calculation for high-stakes decisions. If the list is too small, run fewer variants, test a larger behavioral difference, or collect evidence across comparable sends without pretending each send is identical.

Campaign A/B test process

1. Select one eligible audience

Use the same inclusions, exclusions, consent, frequency, and send window. Randomization should be the only reason otherwise comparable profiles see different versions.

2. Create the control

The control is the current best-known version, not an intentionally weak message. A weak control exaggerates lift and wastes audience.

3. Create one meaningful variant

Change the variable described in the brief. If testing proposition, keep sender, subject, preview, layout, products, offer, audience, and send time constant where possible.

4. Set split and winner behavior

Choose how much of the audience enters the test and whether Klaviyo automatically sends a winner to the remainder. Automatic winner rollout can be useful for time-sensitive campaigns, but it also means the winner is selected on an early window. For durable learning, a full random split may provide cleaner comparison.

5. Freeze the analysis plan

Record metric, attribution settings, end time, guardrails, and excluded anomalies before the send.

6. Analyze the full chain

Review delivery, opens, clicks, onsite behavior, orders, revenue per recipient, unsubscribes, complaints, returns, and margin where relevant. A click winner with weaker conversion may have created curiosity without purchase intent.

Flow A/B testing

Flows introduce additional complexity because profiles enter over time and the business environment changes.

Test one of:

• Message content.
• Time delay.
• Number of messages.
• Branch strategy.
• Offer.
• Channel sequence.

Keep trigger, flow filters, conversion exits, Smart Sending, consent, and re-entry rules consistent across variants.

Flow email tests behave differently from campaign tests in two ways. The winner metrics are limited to open rate and click rate, and Klaviyo recommends click rate because Apple Mail Privacy Protection can inflate opens. The winner is selected as soon as the metric reaches statistical significance or at a date limit you set, whichever comes first, with manual selection still available.

Use a long enough evaluation window

A post-purchase test may need the next expected order cycle. A cart test may resolve within days. Define the observation period from customer behavior, not dashboard impatience.

Avoid contamination

Campaigns and other flows can contact the same profiles during a test. Record frequency rules and major concurrent promotions. If one variant is exposed to a different campaign mix, interpretation becomes harder.

Test incrementality, not only variants

A content A/B test answers which message performed better. It does not answer whether sending either message was better than sending none. Add a randomized holdout when the decision is whether the flow creates incremental behavior.

Form A/B testing

Optimize for valuable consent, not only raw submit rate. A form variant can increase email capture while reducing confirmed opt-in, first purchase, or long-term engagement.

Track:

• Unique form views.
• Submit rate.
• Confirmed subscription rate.
• Email and SMS consent separately.
• Welcome click and first-order rate.
• Unsubscribe and complaint rate.
• 90-day active or customer rate.

Keep targeting and traffic comparable. A mobile-only variant and desktop-heavy control are not a valid creative test.

Klaviyo's own form test reports unique views, unique submits, submit rate, orders, average order value and revenue alongside a win probability, so the platform covers the first half of that list and your own analysis covers the consent and retention half. Note that form variations cannot be edited once a test is live: the test has to be ended first, so settle the creative before launch.

What not to combine in one test

• Subject line and sender name.
• Offer and audience.
• Email body and landing page.
• Send time and segmentation.
• Flow delay and message count.
• Form creative and traffic source.

If the goal is to compare two complete strategies, label it as a package test. You can learn which package wins, but not which component caused the difference.

Build a test log

Add screenshots or exported results because live dashboards and attribution settings can change.

Interpret common outcomes

Statistically significant and commercially useful

Adopt in the tested context, monitor after rollout, and test whether the principle transfers to a comparable audience.

Significant but tiny effect

Estimate annual incremental value and implementation cost. A detectable effect may still be too small to matter.

Promising but uncertain

Retest the same hypothesis on a comparable campaign. Do not change the variant definition to chase a win.

No meaningful difference

Keep the simpler or lower-risk version. A null result prevents unnecessary work.

Guardrail deterioration

Reject or redesign even when the primary metric wins. More clicks with more complaints or lower margin is not a clean improvement.

Common Klaviyo testing mistakes

• Testing many variables at once.
• Selecting the winner metric after launch.
• Using opens without Apple privacy context.
• Stopping when one version briefly leads.
• Declaring a rule from one small test.
• Comparing segments that were not randomized.
• Ignoring attribution settings and concurrent campaigns.
• Testing content but never testing whether the flow is incremental.

Testing methodology matters less if the underlying Klaviyo setup has gaps, so consider a free 30-minute Klaviyo audit before you scale the winning variant.

FAQ

How many recipients does a Klaviyo A/B test need?

Klaviyo requires at least 50 recipients per campaign variation for its significance label, but a useful sample can be much larger. Calculate it from baseline, minimum detectable effect, and the selected metric.

What should I test first in Klaviyo?

Start with a meaningful customer decision, such as audience, offer, timing, or proposition. Test subject lines when inbox recognition is the actual question.

Should the winner be based on opens or clicks?

Choose the metric that matches the hypothesis. Clicks are usually more dependable for engagement because Apple privacy behavior affects opens. Use orders or revenue when purchase is the question.

Can I A/B test a Klaviyo flow?

Yes. Open the email inside the flow builder and select Create A/B test in the email details panel. Variations are weighted equally by default and the percentages can be changed. Keep trigger, filters, and evaluation window consistent, and distinguish variant performance from incrementality.

What if the result is not significant?

Do not force a winner. Keep the control or simpler version, record the result, and decide whether a larger or more distinct test is worth running.

Turn experiments into accumulated knowledge

Testing compounds only when hypotheses, context, results, and decisions are preserved. Deliver helps ecommerce teams build that experimentation system across campaigns and flows, or you can talk to our Klaviyo agency directly. Book a Klaviyo and CRM diagnostic.