Retargeting earns its place when it helps an eligible visitor resolve a buying question and produces enough added value to justify its cost. A previous visit is a useful signal of interest, not proof that another ad or a discount is needed. Start with the hesitation, then choose the message and measurement.
A fictional cart that needs information, not another coupon
Imagine a customer considering a travel bag. They view the interior photograph, choose a color and leave the cart. The team assumes the price is too high. But support questions suggest uncertainty about whether a laptop fits.
A useful retargeting hypothesis is to show the bag’s verified interior dimensions and an honest demonstration. A discount hypothesis is different and has a different margin cost. Keep the ideas separate so you can tell what you are learning.
The example is fictional. You cannot know one visitor’s reason for leaving just from a cart event. Use aggregated support themes and buying-path evidence to make a defensible hypothesis without pretending to read minds.
Define eligibility before building audiences
Document the events, lookback period, exclusions and permission requirements that apply to the actual platform and jurisdiction. Check the provider’s current rules rather than assuming every site visit can be used everywhere.
Keep recent purchasers out of a message that says they have not bought. Review exclusions when returns, cancellations or delayed purchase events could create a mismatch. Treat past-customer recommendations as a separate purpose with appropriate permissions.
Email and SMS are separate channels with separate eligibility and suppression checks. Having an email address or sending an order confirmation does not by itself establish permission for every marketing message. Use current provider guidance and qualified advice where requirements are unclear.
Measure beyond the attributed orders
Fictional illustration: a campaign spends $400 and reports 20 attributed orders. Its reported spend per attributed order is $20. That calculation does not tell you how many of those customers would have bought without the campaign.
Suppose a properly designed, sufficiently informative holdout comparison later estimates eight additional orders under its stated assumptions. The cost per estimated additional order would be $50. These made-up figures show why attribution and incrementality deserve separate labels. They are not a result or an expected lift.
A holdout also needs careful design: assignment, contamination, sample size, timing and uncertainty matter. If the test cannot resolve a useful difference, report that limit rather than treating a noisy estimate as ground truth.
Use a campaign worksheet
Write the buyer question, supporting evidence, eligible audience, exclusions, message, destination, cost limit and review date. Add the contribution calculation that would make the campaign worthwhile. Keep a screenshot of the exact claim and landing page.
Review delivery and frequency in the context of audience size and campaign controls. There is no universal weekly cap that makes every audience comfortable. Watch complaints, irrelevant delivery and repeated exposure alongside orders and contribution.
Your next step is to find one unresolved buying question and decide whether the page should answer it first. Use the CRO guide for that repair. A better destination can be more useful than repeatedly advertising an unclear one.
Editorial revision, October 20, 2026: unsupported recovery percentages, conversion benchmarks and universal audience windows have been removed. All numerical examples are fictional.