# Supertargeting: combining country, device, source and behaviour > Stacking conditions makes the segment sharper and the sample smaller. Here's how to know when the trade is worth it — and when you've segmented yourself into noise. - Source: https://bixlow.com/en/blog/supertargeting-combining-conditions - Language: en - Published: 2026-10-09T09:00:00.000Z - Updated: 2026-10-06T09:28:18.400Z - Author: fernando - Topics: audience targeting, geographic targeting, experiment design, sample size, A/B testing, website personalization, CRO - Publisher: Bixlow --- "New users, on mobile, in Spain, arriving from Meta Ads, viewing a product under €40." That's five conditions. It describes a real and coherent person, and it might be 60 sessions a week. Supertargeting is the ability to express that. Knowing when to use it is a separate skill, and it's where small brands most often go wrong in both directions — some never segment at all, others segment until nothing is measurable. ## The trade, stated plainly Every condition you add does two things simultaneously. It increases the *homogeneity* of the group, which reduces variance and makes an effect easier to see. And it reduces the *size* of the group, which makes any effect harder to see. These pull in opposite directions, and the second usually wins. Halving your sample costs you more statistical power than the homogeneity gains back, unless the condition genuinely separates two populations that behave very differently. So the rule is: add a condition only when you expect the segment to respond *differently*, not merely when you can describe it. ## A test for each condition Before adding a condition, answer: would I expect the variant's effect to be meaningfully different inside this group versus outside it? - "Mobile" for a sticky-button test: yes, obviously. The mechanism doesn't exist on desktop. - "Spain" for a shipping-message test: yes, if shipping norms differ across your markets. - "From Meta" for a message-match test: yes, that's the entire point. - "Chrome users" for almost anything: no. Browser rarely predicts buying behaviour. This is a condition people add because it's available. - "Product under €40" for a guarantee test: maybe. Price sensitivity plausibly changes how much a guarantee matters. If the answer is "I don't know", don't add it as a target. Record it as a segment and read the breakdown afterwards — that costs you nothing and might generate your next hypothesis. ## Target narrow, measure broad This is the technique that resolves most of the tension. Use the minimum conditions needed for the mechanism to apply, and let everything else be recorded rather than targeted. A sticky add-to-cart test targets mobile, because the mechanism is mobile-specific. It does not target country, source or visitor type — but Bixlow records all of them, so when the test finishes you can see whether the effect was concentrated in new visitors or in paid traffic. That post-hoc breakdown is exploratory, not conclusive: you didn't power the test to detect a difference within a sub-segment, so a striking sub-group result is a hypothesis for the *next* experiment, not a decision. Treating it as a decision is how teams end up shipping changes based on the one slice of twelve that happened to look good. ## When narrow targeting is actually right There are cases where heavy targeting is correct despite the sample cost. When the change would be *harmful* outside the segment. Showing an aggressive discount banner to visitors who arrived ready to pay full price costs you margin. Here you target tightly for business reasons and accept the slower test. When the segment is where your money is. If 70% of your revenue comes from one country on desktop, a test targeted there isn't a narrow slice — it's your business. When you're managing risk on a big change. Launching a checkout restructure to 10% of one market first is a deliberate blast-radius decision, not a measurement decision. ## The audit nobody does Supertargeting accumulates. After a year you have nine live personalisations, each targeted at a stack of conditions, and nobody can say what a given visitor actually sees. Keep a list of live targeted experiences with the condition stack and the metric that justified each one, and review it quarterly. Anything whose justifying metric you can no longer find, turn off. A site whose behaviour nobody can predict is worse than a site with no personalisation at all, because you can't reason about the results of anything you test on top of it. ## Key takeaways - Add a condition only when you expect the segment to respond differently, not because the condition exists. - Target the minimum needed for the mechanism; record everything else and read the breakdown afterwards. - Post-hoc sub-group results are hypotheses for the next test, never decisions on their own. - Tight targeting is right when the change would harm the excluded group, or when the segment is where the revenue is. - Audit live targeted experiences quarterly and turn off anything you can't justify. --- ## Meet our AI CRO agent What you just read, put to work by an agent: it understands your visitors, decides what to test or personalize, builds it on your site without a developer and tells you when a result is real. https://bixlow.com/en/ai-cro-agent