Testing an App Install Banner on Mobile Product Pages

If mobile-web losses cluster after the product page in analytics and survey answers show visitors asking about the app, the app may be able to ease that journey, and this banner test can show whether it does. What you should see is installs rising while purchases lost on the web come back in the app. If the losses come from missing information or slow loading on the product page, address that first; if you cannot match web and app identities yet, fix attribution first. Keep the banner out of the payment flow and set it up so it does not reopen on every page.

Cross-channel revenue per visitor

Attribute web and app revenue to the originally assigned visitor over a fixed period; record identity-matching gaps and deduplicate orders.

Also track

• App install rateMeasure installs per assigned mobile-web visitor.
• Web purchase conversion rate (CR)Track whether purchases are lost on the web.
• Session continuation rateMake sure visitors do not drop out of the journey.
• Seven-day app retentionUse retention among attributed installs as a diagnostic and also report returning app users per assigned visitor.

• BannerAdd one dismissible app banner to eligible mobile-web pages.
• PlacementKeep placement and presentation fixed.
• AudienceDefine eligible users before the test and keep the definition fixed.
• OfferKeep existing app incentives unchanged; changing the discount is a separate test.
• AttributionCheck cross-channel identity matching before making a decision from combined revenue.

1. 1A higher app install rate does not justify ignoring a drop in web purchase conversion. Cross-channel revenue per visitor, with orders deduplicated, is the deciding measure; if session continuation falls, the banner may be breaking up the shopping journey.
2. 2This result comes from this placement and presentation on product pages under the current app incentives; test it on other pages before extending it there. Read app revenue and retention over one fixed follow-up period, and report retention both among attributed installs and as returning app users per assigned visitor.
3. 3When the result is murky, check whether identity-matching gaps cluster in one arm, since unmatched orders can hide a difference. Run the web purchase conversion numbers through the A/B test significance calculator to see whether that gap is more than chance. It does not test average revenue, so treat the noisier revenue gap with the same caution.