Increase App Installs From Web Menu Landers: Funnel, Experiment, and Measurement Plan

Context

A food delivery platform wants to increase app installs from users who arrive on restaurant menu pages via web links (they are not yet app users). The goal is to drive installs without harming overall conversion to order (combining web and app).

Assume users may be anonymous or logged in on web, may switch devices (e.g., click on desktop, install on phone), and that app-store privacy policies can limit deterministic attribution of installs.

Tasks

1. Map the funnel and define metrics for the journey:
   • web_lander → deep_link_prompt_view → store_visit → install → first_app_open → order_within_7d
   • Include guardrails: bounce_rate, add_to_cart_on_web, conversion_to_order (web + app)
2. Design an experiment comparing two variants:
   • Variant A: Soft interstitial with smart deferred deep link
   • Variant B: Aggressive full‑screen gate shown after add-to-cart
   • Specify: randomization unit (session vs. user), cross-device identity pitfalls and mitigations, power/MDE assumptions, holdout design, and exposure caps.
3. Define success criteria and a trade-off policy if installs rise but immediate web orders fall.
4. Outline how to segment effect sizes by intent (e.g., cart_abandoners, repeat restaurants, delivery_distance) and how to avoid Simpson’s paradox.
5. If app-store policies or tracking limits block attribution, propose measurement alternatives (e.g., geo-level switchback, pre-post with CUPED, modeled install lift via synthetic controls) and how to validate incrementality.