ASO Patents

Patent number	US12332765B2
Official title	Systems and methods for variant testing at scale
Assignee	Apple Inc.
Inventors	Jae Hyeon Bae, Kurt M. Fredericks, Nicholas Kistner, Andrew T. Maher, Nihar Ranjan Hati, Mahesh Molakalapalli, Srivas Chennu, Jamie J. Martin
Priority date	September 23, 2022
Filed	February 10, 2023
Granted	June 17, 2025
Status	Active. Anticipated expiry March 8, 2043
Family	US20260050535A1 (continuation, pending)
Source	patents.google.com/patent/US12332765B2

What This Patent Covers and Why It Matters for ASO

Feature it describes: the measurement and statistics layer underneath product page testing.

This site already covers US20230176843A1, which describes how users are assigned to product page treatments. That document says nothing about how a winner is decided. This one does, and it answers the single question every ASO practitioner asks about product page optimization.

The statistical methods named are Bayesian sequential testing and chi squared tests, applied to conversion metrics aggregated daily.

That matters more than it sounds. A Bayesian sequential test is designed to be looked at repeatedly while it runs. Classical fixed horizon testing punishes you for peeking, because every extra look inflates the false positive rate. A sequential design assumes continuous monitoring and accounts for it. So if this reflects the platform, checking your test daily is not the error that ASO advice usually claims it is.

The other details that change how you should run tests:

• The outcome is binary and measured on a user’s first visit. The system records, for each user’s first visit, whether they downloaded the app. A user who returns three times and converts on the third is not three observations.
• Results are available the day after launch. Aggregation runs daily, so a number exists almost immediately. An early number existing is not the same as an early number meaning anything, and the sequential design is what keeps those two apart.
• The pipeline is built for millions of concurrent experiments. It uses a single pass over instrumentation data rather than rescanning raw logs per experiment, with per user statistics tables and a facts database partitioned by date. This is industrial infrastructure, not a feature bolted onto App Store Connect.
• What is varied is icons, screenshots and descriptions, and the metric being optimized is visits that lead to downloads. Nothing downstream of the install appears in the claim.

Put this together with the 80 to 20 traffic split described in the sibling patent and the practical guidance becomes concrete: variant traffic is scarce, the outcome is one binary event per user, and the test is designed to be monitored continuously rather than read once at the end.

Patent Summary

An app store generates variant versions of an application’s product page, varying visual elements such as the icon, screenshots and description. Users are assigned to either the unchanged control page or to a variant. For each user’s first visit the system records the visit date, a user identifier, and whether a download followed, storing these as timestamped conversion records in a facts database partitioned by date.

Conversion rates are then computed for the control and for each variant over a collection period, statistical tests are applied to those rates, and the resulting performance is displayed to the developer.

The scale mechanism is the part that gives the patent its title. Rather than scanning raw instrumentation separately for every running experiment, the system aggregates metrics across all active experiments in a single daily pass, using per user statistics tables to avoid redundant processing. This is what makes millions of concurrent experiments tractable and what makes next day results possible.

A continuation, US20260050535A1, was filed in June 2025 and is still pending, so Apple is continuing to pursue protection on this line.

A detailed breakdown of the pipeline, the figures and the full claim language will follow in a later update to this article.

Related reading: US20230176843A1 on how treatments are assigned and what is measured.