Logo
By adopting Statsig for Experimentation and Feature Management, Scribd, Inc. accelerated testing velocity and increased subscriber growth through real-time personalization with Multi-armed Bandits.

How Scribd, Inc. increased subscriber growth with Multi-armed Bandits

How Scribd, Inc. increased subscriber growth with Multi-armed Bandits

3%

increase in new paying subscribers

30%

growth in experimentation velocity

Scribd, Inc. is on a mission to democratize access to knowledge, providing millions of high-quality documents, presentations, ebooks, audiobooks, and community-driven literary discussions. With a diverse product suite—including Scribd, SlideShare, Everand, and Fable— and a growing user base, the company is committed to delivering personalized user experiences at scale. To sustain this growth and optimize user journeys from acquisition to retention, Scribd, Inc. has made experimentation and data-driven decision-making central to their strategy.

Life before Statsig: outgrowing internal tools

Scribd, Inc. has always valued rapid experimentation and iteration, initially supporting this culture with an internal experimentation platform. However, as the company expanded, their in-house solution struggled to keep pace:

  • It couldn’t scale with the increasing volume of experiments from different teams
  • Personalization was limited due to platform constraints
  • Internal resources for maintaining and upgrading the system were scarce
  • Crucially, the system lacked the capability to run multiple simultaneous experiments without overlap—known as swimlanes—which undermined the validity of their experimental results

Realizing these limitations, Scribd, Inc. needed a robust, scalable third-party solution that could enable advanced experimentation without sacrificing internal resources.

Building a scalable experimentation culture with Statsig

Since adopting Statsig in early 2024, Scribd, Inc. has significantly enhanced its data-driven product development process. Today, Statsig is predominantly used by Scribd, Inc.'s growth team for experimentation and all product teams for progressive feature rollouts:

  • Growth Team: Primarily leverages experimentation, including advanced techniques like multi-armed bandit algorithms, to rapidly test and personalize user experiences
  • All product teams: Utilizes Statsig’s feature management capabilities to progressively roll out new features, ensuring stable user experiences and controlled risk management

Statsig has streamlined experimentation at Scribd, Inc., providing intuitive, powerful tools that significantly reduced the overhead previously required from data scientists and engineers.

“As a data scientist, I don't have to build every single personalization model from scratch. Statsig’s out-of-the-box solutions free up a lot of resource-intensive work we would otherwise have to tackle ourselves.”
Regina Chan

Regina Chan

Senior Data Scientist, Scribd, Inc.
Scribd, Inc. Logo

Increasing new subscriber growth by 3% with Multi-armed Bandits

One of Scribd, Inc.’s standout successes with Statsig has been their adoption of Statsig’s multi-armed bandit capabilities. Multi-armed bandits use machine learning to dynamically allocate traffic among different experiment variants, continuously optimizing for the best-performing option. Statsig’s AutoTune AI specifically leverages contextual bandit algorithms to personalize experiences by automatically selecting optimal content or features based on real-time user context and performance data.

The Growth team leveraged a top-requested feature—identified through user research and launched by the core Scribd team—as the foundation for their first contextual multi-armed bandit experiment. They used it to test different paywall messaging variants shown when users attempted to copy text from documents. They hypothesized that different messaging could significantly influence users’ decision to sign up, depending on the context.

Using AutoTune AI, Scribd, Inc. tested four messaging variants based on two simple user metadata attributes: device type (desktop or mobile) and the time of day the interaction occurred. The AutoTune AI tool streamlined the process by dynamically adjusting which messaging variant users saw, continuously learning and optimizing to show the most effective message for each context, making it easy and intuitive for Scribd, Inc.’s growth team.

“Statsig has made it possible for us to run tests that weren’t feasible with our internal solution before—like contextual bandits that personalize experiences in real time.”
Tess Peppers

Tess Peppers

Product Manager, Scribd, Inc.
Scribd, Inc. Logo

For an initial test with Multi-armed Bandits, the results were impressive. The test showed nearly a 3% increase in new paying subscribers by leveraging just device type and time of day.

“With Statsig’s Autotune feature, we were able to prove that incorporating even lightweight personalization with just a little bit of user metadata is already driving meaningful improvement in our core business metrics.”
Regina Chan

Regina Chan

Senior Data Scientist, Scribd, Inc.
Scribd, Inc. Logo

Scribd, Inc. valued Statsig’s partnership during this initial experiment, especially the direct support provided by the Statsig team throughout the process.

“Given that multi-armed bandits were a new feature for us, working directly with the Statsig team was very valuable. They clarified settings and updated documentation, ensuring everything functioned exactly as expected.”
Regina Chan

Regina Chan

Senior Data Scientist, Scribd, Inc.
Scribd, Inc.

Building on this success, Scribd, Inc. plans to expand their use of multi-armed bandits by incorporating richer user metadata—such as country and content consumption patterns—from their internal feature store. These attributes will power more personalized experiments aimed at improving activation, retention, and reducing churn, all while efficiently running in the background to support smarter, resource-light optimization.

Accelerating experimentation velocity by 30%

Beyond specific experiments, Statsig has significantly boosted Scribd, Inc.'s overall experimentation velocity, achieving a 30% increase in the number of experiments run by their growth team. This increase in efficiency is attributed to Statsig’s intuitive UI, robust infrastructure, and self-service capabilities.

Looking ahead, Scribd, Inc. aims to extend their data-driven culture across more teams and user journeys. The growth team particularly envisions deploying advanced personalization models continuously in the background, further refining user retention and reducing churn. With integrated experimentation and feature management, Statsig provides Scribd, Inc. with the infrastructure to scale controlled rollouts and personalized user journeys to support their next stage of growth.

About Scribd, Inc.

Scribd, Inc. provides access to the written and spoken word across a portfolio of unique products: Scribd®, SlideShare®, Everand™ and now Fable. Scribd is home to the world’s documents, with specialized information shared by a global community of thinkers. SlideShare provides millions of presentations and other professional content. Everand is a destination for stories and knowledge, offering millions of ebooks, audiobooks, and more. Fable is the leading social reading platform, where readers discover, discuss, and share books through curated book clubs and personalized reading experiences. These apps are available on iOS and Android devices, as well as web browsers.

We use cookies to ensure you get the best experience on our website.
Privacy Policy