Skip to main content
beta

​
What are Contextual Bandits?

A Contextual Bandit is a bandit that personalizes which variation a user sees based on their context — attributes like country, device, or past behavior. Like a standard multi-armed bandit, traffic weights change while the experiment runs to favor better-performing variations. The difference is that a contextual bandit learns a separate set of weights for each context, instead of a single global set of weights for all users. In other words, a multi-armed bandit asks “which variation is best?” A contextual bandit asks “which variation is best for this kind of user?” Where a global winner would be applied to everyone, a contextual bandit can send users in one country to one variation and users in another country to a different one — whatever performs best on the Decision Metric within each context.

​
When should I run a Contextual Bandit?

You should run a contextual bandit when:
  • You expect the best variation to differ across user segments (e.g. by region, device, or plan), rather than there being one global winner.
  • You have a single, clear Decision Metric to optimize toward.
  • You can express the segmenting attributes as columns in your assignment query so the bandit can split on them.
  • You care more about shipping the best experience to each segment than about producing unbiased, per-metric learnings.

​
How is it different from a multi-armed bandit?

​
What you need to run one

A contextual bandit splits traffic on context attributes, so it needs an assignment query that selects those columns in addition to the user identifier and timestamp. In GrowthBook these are defined on a Contextual Bandit Query, which holds the assignment SQL plus the list of targetingAttributeColumns the bandit is allowed to split on. Each context column you want to use must appear in the query’s SELECT, and at least one targeting attribute is required — a bandit with no context to split on is just a multi-armed bandit. To get started you will need:
  • A connected Data Source with a warehouse the bandit can query.
  • An assignment query that selects your user id, timestamp, and one or more context columns.
  • A single Decision Metric to optimize toward.
  • Two or more variations to test.
  • An SDK that supports contextual bandits — version 1.7.0+ of the JavaScript, React, or Node.js SDK.
For a step-by-step checklist covering SDK installation, tracking callback changes, and the rest of the setup, see Setting up a Contextual Bandit.

​
GrowthBook’s Contextual Bandit implementation

Like multi-armed bandits, contextual bandits use Thompson sampling, a Bayesian algorithm that balances exploration (trying variations to learn how they perform) and exploitation (sending more traffic to the variations that look best). The key addition is that GrowthBook fits a decision tree over your context attributes, partitioning users into leaves — groups that share similar context — and then runs Thompson sampling within each leaf. This is how the bandit can converge to different winners for different kinds of users. As with standard bandits, GrowthBook ensures every variation keeps at least a small share of traffic within each context, so the bandit can keep adapting if user behavior changes over time. The statistical details are covered in the Contextual Bandit technical reference.

​
FAQ

  1. When is a contextual bandit worth it over a multi-armed bandit?
    Only when you have a real reason to believe the best variation differs across user segments. If one variation is best for everyone, a contextual bandit adds complexity (splitting your data across contexts, which needs more traffic per context) without a payoff. A plain multi-armed bandit is the better default.
  2. Can I run a Contextual Bandit using the frequentist engine?
    No. Like multi-armed bandits, contextual bandits are available only under the Bayesian engine, where Thompson sampling is used.
  3. What happens if a context has very little traffic?
    GrowthBook controls how finely it splits using settings like minUsersPerLeaf and maxLeaves.
  4. Do contextual bandits suffer from the same biases as multi-armed bandits?
    Yes — because traffic weights change adaptively, the same adaptive-experimentation biases apply, and splitting by context can make them more pronounced in low-traffic segments. If unbiased per-metric effect estimates are your goal, a standard experiment is still the better tool.