Sorting hat
A sorting hat is a set of questions inside the product that asks the user about their situation and uses the answers to route them into an experience matched to what they're trying to do. A travel app might ask whether someone is traveling with kids and whether they've booked accommodation yet; a fitness app might ask what equipment is available and how many days a week they can train. Users who answer differently land in different versions of the product, rather than all seeing the same default.
Definition
The in-product question-branching instrument that routes users to differentiated paths based on classified intent, operating as one signal stream within intent detection.
How users encounter it
The user answers one or more questions about their situation. Each answer narrows the classification — the first question might separate broad categories of intent, subsequent questions resolve ambiguity within those categories. At the end of the sequence, the user is routed to a downstream path matched to their classified intent, or to the OSFA default if no matched path exists.
The questions ask about factual circumstances rather than identity or mental state — what the user's situation actually is, not what they want or who they are. The intent classification is inferred from circumstantial answers, not asked for directly.
Branches without paths route to OSFA
When a sorting-hat branch terminates without a corresponding path built for it, the user routed to that terminal experiences the OSFA default rather than a differentiated experience. No behavior breaks, no data is lost, no force-feeding produces misclassification.
This is what makes high-cardinality classification compatible with low-cardinality investment. A sorting hat with five branches but only two paths is operating as a two-path instrument with three branches that route to OSFA. The count of branches isn't the count of paths. When the gap is large, the sorting hat is doing more classification work than routing work — appropriate when classification data is itself valuable, such as identifying which branches are common enough to justify path investment.
One signal stream among several
Intent detection integrates multiple signal streams: acquisition-channel data, search-term referrals, campaign attribution, sorting-hat answers, and any other ambient signal available at runtime. The sorting hat is one of these streams; intent detection is the broader runtime operation that integrates them all. Conflating instrument with operation obscures which discipline is being run — sorting hat design is branching-tree construction and question wording; channel signals require attribution hygiene and channel-to-cluster mapping; search-term signals require keyword-clustering and intent-inference from query patterns.
The most controllable lever
In activation engagements, the sorting hat is typically the most direct lever because the team builds it, owns it, and can iterate on it independently of marketing or attribution-tooling decisions. Other signal streams sit further from the team's direct control — acquisition data depends on campaign structure, search-term data depends on what users type. The sorting hat is the part of triangulation a product team can change on its own.
Example
A project management SaaS asks new signups two questions during onboarding: "Is your product launched?" and "Are you currently tracking retention?" A user who answers post-launch and yes-to-retention routes to a path focused on churn analysis. A user who answers pre-launch routes to a path focused on setting up initial measurement. A user who answers post-launch but no-to-retention routes to OSFA — the team hasn't built that branch yet, and the data on how many users land there tells them whether to invest in building one.