Approximate intent
Approximate intent is the stance that every detail of what users are trying to accomplish doesn't need to be pinned down precisely. Instead, it means making an educated guess from whatever signals are around (what someone searched for, what ad brought them in, what they answered on a sign-up question). In design, it means not building separate flows for every possible reason behind the same activity: booking an Airbnb equally serves someone traveling to visit family, or on business, or only to redeem expiring air miles.
Definition
The principle that intent operates as approximate inference rather than atomic resolution — at runtime, detection works from noisy signals; at design, the offering serves the common child transition where concurrent parent transitions converge.
The two registers
The discipline operates in two structurally distinct places:
-
At runtime: reading the user means reading noisy signals — search terms, referral context, campaign attribution, questions placed at the transition's entry. Intent detection works by approximate inference across these signals. The aim is trajectory-reliability at an actionable altitude: the inference needs to be reliable enough to route the user to a path that improves their route efficacy. Classifying the user's exact internal state is a different and harder bar the framework doesn't try to clear. Intent detection is the operational form of runtime approximation — multiple signals combining into a best-guess intent classification, with confidence varying by user and signal strength.
-
At design: when concurrent parent pursuits converge on the same offering-served child pursuit, the offering serves the common child and stays parent-naive. Airbnb serves "book a place to stay" without modeling visit-family / vacation / air-miles redemption as separate cases. Parent-naive design trades route-efficacy and downstream margin-on-efficacy for design simplicity, and the trade is correct when the served child already clears the four-filter test cleanly and the parents don't individually justify their own designed-for treatment. When they do, the prism move is the alternative — surfacing parent intents and designing paths for some of them specifically.
Approximate routing outperforms no routing
The positive claim doesn't live at the level of specific accuracy. Approximate intent-based routing produces better outcomes than no intent-based routing, because routing without intent information has no basis for matching affordances to where users are headed. The framework rejects intent-atomism — the expectation that intent should resolve to atomic units before action is warranted — in favor of pragmatic intent reading and design.
Intent-oriented routing also differs from demographic routing in what it routes on. Demographics describe who the user is; intent describes what the user is pursuing. The framework treats the latter as the variable that actually predicts whether a given affordance configuration will serve the user, and as the variable that gives the team a usable brief that demographic data does not.
Example
A user searches "place to stay in Lisbon, October, 2 people" and lands on Airbnb. The signals available — destination, dates, party size, time-of-year, the language of the search — don't tell Airbnb whether the user is going to a wedding, a business retreat, a family visit, or a vacation. They tell Airbnb enough to surface listings that fit the inferred trajectory, route the user through the appropriate booking flow, and adjust post-booking communications band-matched to the inferred scope.
At the design level, Airbnb doesn't try to detect or serve those parent intents separately. The common child — "find and book a place to stay" — is what the offering serves, and the design stays naive to the wedding-vs-business-retreat parent variation. The runtime inference is approximate; the routing it supports produces better outcomes than no intent-based routing would.