Behavioral proxy
A behavioral proxy is an observable action a user takes in the product, used as a stand-in for a business result the team can't see yet, like whether someone eventually becomes a paying customer. The team watches the action because it shows up right away, while the result it stands in for only lands much later.
Definition
A behavioral proxy is an observable user action used to stand in for a commercial outcome, measured either as a binary — did the user perform action X — or as a count — did the user perform action X at least N times.
Binary or count
The action being measured can be atomic or composite. Atomic proxies track a single event: sent an invoice (N=1), followed 30 accounts (N=30), returned for 3 sessions (N=3). Composite proxies layer conditions onto that base — filtered by population (mobile users who complete checkout), combined across actions (signed up and created a project), or sequenced (created a project, then invited a collaborator). Facebook's "7 friends in 10 days" is composite: a count paired with a time-window filter.
This binary-or-count shape is shared by every behavioral proxy, including the three tiers below and the value milestone. What separates one proxy from another is what the underlying behavior is anchored to and what relationship that behavior has to the commercial outcome.
The three tiers
Behavioral proxies sort into three tiers, ordered here from least to most reliable. Reliability tracks intent-based explanatory power — how much the proxy accounts, at a model level, for why it correlates with the commercial outcome.
- Engagement event: keyed to undifferentiated activity — that the user is present at all (retained at week zero, three sessions logged). Near-zero explanatory power, since presence says nothing about why a user would convert. It works as a health metric, not a value metric.
- Feature-usage event: keyed to which in-app feature the user engaged (created three projects, followed 30 accounts). Severely diluted explanatory power — it gestures at intent but is mostly noise, and typically fails at scale.
- External-impact event: keyed to an in-app event that produces an effect outside the offering, like an invoice that reaches a real client. Highest explanatory power, conditional on diagnosing the user's intent correctly, because the behavior is causally tied to the state the user is actually after.
From proxy to lever
Naming a behavioral proxy says nothing about whether acting on it works. A candidate behavior is a proxy under active hypothesis-testing; a value milestone is one that has earned lever status by surviving scale-testing, where interventions that amplify the behavior produce predictable lift in the commercial outcome. All three are the same structural object — an action measured as a binary or a count — at different stages of verification.
Example
An invoicing tool watches "sent an invoice" as a behavioral proxy. Measured as a binary with N=1, it asks one thing: did this user send at least one invoice? The action stands in for a commercial outcome the team can't observe quickly — months of retention data compressed into a single early signal. It sits at the external-impact tier because the invoice has effects beyond the app, a client receives it and may pay it, which is why it carries more explanatory power than "logged in three times" would.