Double proxy
A double proxy is one event that stands in for two different things at once: that the user got somewhere with what they came to do, and that the account is likely to convert or stick around. Most tracked events only cover the second half.
Definition
An event that stands in for both value streams at once — proxying for the user's progress toward their sought state and correlating with the provider's commercial outcome.
One event, two streams
A behavioral proxy is normally selected against the provider's side alone. The event correlates with conversion, and whether the user got what they came for goes unaddressed. "Logged in three times" works this way: it may predict conversion accurately while saying nothing about what the user was pursuing when they logged in.
A double proxy carries both halves. "Sent an invoice" correlates with conversion, and the send is the kind of action whose point lies outside the offering, so the team's theory of what the event implies is anchored to what the user came to accomplish. One measurement covers user value and business value — the two streams the double-value problem says a single user has to produce from one finite supply of energy. This is the structural reason external-impact events make the strongest behavioral proxies: the tier is defined by the event's significance living outside the offering's surface, which is the same property that gives it a user-side half.
Relation to pursuit, not data source
Double-proxy status depends on how the event relates to the user's pursuit, not on where the event fires or who fires it. An event triggered by a third party — the user's client paying an invoice — can be a double proxy, because payment received is progress on what the user was after. An event the user performs directly, like opening a feature panel, often isn't, because its significance is internal to the offering. The test is whether the event's meaning is anchored to the user's intent.
How each half is known
The two halves sit on different epistemic footing:
- Commercial half — checkable. The event either correlates with the commercial outcome or it doesn't, and a cohort comparison settles which.
- User half — theorized. Instrumentation witnesses events, and an event is not itself progress toward anything.
This is not a limit on what's trackable. Some outward effects are perfectly trackable — payment received, invoice opened, document exported. The limit is on what any observation can establish. That a paid invoice advanced what the user came for is inferred from the nature of the action, never observed directly, no matter how good the instrumentation gets. The property that makes these proxies strongest is the one analytics cannot confirm, so establishing it means leaving analytics for user value research and outcome interviews.
Example
A freelance invoicing tool is evaluating two candidates against first paid conversion in its signup cohort. Both correlate: signups who sent an invoice in their first 14 days convert at a higher rate, and so do signups who logged in at least three times in week one.
The login count stands in for conversion only. If the team drives it up with reminder emails and the conversion rate among performers collapses, the team has no theory to check the failure against — they only ever knew that logins co-occurred with converting.
The invoice send stands in for both. Sending an invoice is how a freelancer gets paid, which is what brought them to an invoicing tool, so the correlation comes with an account of why it exists. If interventions lift sends without holding conversion, the team has something specific to examine: whether the sends they manufactured were the same kind of action as the organic ones, or whether the intent diagnosis was wrong to begin with.