Portfolio of milestones
A portfolio of milestones is the set of behavioral milestones a team has built and proven as reliable levers: multiple of them rather than one, each anchored to a specific reason users came in, each verified at scale to actually move the business outcome the team is accountable for.
Definition
A set of value milestones — one per theorized intent the offering serves — used in place of a single milestone that would aggregate distinct intents into one signal.
Single milestones collapse intents
Users arrive with different intents (see multi-intent model). A team that picks one milestone for the whole signup cohort is implicitly betting that either every user has the same intent or the milestone correlates well enough across intents to function as an aggregate. Neither typically holds at scale. When intents diverge across the user base, the singular milestone's correlation reflects the dominant intent in whatever cohort produced the original analysis; once the intent mix shifts, the correlation moves with it, and the team can't ask which intent moved because the milestone never distinguished them.
The portfolio follows from multi-intent. Once more than one intent is in play, the measurement framework structurally requires intent-anchored milestones to keep correlations legible at scale.
One milestone per theorized intent
The anchoring discipline does the work. Each milestone in the portfolio has a theorized intent attached — a named user pursuit the team believes the milestone correlates with, which defines what the milestone is measuring progress toward. N milestones because there are N intents worth serving distinctly, with each milestone serving as the proxy for one of them.
The portfolio addresses the singular-metric aggregation problem by construction. There is no framework-level milestone that averages across intents; the measurement object is the portfolio itself, and its elements are intent-specific by definition.
Diagnosability under per-milestone drift
The portfolio doesn't prevent any individual milestone from drifting. A milestone's anchored intent can shift away from what's actually driving its firings, through correlation drift at the per-milestone level — whether intent-mix drift inside that intent's sub-cohort, offering-surface drift, or other condition changes.
What the portfolio offers against per-milestone drift is diagnosability. With a singular milestone, the team can see the correlation moving but not what's driving the movement — the referent is an aggregate, and there's no apparatus for asking which part decomposed. With a portfolio milestone, the team can ask which intent's referent shifted and what about that intent or its serving infrastructure changed, because each milestone has its own theorized referent rather than an averaged one. The portfolio's payoff is structural diagnosability, not structural permanence.
Example
A project management tool serves two distinct intents: solo task-tracking (individual contributors organizing their own work) and team coordination (managers running shared workflows). A singular milestone like "created 5 tasks" would aggregate both — it correlates with retention overall, but the two cohorts are doing different things when they hit it. A portfolio splits them: "created 5 tasks in a private workspace" anchors to the solo intent; "assigned a task to a teammate" anchors to the team-coordination intent. Each sub-cohort is identified at intent detection and measured against its own milestone.
When the team-coordination milestone's correlation later weakens, the team can ask whether the team-coordination intent's distribution shifted, whether the assignment affordance changed, or whether something specific to that sub-cohort moved. With a singular milestone, the same weakening would surface as a correlation drop with no anchored referent to interrogate.