Intent-mix fit
Intent-mix fit is the running check on whether the team's picture of why users show up still matches who's actually showing up. The team built the product around a theory of the user base (these goals come up most often, in roughly these proportions), and intent-mix fit is the comparison between that theory and what the live usage data keeps showing: which kinds of users are landing, in what numbers, converting at what rates. When the gap widens, either the user base has shifted, the original theory was wrong, or both, and the team has cause to revisit which goals the product is committed to serving.
Definition
An ongoing audit signal about whether the offering's theorized distribution of user intent still matches the actual user base.
Where the signal comes from
The deployed intent detection apparatus — most directly the sorting hat — generates the signal, which gets read against the team's prior theory of which intents are present and in what proportions. Before deployment, the team commits to the intents the offering will serve and an approximate sense of how common each one is. After deployment, the apparatus produces continuous data about who's actually arriving: sorting-hat answer distributions, per-intent conversion rates, and OSFA-routing rate. Intent-mix fit is the comparison between the theory and the data.
The pre-deployment form is what the team gathers when intent-clarifying questions go in before any paths exist — the distribution of intents across the user base and the conversion performance of each intent cohort. Intent-mix fit extends those data streams into post-deployment monitoring.
The two axes
The deployed sorting hat produces signal on two axes whose movements indicate different territory shifts:
- OSFA-routing rate: the fraction of users whose classification routes them to the OSFA default rather than to a built path. Climbs when more users land in branches the sorting hat has but the offering hasn't built for.
- Sequence-retention rate: the fraction of users who complete the sorting-hat sequence rather than dropping out mid-flow. Tracks whether the sequence itself functions as a survivable signal-collection instrument.
Neither axis is interpretable at a point in time. A high OSFA-rate isn't itself a problem — it's expected when classification cardinality exceeds investment cardinality. What intent-mix fit reads is how the axes move over time.
Reading axis combinations
Combinations of movement signal different things:
- OSFA-rate stable, retention climbing: sequence degradation. More users finishing the sequence not because routing improved but because the questions have lost discriminating power.
- OSFA-rate climbing, retention stable: intent-category drift. Same number of users complete the sequence, but more route to OSFA — either theorized categories are drifting from actual users, or acquisition has shifted to bring in intents the sorting hat doesn't have branches for.
- Both climbing: a deeper shift. Offering-surface drift, acquisition-mix drift, or compounding sequence-edits that have eroded both legibility and category-fit. Diagnostic work disentangles which.
The interpretation discipline runs the corresponding diagnostic when a single axis moves; when both move, the team picks which to investigate first based on which is more likely to have produced the other.
Audit signal vs. instrument failure
A predictable failure mode: a climbing OSFA-rate or dropping retention reads more naturally as "the sorting hat isn't working anymore" than as audit signal about the underlying intent mix. Teams respond by reworking the sequence to push the metrics back, or by removing the instrument entirely. Both moves forfeit the signal that would have surfaced the drift earlier.
The corrective is to run the diagnostic before intervening. A climbing OSFA-rate is signal about the territory, not necessarily about the instrument; removing the instrument removes the audit.
Example
A productivity offering's sorting hat asks three questions to classify arrivals into one of four intents: solo project management, team collaboration, client-facing work, and personal task tracking. At deployment, the team theorized a roughly even split and built paths for the first three. OSFA-routing rate sits around 25% — the personal-task-tracking branch routes to OSFA, as expected.
Six months later, OSFA-rate has climbed to 45%. Sequence-retention is unchanged. Reading the table: same number of users completing the sequence, more routing to OSFA. The intent categories are drifting from the actual user base — either acquisition has shifted toward personal-task-tracking users, or a fifth intent the sorting hat doesn't have a branch for is showing up in volume. The team treats the climbing OSFA-rate as audit signal, runs the diagnostic on acquisition channels and on freeform-text data, and discovers a wave of solo freelancers whose intent doesn't match any of the four existing categories. The signal pointed to where the territory had moved; the question now is whether the offering's path portfolio should expand to serve them.