Defining "active member," and locating the leak in the consumption funnel.
Three teams at HD Hyundai Medi Plus had three working definitions of an active user. I led the KPI committee that fixed the vocabulary, mapped a five-stage funnel from app download through repeat purchase, and built the Power BI dashboard the leadership team uses for weekly review. Two findings reframed how the team thought about its own growth.
One executive review, three different denominators.
HD Hyundai Medi Plus operates a consumer digital-health app with a layered product surface: content, feature use, scheduled care, and an in-app commerce channel. By late 2023 each team had its own working definition of an active user: one based on login, another on feature breadth, another on payment status. Numbers from different decks didn't reconcile, retention conversations stalled on terminology, and growth-budget discussions had no shared denominator.
I was asked to operate the KPI Metrics Establishment Committee: a three-person team across AI/DX and R&D. My role spanned the committee facilitation and the entire data-and-dashboard build; the two R&D analysts contributed prior-art review, hypothesis framing, and analytical sanity-checking. The mandate was specific: produce a single canonical KPI vocabulary, operationalise the core metrics, and put them in front of leadership weekly.
Member-level event logs, deduped and joined in PostgreSQL.
I assembled a member-level panel from internal sources:
- Behavioural event logs: login, screen views, feature interactions, content consumption, scheduled-care actions, commerce events.
- Member metadata: registration date, demographic context, acquisition channel.
- Subscription and purchase state: plan, payment events, lifecycle stage.
Cleaning was the unglamorous half: deduping sessions across device boundaries, joining heterogeneous event tables on member id, normalising timezone offsets, and folding raw event types into canonical action categories that mapped to the product team's mental model. The first round scoped to B2C; B2B was queued for a later round.
Pragmatic definition, five-stage funnel, five KPIs.
Defining "active": a deliberate minimum bar
Rather than over-engineering the definition with feature-weighted scores or behavioural clustering, the committee chose a deliberately simple operational rule: a member is active if they have logged in within the period of interest, regardless of feature use beyond login. This made the funnel boundary unambiguous, comparable across teams, and easy to recompute as the product evolved. Stronger definitions would be layered in later rounds.
The funnel: five stages, two phases
I split the customer journey into two structural phases (a signup phase and a consumption phase), each with its own gating actions:
Five canonical KPIs
On top of the funnel I defined a small set of metrics designed for weekly executive review:
- Active User: DAU, WAU, MAU, with the ratios DAU/WAU and DAU/MAU as engagement-stickiness signals.
- Weekly Retention Rate: % of new-member cohorts still active by week N. Reported on a 12-week horizon.
- Carrying Capacity: steady-state retainable user count under current acquisition and churn dynamics.
- Churn Period: median time-to-disengagement from a cohort.
- Conversion Funnel: stage-by-stage drop-off across the five-stage funnel.
Power BI dashboard: exec view + analyst drill-down
I designed two viewing modes against the same semantic model. The executive view shows one page, the five KPIs, daily / weekly / monthly toggles, and trend lines, with no drill-down. The analyst view exposes the cohort matrix, weekly retention curves with cohort sizes, segment composition, and stage attribution, full filtering. Both fed from the same Power BI data model, so reconciliation between the two was automatic.
Two findings that reframed the growth conversation.
Three things came out of the first round of analysis that the team did not expect:
- The biggest leak is in the consumption phase, not signup. The signup phase loses about 73% of users between download and login. The consumption phase loses about 78% between feature use and the first purchase. Marketing dollars were going into the signup top of funnel; the worst leak was downstream of where the team was looking.
- Ad spend hit a carrying-capacity ceiling. After the first and second ad campaigns, the daily-login carrying capacity converged to roughly the same level. Additional ad spend stopped translating into additional retainable users. The capacity ceiling was a real-world constraint, not an aspirational target.
- The first commerce action was the strongest retention signal in the data. Members who completed at least one in-app purchase retained at ~80% over a 2+ month window. Non-purchasing members retained at ~29%. Crossing the purchase threshold predicted long-tail engagement more reliably than any pre-purchase behaviour the team had been tracking.
Together, these said the binding constraint on growth was not acquisition spend. It was conversion to commerce, getting users from feature engagement to their first purchase. That single stage transition was the most valuable lever the team had, and the one nobody was actively working on.
From bespoke decks to a stable weekly review.
- Single canonical KPI vocabulary across the teams. Reconciliation problems disappeared.
- Weekly executive review now runs against the dashboard rather than against bespoke decks.
- Growth strategy reframed. The conversion-to-commerce stage moved from a peripheral product concern to a primary growth lever. Acquisition spend conversations became more honest about the carrying-capacity ceiling.
- Foundation for the next rounds: a documented Deep Dive list and a B2B equivalent of the same KPI structure were already drafted at the close of round one.
I owned 90% of the dashboard build (model, measures, layout, governance); the remaining contribution came from the two analysts on hypothesis review and the BI platform team on access-control configuration.