Outcome-event substitution
Outcome event substitution is running an analysis against a different event than the one that actually represents the conversion the business cares about, usually because that conversion event isn't tracked, isn't tracked reliably, or has too little sample to produce stable numbers. The substitute is typically a pre-conversion event like "checkout initiated" or "trial started" that can be measured cleanly, or a downstream event like "still a customer at month two" that strips out noise the raw conversion event carries (one-payment-then-churn, gamed trial flows). The cost is interpretive distance: whatever the analysis finds about the substitute has to be translated back into a claim about the actual conversion, and that translation isn't always clean.
Definition
Running multiplier analysis against a substitute event when the ideal commercial-outcome proxy isn't reliably measurable, or when a different downstream event better answers the analytical question being asked.
Two motivations for substitution
Multiplier analysis is computed against a commercial outcome event — the conversion the business actually cares about. Substitution becomes necessary in two structurally different cases. In the first, the commercial outcome event exists conceptually but can't be reliably measured, so analysis swaps in a measurable pre-conversion event and accepts interpretive distance as the cost. In the second, the commercial outcome event is measured cleanly, but a different downstream event carries information the team wants to bake into the analysis — typically fit-signal or retention — and substitution sharpens what the multipliers track.
When the commercial outcome can't be measured
Three failure modes show up in practice:
- Sample too small. A reliable proxy exists (e.g., backend "transaction complete" from the payment processor) but its absolute volume is too low for stable multiplier computation. Rough thresholds: under ~1000 commercial-outcome events warrants considering substitution; under ~500 usually requires it; under ~200, direct analysis on the commercial outcome isn't informative.
- Conversion-named event is unreliable. An event with a conversion-suggesting name fires under conditions that don't actually represent the commercial outcome. The canonical case: a frontend "click buy" or "purchase initiated" event that fires regardless of whether the transaction completed. Using it treats intent-to-pay as if it were payment.
- No commercial-outcome proxy tracked at all. Nothing in the analytics tool represents the conversion event. Common in early-stage offerings where instrumentation hasn't caught up to the business model.
Substitute candidates in these cases are pre-conversion events whose measurement is reliable and whose relationship to conversion is well-understood: "trial started," "paywall viewed," "checkout initiated," "credit card entered." Each is closer to conversion than top-of-funnel events and farther from it than the commercial outcome itself.
When a downstream event sharpens the question
The motivation here is different: the commercial outcome event is tracked correctly, but it contains noise that distorts the analysis. First paid conversion is the canonical case — operationally clean, but poor-fit customers often churn within the first month after paying, and the conversion event can be gamed slightly. Substituting "still a paying customer at month 2" filters out one-payment-then-churn noise before the multiplier is computed. Resulting multipliers reflect what produces fit, not just what produces first conversion. Operational instrumentation is either a continuous-payment check (no cancellation in window) or a recurring-revenue check (second monthly payment landed).
Picking the substitute
Choose the closest event to the commercial outcome that's reliably tracked and carries enough sample to support analysis. The closer the substitute to the commercial outcome, the smaller the interpretive gap. Behaviors that correlate strongly with a far-upstream event ("paywall viewed") may correlate weakly with the commercial outcome itself — a high multiplier on paywall-viewed might mean the behavior drives users to consider paying without closing them.
Before substituting at all, verify what each available event actually represents. A click-buy that fires pre-transaction is structurally different from a transaction-complete, even when teams treat them as interchangeable in casual conversation. When no reasonable substitute exists, the right move may be adding tracking for a reliable commercial-outcome proxy rather than working around its absence; analysis output can then include recommended tracking rather than just multipliers against weak substitutes.
Example
A subscription product has a small monthly volume of paid conversions — about 350 per month — but signup cohort volume is high. Running multiplier analysis against first paid conversion produces multipliers that swing meaningfully from month to month without underlying behavior change. The team substitutes "checkout initiated" as the result event. Volume is now sufficient for stable analysis, and multipliers are interpretable. The output ranks behaviors that drive users into checkout, with the explicit understanding that some of those behaviors may produce checkout-starters without producing payers. The team supplements with a slower-running analysis directly on first paid conversion to sanity-check the top-ranked behaviors.