Skip to content

Measurement plan

In the admin: Analytics — the screen this page documents (opens in the running app).

The rationale states a set of falsifiable bets. This page is how we test them: the hypotheses we intend to support or refute, the metrics that do the testing, and the end-to-end path that connects a visitor's clicks to a downstream bid outcome.

Everything here is designed to be measurable — most of it can be collected now, from data the app already captures. Where a signal depends on the Eloqua integration or real bid data, that's called out.

Hypotheses → metrics

Each row is a bet from the rationale, the prediction it makes, the metric that confirms or refutes it, and where that metric comes from. The right-hand column ties to the vision success metrics and the behavioral metrics dictionary.

#Bet (principle)PredictionMetricSource
H1Guide, don't filterGuided sessions converge on a recommendation where open-filter sessions abandonRecommendation-reached rate across the funnel stagesSession snapshot
H2Right-size, don't up-sellA cost-appropriate cheaper alternative is surfaced in the majority of over-spec-prone sessionsRight-size / alternative-surfaced rateRecommendation output + recommendation-quality loop (#352)
H3Right-size, don't up-sellRecommended tier trends down toward sufficiency vs. historical selection tierOver-specification delta (recommended vs. historical tier)Recommendation output vs. baseline
H4Educate through every stepTeaching the trade-off raises confidence and reduces churn at each questionTime-on-question, back-navigation / answer churn, "engaged" signalBehavioral event stream
H5Educate / guideQuestions that lose people are identifiable and fixablePer-factor drop-off (#350)Event stream + snapshot
H6Product-market fit of the catalogueWhat buyers actually need is visible as demandDemand intelligence — answer distributions (#353)Genuine-human answer rollup
H7Guided path effectivenessSome tracks/persona paths convert better than othersTrack / persona-path effectiveness (#354)Snapshot by track
H8Selection precedes commitmentSelection feeds real pipeline — sessions become known leads and, ultimately, bidsConversion rate → downstream bid outcomeEloqua association (#351) + bid data (later)

A bet is refuted as readily as confirmed — H3 trending the wrong way, or H1 showing guided sessions abandoning as often as filters, is a signal to change the design, not to hide the number.

The end-to-end measurement path

The chain from an anonymous click to a bid outcome is deliberately staged, so each link is measurable on its own and the whole is auditable.

 anonymous session            known lead                 downstream
 ────────────────────         ──────────────             ──────────────
 guided-selection flow  ──►   lead capture / sync   ──►  Eloqua contact  ──►  bid / sale
      │                            │                          │                   outcome
      ▼                            ▼                          ▼
 session snapshot +           eloqua_session_contacts   lead intelligence
 behavioral event stream      (durable association)     (#351)
 (funnel, per-factor,         = the "converted"         scored/segmented
  demand, engagement)          funnel stage             for sales follow-up
  1. Session behaviour — the anonymous session snapshot is the source of truth for funnel counts; the behavioral event stream adds inside-a-step signals (time-on-question, answer churn, /compare, media views, CTA click, engagement). Both are PII-free by construction and gated by consent.
  2. Lead intelligence — when a visitor is captured, a durable Eloqua association (eloqua_session_contacts) is written; this is the converted funnel stage, deliberately the stable lead fact that survives the transient sync queue and PII purge. Lead intelligence (#351) scores and segments those leads for follow-up. See Eloqua integration.
  3. Downstream sale lifecycle — the association is the join key that will let session behaviour be tied to the eventual bid / sale outcome (won, value-engineered out, lost to a competitor). That last link depends on bid data landing (via the Eloqua integration, #350/#351) — the plan is built so the join exists before the outcome data does.

Data integrity — why the numbers are trustworthy

The measurement plan is only as good as the data under it. Three properties make it defensible:

  • Genuine-human filter. Metrics are computed over a trustworthy base that excludes bots and non-human traffic — see the trustworthy base.
  • PII-free by construction. Event payloads carry only knowledge-base ids and bounded numeric measures; free text is dropped at the write layer, so no email, name, or note can reach the analytics stores. Conversion is tracked by a durable association, not by storing personal data.
  • Consent-gated. Behavioural capture respects the 2-tier consent model; a session without measurement consent is not tracked.

What's live vs. planned

  • Live now: the session snapshot funnel, the behavioral event stream, and the analytics read paths (funnel, per-factor drop-off, demand intelligence, track effectiveness, recommendation quality) — visible in the admin Analytics screen.
  • Planned: the downstream join to bid outcomes, which lands with the Eloqua lead-capture and lead-intelligence work (#350/#351). This page documents the plan so the collection is in place before the outcome data is.

For admins

The metrics behind H1–H7 are already visible in the admin Analytics screen. A contextual link from that screen back to this rationale is delivered by the admin↔docs cross-referencing work (#568).