Digital funnels often report growth while hiding a more important failure: the system may be acquiring traffic that does not represent real demand. In EdTech, this problem is especially expensive because a "lead" may be a bot, an accidental click, a confused user, a parent exploring on behalf of a learner, or a serious prospect who needs a different next step. This paper proposes funnel data science as a practical framework for distinguishing behavioral lead quality from raw lead volume. The framework emerged from product work around a large EdTech environment serving more than 100,000 learners across 70 countries, where acquisition quality, platform migration, messaging funnels, and learning-intent signals became more important than simple conversion counts. The paper argues that lead scoring should move from static demographic or channel-based rules toward behavioral evidence: dwell time, click latency, session depth, path entropy, event rhythm, recency, completion signals, and downstream learning behavior. We define a behavioral lead taxonomy, propose an event schema for EdTech funnels, introduce operational metrics such as bot leakage rate and cost per qualified learner, and describe a maturity path from rules and RFM segmentation to supervised scoring, Markov journey modeling, and intervention design. The contribution is not a private-data benchmark; it is an implementation-oriented framework for teams operating in constrained or high-friction markets, including environments where advertising platforms, messaging channels, and payment infrastructure may change under pressure.
Ilya Emelianov (Thu,) studied this question.