Too many elicitation questions cause dropout before first use

What it is

To keep the first screen from being cold, a product lines up questions at the door. As the list grows, people who have not yet seen any content leave. Elicitation dropout is the preference form acting as a gate to content, whose length itself causes loss — the loss happens before recommendation quality has a chance to be judged.

This is front-loaded attrition: people do not leave because recommendations are bad. They leave because they have not been allowed to see any.

Why it happens

A new user’s goal is “see what this product is,” not “complete a taste census.” Each elicitation item spends tolerance that has not been earned. When a question has no immediate return (select, then another question, still no content), cost is entirely now and benefit is entirely in a hypothetical later; after discounting, many exit.

Gated elicitation also creates an escalation of commitment: after eight items people try to estimate “how much longer,” guess wrong, and close. A skip control that is hidden or worded as punishment (“skip and we cannot personalise”) will push some people through, but the completers are the trapped, not the better seeds.

Studying it

Make elicitation length an experimental factor: 0, 3, 8, 15 items; skip allowed or not; preview of matching items after each choice or not. Dependent variables: fraction who reach the first screen, first-minute use after arrival, week-one retention, agreement of seeds with later behaviour. Independent variables must include reaching the first screen — that is where dropout happens. Do not inspect recommendation metrics only among completers.

Cut the funnel by question index. A steep drop after item three is about that item’s burden or discriminability, not about “elicitation.” Do not generalise quality gains on completers to the people who were questioned away.

Where it stops holding

In high-involvement, one-time configuration (tax filing, a medical record), the long form is the task, not a gate in front of recommendations. After payment, or under external mandate (a school account), dropout is suppressed and the cost turns into resentment rather than install loss. This entry is about asking so much that people leave before seeing content. It does not deny elicitation as a cold-start patch, and it does not set three items as a cap.

Applying it

• At the door, ask only as many seeds as fit on one screen. Move the rest into use (“want more like this?” after an item).
• Give each question an immediate preview: two matching items appear after a tick, so benefit is not postponed until the form ends.
• Skip is a first-class path. After skip, land on the popularity channel, not an empty state or a threat.
• Check: conversion from “opened the app” to “saw the first content item,” contrasted by elicitation length. People who drop before content appears are a cost of elicitation. Do not file them under recommendation quality.

Related

• Same group: L6.03.1 Recommendation quality is lowest when there is no history · L6.03.2 Explicit preference elicitation is the usual workaround
• Nearby: L6.05 Turning Personalization Off · L2.03 Examples and Template Guidance · L1.12 Latency and Streaming Experience
• Search terms: elicitation dropout · onboarding abandonment · preference questionnaire length