Conversion

Run it yourself

Confusion Detection

This tactic shows contextual help after three or more help-page visits in a declining session, catching frustrated subscribers while activation is still recoverable.

Engagement line with two anomalous spikes circled for review
Validated

22.6 percentage points separate the two help-seeking paths

first-week session data across hundreds of organizations and 14.3M end users

Validated against propensity-matched comparison groups.

How we grade evidence →

Threshold trigger · Edition 1 · June 2026


What is it?

Help-seeking points in two directions: for one group, it is learning, the subscriber digging into documentation is on the way to deep activation; for the other, it is frustration, the subscriber circling help pages because the product is not making sense for them is on a path toward churn. The outcomes of the two paths differ by 22.6 percentage points, which makes reading the direction worth automating.

This tactic exists for the latter, frustrated group. When a subscriber logs three or more help or support page visits inside a session whose engagement is declining, the tactic shows a help pop-up with guidance targeted at what they were looking for, turning the moment confusion becomes visible into the moment it gets answered.

When it fires

Three or more help or support page visits matter only inside a declining session. Help visits inside a rising session read as learning and are left alone. Interrupting a subscriber who is teaching themselves the product would break the momentum the signal indicates.

Firing happens in-session, while the subscriber is still present. Confusion answered an hour later by email is a support ticket, while confusion answered in the moment is an activation save.

What the evidence shows

Subscribers on the two help-seeking paths end up 22.6 percentage points apart in terms of retention, validated across first-week session data from hundreds of organizations. The signal is strong because the same behavior carries opposite meanings, and the session’s direction tells them apart.

The stakes are familiar from churn-reason data: usability issues account for 14 percent of churn, and they show up in first-week URL patterns long before they show up in a Cancel Flow. Confused subscribers are detectable in days one through seven, early enough that a well-placed answer changes the outcome.

How it runs

In production, this tactic reads session data continuously, scores each session’s direction, and arms when help-page visits accumulate inside a decline. The pop-up cue renders with content matched to the help pages the subscriber visited, not a generic prompt.

Guardrails keep the tactic light: one pop-up per session, suppression for subscribers whose session reads as learning, and a cooldown after a dismissal so a subscriber who waves off help is not chased with it.

Run this for your business

Want to run Confusion Detection for your business? Connect the Churnkey MCP to your favorite AI agent. It reads your own usage and billing data and recommends the growth and retention plays most likely to move your LTV—starting with whether this one fits.

npm install -g @churnkey/mcp
Read the MCP docs →
Prefer we run it for you, measured against a holdout? We're piloting managed growth tactics with a handful of subscription companies. Talk to us about a pilot →

Churnkey's retention products run on the same dataset behind this tactic.

Schedule a demo

Put the evidence to work.

The same dataset behind these tactics powers Churnkey's retention products. See what it finds in your subscription data.