Funnel Drop Alerts: An Issue When Step-to-Step Conversion Falls

Last verified on Nov 10, 2026.

What it catches

A funnel is a sequence of steps people take: start signup, verify email, finish setup. A funnel drop is when a step that usually leads to the next one stops doing so. The users are still arriving. They are just not getting through.

This is one of the costliest failures that never throws an exception. A form validates the wrong field. A button handler fails in one browser. A third-party script covers the submit action. A redirect lands on a blank page. Every service reports healthy while revenue quietly stops moving.

The check is about the share of users who get from one step to the next, not the absolute counts. Traffic can double and the check still fires if the share who complete falls.

Typical causes:

• A release changes a form and a required field is never filled.
• An auth or payment provider returns an error only on one step.
• A script update breaks a handler on a popular browser.
• A redirect, link or feature flag sends users to a dead end.
• An email with a verification link stops being delivered.

Example: the second step of onboarding

This is an example, not output from a real project. Your frontend logs a step event when users enter and finish the profile step of onboarding:

2026-11-08T10:30:01Z INFO web onboarding_step_started step=profile session=s_5521
2026-11-08T10:30:48Z INFO web onboarding_step_completed step=profile session=s_5521
2026-11-08T10:31:12Z INFO web onboarding_step_started step=profile session=s_5522
2026-11-08T10:36:40Z INFO web onboarding_step_started step=profile session=s_5530

Most started events now have no matching completed event. Sazabi opens an issue. An example of what it could say:

• Name: Onboarding profile step completion has fallen
• Description: Far fewer sessions that start the profile step are completing it than is usual. Starts are at their normal rate, so users are arriving and not finishing. Completions fell right after the 10
  release.
• Severity: high

The agent points at the step where users stop. You do not have to guess which of five funnel stages broke.

How Sazabi sets it up

The Sazabi agent sets up sensors for your project and maintains them over time. For this signal it reads your logs to find events that pair as a step and its completion, then writes the question. In plain words: "Tell me when the share of started profile steps that complete falls well below its usual level."

You set nothing up. The agent writes, backtests and maintains the sensors. Regular daily and weekly cycles do not count as a drop. Each flow can be judged per plan or per platform, so a drop on one of them does not hide behind healthy totals, and the issue names which one.

When it fires, Sazabi records a signal and an investigation starts. It opens an issue if the drop is real.

What you see and how you get told

You see an issue naming the two steps, how far conversion has moved and what the agent found around the time it began, such as an error on the step or a recent release. Ask follow-ups in the thread, or hand the fix to a coding agent with sazabi issues agent-prompt. Issue triggered goes to the in-app feed and, with a delivery rule, to a shared Slack or Microsoft Teams channel for the product team. Use sazabi signals list --type sensor for the raw firings.

What you need

Events from your product that mark the steps of a flow. These come from frontend or mobile telemetry, or from backend services. For browser and app events, see client errors and frontend breakage. Connect logs through OpenTelemetry or AWS.

What it does not cover

• Behavior with no event: a click or a scroll your app does not log.
• Cohort analysis, experiment readouts and attribution.
• Detecting that an event stopped. That is vanished events.
• Broken sequences and order changes are not covered today.

Issues are metered; see pricing.