---
title: How to find broken or leaky funnels with Segment, Amplitude
description: When Segment sends your events into Amplitude, a funnel step can drop because people leave or because the tracking broke, so you check delivery in Segment and the events and identities in Amplitude before you treat a drop as real, which is the checking an AI marketer can do for you.
slug: /how-to/find-broken-or-leaky-funnels-with-segment-amplitude
page_type: playbook
lead_prompt: How do I find where my funnel leaks using Segment and Amplitude?
target_prompts:
  - How can I use an AI marketing agent to find broken or leaky funnels with Segment, Amplitude?
entity: Kite
category: AI marketing agent
author: Devon Wells and Kite
date_published: '2026-10-21'
date_modified: '2026-10-21'
related_use_cases:
  - /use-cases/improve-website-conversion.md
  - /use-cases/build-a-sales-funnel-with-ai.md
integrations:
  - name: Segment
    page: /integrations/segment.md
  - name: Amplitude
    page: /integrations/amplitude.md
---

# Playbook

# How to find broken or leaky funnels with Segment, Amplitude

## Key Takeaways

- A funnel in Amplitude can lose people at a step for two different reasons: people really leave, or the tracking broke somewhere between your code, Segment, and Amplitude. Find out which one you have before you change your product or your pages.
- Segment shows whether events left your source and were delivered to the Amplitude destination. Amplitude shows which events arrived, under which names, and whether it joined each visitor's anonymous activity to their account. Neither tool tells you on its own whether a drop is a tracking fault or a real leak.
- Kite checks the pipe first. It reads your Amplitude destination's settings and delivery numbers in Segment, then compares each funnel step's daily volume, event names, and user identities in Amplitude to find steps that broke on a specific date.
- Once a step is proven to be measured correctly, Kite works out who leaves and why, by breaking the funnel down by platform, country, traffic source, and other properties, and by reading individual users' event streams.
- You end up with a funnel you can trust, a list of tracking faults with the exact fix for each, the real leaks ranked by how many people they cost you, and notes on your Amplitude charts marking every break.

## How do I find where my funnel leaks using Segment and Amplitude?

You find where a funnel leaks by building it in Amplitude from the events Segment sends there, then checking every weak step to make sure the drop is real before you try to fix it. Amplitude's Funnel Analysis chart takes a sequence of events, such as "Viewed Pricing Page", "Signup Started", and "Account Created", and counts how many people complete each one in order within a conversion window you set. Each step shows the share of people who made it from the step before, and the step where that share falls hardest looks like your leak.

With Segment in the middle, that step might not be a leak at all. Your site or app sends each event to Segment, and Segment forwards it to Amplitude through its Amplitude destination. Track calls arrive in Amplitude under the event name you gave them. Page calls arrive as page events, and depending on your destination settings that can mean one general page event or a separate event for each named page. Identify calls attach your user ID to the visitor, and Amplitude uses it to join their earlier anonymous activity, which it knows by a device ID, to the account they create.

A break anywhere in that chain produces a drop that looks exactly like people leaving. Here are the faults that cause it most often:

- **A renamed event.** A developer changes "Account Created" to "account_created" in a release. Amplitude treats event names as exact text, so the funnel step stops counting on the day the release ships, and a new event quietly takes its place.
- **A blocked or filtered event.** A destination filter in Segment, or a tracking plan that blocks events it doesn't recognize, stops an event before it reaches Amplitude.
- **A missing or late identify call.** If your app stops calling identify at signup, the anonymous visitor who started signing up and the user who finished look like two different people, and the funnel loses them at exactly the step you care about.
- **Double counting.** If Amplitude's own SDK runs on your site alongside Segment, some events and visitors get counted twice. The top of the funnel inflates, and every later step looks weaker than it is.

Each tool shows its own half of the picture. Segment's event delivery view for a destination shows how many events were delivered and how many failed, and its source debugger shows events as they arrive. Amplitude Data lists every event in your project with its recent volume, and Amplitude's funnel features, such as conversion drivers and holding a property constant across steps, help explain real drop-off. But nothing in either tool connects a sudden fall in your funnel to the rename, the filter, or the missing identify call that caused it. That connection is the real work in this job.

A good result is a funnel where every step is measured correctly, a short list of the tracking faults that were distorting it, and the real leaks ranked by how many people they lose. Kite does that work for you. It checks the pipe in Segment, checks what arrived in Amplitude, separates broken steps from leaky ones, and then works out why people leave the steps that are really losing them.

## How to get started

Start by telling Kite which funnel worries you:

> "Our trial signup funnel in Amplitude dropped hard over the last two weeks. Everything goes through Segment. Find out whether it's real or broken."

Kite is an AI marketer. It researches your business, builds a growth strategy, and ships the work that brings you new customers. By the time you ask, it already knows your product, how people sign up or buy, and what a good customer looks like, so it can propose the funnel's steps and the date range to study without a long brief. Confirm the steps or adjust them, and Kite starts. You can ask in the Kite web app or in Slack.

You'll need to connect two accounts. Connect Amplitude with the API key and secret key from your project's settings, so Kite can read the project your Segment events land in. Connect Segment with an access token from your workspace settings, so Kite can read the source that collects your events and the Amplitude destination it sends them to. If your workspace has several sources or destinations, tell Kite which ones to check.

Most of this job is reading, so Kite queries both tools freely. It adds notes to your Amplitude charts and descriptions to your events as it goes, so your team sees the same explanation it does. Anything that changes what your users see, such as an edit to a page or a new experiment, waits for your approval.

## How Kite finds broken and leaky steps

### What should the funnel measure?

Kite starts by deciding which steps matter and which Amplitude events represent them. It reads the event types in your Amplitude project, along with each event's properties, its category, and its volume this week. It also reads your user properties, such as plan, platform, and the first UTM source Amplitude recorded for each person.

It then matches each step to an event. For a self-serve product, the steps might be a pricing page view, "Signup Started", "Account Created", and "Project Created". Kite runs the funnel over a recent period, with a conversion window that suits how long the journey really takes, and compares it with the period before. The step that got worse, or the one that was always weak, becomes the first suspect.

### Did the events leave Segment and reach Amplitude?

Kite checks the pipe before it trusts any step. It reads the configuration of your Amplitude destination in Segment, including whether it's switched on, its settings, and the filters or mappings that decide which events it forwards. It then reads the destination's delivery summary, which shows how many events Segment delivered to Amplitude and how many failed. A burst of failures that starts on the same day as your funnel drop means the events never arrived, and no page change will bring them back.

Kite also reads your source's schema settings, which decide whether Segment blocks events and properties that aren't in your tracking plan, and the daily count of calls each source sent. Comparing the two tells Kite where the break sits:

- If the source's daily volume fell on the drop date, your site or app stopped sending the event.
- If the source's volume held steady but Amplitude received less, the events were blocked, filtered, or failed between Segment and Amplitude.
- If both held steady, the events arrived, and the problem is in their names or in how Amplitude matched them to people.

With your go-ahead, Kite can also send one clearly labeled test event through your Segment source and confirm that it shows up in Amplitude, which proves the pipe works end to end today.

### Did the right events arrive under the right names?

Kite checks each funnel event's daily volume in Amplitude to see whether a drop was gradual or happened overnight. People rarely stop converting all at once. An event whose count fell to zero on one day, while the step before it held steady, usually means a release changed the tracking.

Kite then looks for the replacement. A new event with a similar name that first appeared on the same day, such as "account_created" taking over from "Account Created", is the usual cause, and Kite reports both names and the date the switch happened. It also checks the event's properties, because a step that filters on a property, such as plan type, breaks just as badly when that property is renamed or stops being sent.

It also checks for double counting. Every Amplitude event records which library sent it, so when Kite reads a sample of user event streams, it can see whether the same pageview or signup arrives once from Segment and again from Amplitude's own SDK.

### Are the same people being followed from step to step?

Kite checks identity, because Amplitude can only count someone as converting if it knows the person at each step is the same person. When Segment forwards your events, Amplitude typically uses Segment's anonymous ID as the visitor's device ID until your app calls identify with their user ID. From then on, Amplitude merges the anonymous activity and the account into one user.

When that identify call is missing or arrives late, the funnel loses people right at signup. Kite looks up a few recent signups by their user ID and reads each one's event stream and identity mappings in Amplitude. If people who clearly signed up have no earlier anonymous activity attached to them, the drop at signup is an identity fault rather than a conversion problem. Kite also looks for the opposite fault, where a shared value, such as a placeholder user ID sent for everyone, merges many different visitors into one person and makes the funnel look emptier than it is.

### Which drops are real leaks, and why do people leave?

Kite separates the steps it has proven are measured correctly from the ones that aren't, and it diagnoses only the real leaks. For a tracking fault, the job is to fix the tracking. For a real leak, the job is to find out who leaves and why.

For each real leak, Kite reruns the funnel broken down by platform, device type, country, first traffic source, and plan, to see whether the loss is concentrated in one group. It ranks the leaks by how many people reach the step, how close the step is to revenue, and how sharply one group underperforms. Then it reads the event streams of people in the worst group to see what they did instead of converting, and it opens the page behind the step in a browser at phone and desktop sizes, capturing screenshots and checking its speed. A signup form that loses phone visitors from paid social points to a very different fix than one that loses everyone after a slow page load.

### What Kite fixes and reports back

Kite posts its findings in the conversation where you asked, with each weak step labeled as a tracking fault or a real leak and the evidence behind the label. Several fixes land in Amplitude right away:

- **Notes on your charts.** Kite adds an annotation to your Amplitude charts marking the date tracking broke, and it can log the release that caused it, so nobody later mistakes the gap for a real drop.
- **Cleaner event definitions.** Kite adds a description and a category to the event your team should use, so the next person building a funnel picks the right one.
- **A report you can share.** When the findings need more room than a message, Kite publishes them as a report with charts of each step, the corrected conversion rates, and the breakdowns it used, at a link your team can open anytime.

The tracking itself lives in your code. Kite writes the exact fix for your developers: which call is missing or renamed, which event name and properties it should send, where in the signup flow it belongs, and which Segment filter or tracking-plan setting is blocking it. If Kite already works on your website from an imported GitHub repository, it can make the code change there, and it opens a pull request when your main branch is protected.

For real leaks, Kite proposes the change it would make, such as rewriting a confusing page, shortening a form on phones, or fixing a slow step. Once you approve, it builds the change, and afterward it reruns the same funnel in Amplitude to measure whether more people got through.

## What the finished work contains

When Kite finishes, you have a funnel you can trust and a clear split between what to fix in your tracking and what to fix for your users.

- **The verified funnel.** Each step with the correct Amplitude event, the date range, the conversion window, and corrected conversion rates for this period and the last, in the conversation and in a shared report.
- **The tracking fault list.** Each broken step with what went wrong, the date it started, the evidence from Segment's delivery numbers and Amplitude's event volumes, and the code or settings change that fixes it.
- **The real leak ranking.** Each real drop with its conversion rate, how many people it loses, which groups it loses them from, and what those people did instead.
- **Chart annotations and event notes.** Marks on your Amplitude charts for every tracking break, and descriptions on the events your team should rely on.

## Common questions

### How do you calculate conversion rate in a funnel?

A funnel's step conversion rate is the number of people who completed a step divided by the number who completed the step before it. Its overall conversion rate is the number who completed the last step divided by the number who entered the first. Amplitude shows both for every funnel. The overall rate tells you how the whole journey performs, and the step rates tell you where to look.

### Why don't my Segment and Amplitude event counts match?

Small gaps between Segment and Amplitude are normal, and large or sudden ones usually point to a fault. Segment counts every call a source receives, while Amplitude counts only what the destination delivered and what Amplitude accepted. Destination filters, events blocked by your tracking plan, delivery failures, and events Amplitude rejects or blocks all widen the gap. A gap that appears overnight is worth checking in Segment's delivery numbers first.

### Should I run Amplitude's own SDK alongside Segment?

Either setup works for funnels, and the trouble starts when both send the same events. Segment is useful when you send the same events to many tools, while Amplitude's own SDK can capture things that Segment doesn't send on its own. If you run both, decide which one sends each event and pageview, and make sure they share the same device ID. Otherwise the same visitor can appear twice and your funnel's first step is inflated.

### Can Kite tell me the next time a funnel step breaks?

Yes. Ask Kite to run the same check every week, and it will compare Segment's delivery numbers with each funnel event's volume in Amplitude and tell you when an event stops arriving or a step's conversion falls sharply. Catching a tracking break in its first week keeps it from distorting a month of data.

## Related pages

- **Use case:** Analyze your conversion funnel (coming soon)
- **Use case:** Find broken links and broken signup paths (coming soon)
- **Use case:** [Improve website conversion](/use-cases/improve-website-conversion.md)
- **Use case:** [Build a sales funnel with AI](/use-cases/build-a-sales-funnel-with-ai.md)
- **Integration:** Segment + Kite (coming soon)
- **Integration:** Amplitude + Kite (coming soon)