How to Build a Funnel Analysis in Amplitude

Open most teams' Amplitude workspace and the funnels look the same: a chart built once during onboarding, three or four generic page-view events strung together, a single conversion percentage at the bottom that someone screenshots for a slide, and then nothing. Nobody comes back to it, nobody segments it, and nobody notices when that percentage quietly drops ten points over a month. That's not an Amplitude problem, it's a setup problem, because a funnel built around page views instead of a real user journey was never going to tell anyone why users actually leave. Amplitude's funnel analysis is genuinely built to answer that question, step by step drop-off, segmented by who's dropping and why, not just a top-line rate. For background on what the platform covers and where it fits against other product analytics tools, see our full Amplitude profile. Here's the workflow that turns a funnel into something a team actually acts on.

Step 1: Define the Event Sequence Around a Real User Journey, Not a String of Page Views

The fastest way to build a useless funnel is to chain together whatever page-view events already exist in the tracking plan: viewed homepage, viewed pricing page, viewed signup page. That sequence measures navigation, not intent, and it will always show a steep drop-off because most visitors were never going to convert in the first place. A funnel worth building starts from the actual decision points in the journey, events like started checkout, added payment method, and completed purchase, or for a product-led flow, created workspace, invited a teammate, and activated a core feature for the first time. Those are the moments where a user is actually choosing to move forward, and they're the ones where a drop-off means something specific went wrong.

Getting this right means sitting down with whoever owns the tracking plan before opening the funnel builder, not after. If the event that should mark "started checkout" actually fires on a button click that doesn't guarantee the checkout page loaded, or if "activated" is defined differently in the data than in the team's own retention conversations, the funnel will report a number that looks precise and means the wrong thing. A few minutes confirming what each event actually captures saves a much longer argument later about why the numbers don't match what the product team already believes.

Step 2: Set a Conversion Window That Matches How Long the Journey Actually Takes

Amplitude's funnel chart defaults to a one-day conversion window, meaning a user only counts as converted if they complete every step within one day of entering the funnel. That default quietly punishes any journey that legitimately takes longer, a B2B trial that converts to paid over two or three weeks, or an e-commerce cart that gets finished on a second visit after payday. Left on the default, a funnel like that will report a conversion rate far lower than reality, and a team that doesn't know to check the window setting will chase a "drop-off" that's really just people converting on day four instead of day one.

The window is adjustable from as short as one second up to 366 days, so the right move is to set it to match the journey being measured rather than accept whatever Amplitude assumed. A same-session checkout funnel might reasonably use a window measured in minutes. A trial-to-paid funnel needs a window covering the actual length of the trial. Getting this one setting wrong is the single most common reason a funnel's headline number gets dismissed as inaccurate.

Step 3: Choose Ordered or Any-Order Funnel Logic Based on How the Journey Actually Behaves

Amplitude gives a funnel two ways to treat the sequence of steps, and picking the wrong one produces a number that looks confident and is quietly wrong. An ordered funnel in its default, looser mode still counts a user as converted even if same-day events happened out of the exact sequence, but events on different days still have to follow order. Switching to strict mode requires the exact sequence every time, with no flexibility, and it's what unlocks the shortest conversion windows, down to two seconds, for journeys where sequence and speed both matter. An any-order, unordered funnel drops the sequence requirement entirely and just checks whether a user completed every step at all, which is the right choice when steps genuinely can happen in any order, like a user exploring three different product features before adopting one, but the wrong choice for a checkout flow where doing steps out of order usually means something broke.

The mistake to avoid is defaulting to whichever setting Amplitude opens with and never revisiting it. A strict-order funnel applied to a flow where users legitimately branch and circle back will report a conversion rate that's artificially low, penalizing normal behavior as a failed conversion. Matching the setting to how the journey actually works is what makes the resulting percentage trustworthy enough to act on.

Step 4: Segment the Funnel by User Property and Cohort to Find Where Specific Groups Drop Off

A single aggregate conversion number hides more than it reveals. A funnel converting at forty percent overall might be converting at sixty percent for users on desktop and twenty percent for users on mobile, or converting well for users acquired through organic search and badly for users from a specific paid campaign. None of that is visible until the funnel gets broken down, and Amplitude's Group By control is what does it, adding a user property like device type, acquisition channel, plan tier, or geography as a breakdown dimension so the same funnel chart shows conversion separately for each value instead of blended into one line.

Property breakdowns answer "which segment," but a saved cohort answers a more specific question: exactly which users, built from any combination of events, properties, and time windows, are stalling at a particular step. Building a cohort of everyone who hit "added payment method" but never reached "completed purchase" turns a percentage into an actual list of people, and that cohort can be reused directly in a retention chart, a path analysis, or an experiment without rebuilding it from scratch. The Compare control on a Group By breakdown adds one more layer, letting a specific segment's conversion rate become the baseline so every other segment's difference shows up directly. Skipping segmentation is how a funnel keeps reporting "forty percent" for months while the real story, one channel or device quietly failing, sits unnoticed underneath it.

Step 5: Compare the Funnel Against a Prior Period to Catch a Regression Before It Compounds

A funnel checked once and left alone will not tell anyone when something breaks, because a static chart has nothing to compare itself against. The comparison control next to the date picker, Previous Period vs., fixes this by overlaying the current time range's conversion against an earlier one directly in the same chart, so a step that dropped five points after a release is visible immediately instead of getting lost in a number that just looks unremarkable on its own. This is the difference between noticing a regression the week it happens and noticing it a quarter later when someone finally asks why signups are down.

The habit worth building here is checking the comparison after every release that touches a step in the funnel, not just during a periodic review. A checkout redesign, a pricing change, or a new onboarding flow should each get a before-and-after look at the specific step it touched, because a drop that shows up right after a known change is far easier to trace than the same drop discovered weeks later with other changes shipped in between.

Step 6: Turn the Finding Into a Specific Action Instead of a Chart Nobody Revisits

A funnel that ends at "here's where users drop off" has done half the job. The number only earns its place on a dashboard if it leads somewhere: a UX fix for a step that's failing broadly across all segments, or a targeted campaign, an email or in-app message aimed specifically at the cohort of users who stalled at one step, when the drop-off is isolated to a segment that's otherwise valuable. A checkout funnel showing a sharp drop right after a shipping cost gets revealed points at pricing transparency earlier in the flow, not a vague note to "improve checkout." A trial funnel showing mobile users stalling at account setup points at a specific screen to fix, not a general mobile initiative.

Whichever action gets taken, the funnel isn't finished once it ships. Re-running the same funnel, with the same event sequence, window, and segmentation, after the fix goes out is what actually confirms whether the drop-off closed or just moved to a different step. A team that treats the funnel as a one-time diagnosis rather than a chart to return to will keep rebuilding the same analysis from scratch every time the question comes up again, instead of watching one saved chart do that job continuously.

What This Workflow Actually Produces

Done this way, an Amplitude funnel stops being a screenshot of a single conversion percentage and becomes an actual diagnostic tool: an event sequence built around real decision points instead of page views, a conversion window that matches how long the journey genuinely takes, ordered or any-order logic chosen to fit how users actually behave rather than left on default, segmentation by property and cohort that shows exactly which group is struggling, a period comparison that catches a regression while it's still fresh, and a finding that ends in a shipped fix or a targeted campaign rather than a static chart. That's a meaningfully different outcome than the funnel most teams end up with, and it's the difference between a dashboard that quietly drives product and marketing decisions every week and one that got built once, presented once, and forgotten.

If you're still working out which events belong in your core funnels or how Amplitude's segmentation and cohort tools compare to what you're already using for reporting, our Amplitude tool profile covers the feature set and where it fits alongside the rest of an analytics stack in more detail.