How to Build Path Analysis in Amplitude

Path analysis is only as good as the events feeding it. Before building any paths, audit your event taxonomy to ensure it captures meaningful user actions at the right granularity. You need events for every screen view, every key interaction (button clicks, form submissions, feature activations), and every transition point. Avoid overly granular events like individual field focus events that create noise, and overly broad events like a single 'page_view' without page name properties.

• List the 20-30 most important user actions in your product and verify each has a tracked event
• Ensure screen or page view events include a property identifying which screen was viewed
• Check that your event naming convention is consistent (e.g., 'noun.verb' like 'checkout.started')

Do not open Pathfinder and explore randomly. Start with a specific question: 'What do users do immediately after signing up?' or 'Where do users go after viewing a product page without purchasing?' or 'What paths lead to users activating feature X?' Write down the question, the starting event, and the ending event (if applicable). This focus prevents you from drowning in a sea of branching paths.

• Write the question in plain language before opening Amplitude
• Identify the anchor event (starting or ending point for the path)
• Decide whether you are analyzing paths forward from an event or backward to an event

In Amplitude, create a new Pathfinder analysis. Set your starting event (or ending event if you are working backward). Configure the user segment to match your target audience — for example, new users in their first 7 days or power users who have completed onboarding. Set the time range to at least 30 days for statistical significance. Amplitude will generate a Sankey-style diagram showing the most common paths users take from your anchor event.

• Select 'Starting Event' and choose your anchor event from the dropdown
• Apply user segments to focus on the cohort you care about
• Set the path depth to 5-7 steps to balance detail with readability

The raw Pathfinder output will likely show dozens of paths, many of which are noise. Use Amplitude's event grouping to collapse related events (e.g., group all 'settings.*' events into a single 'Settings' node). Hide events that appear in paths but are not relevant to your question (e.g., 'notification.dismissed'). Expand only the paths with the highest user volume to focus on what matters most.

• Use the 'Hide Events' feature to remove low-signal events from the visualization
• Group related events using custom event categories
• Click on specific path branches to drill into the user count and conversion rate at each step

Look for two patterns in your path analysis. First, drop-off points: steps where a large percentage of users exit the flow entirely. These indicate friction, confusion, or unmet expectations. Second, unexpected detours: steps where users navigate to a screen you did not expect them to need. These indicate missing information or unclear UI in the preceding step. Annotate both findings directly in Amplitude or in a shared document.

• Calculate the drop-off percentage at each step by comparing user counts between nodes
• Flag any step where more than 30% of users leave the flow
• Note unexpected paths taken by more than 10% of users — these are worth investigating

The most powerful insights come from comparing paths between different user groups. Create two Pathfinder analyses with the same starting event but different segments: users who converted versus those who did not, free users versus paid users, or mobile versus desktop users. Overlay the results to see where the paths diverge. The divergence point is usually where the converting group found value and the non-converting group did not.

• Create a 'Converted' segment and a 'Did Not Convert' segment for your key outcome
• Build identical Pathfinder charts for each segment and compare them side by side
• Document the specific steps where the paths diverge as hypotheses for experimentation

Path analysis insights are only valuable if they lead to changes. For each finding, create a hypothesis and a proposed experiment. Drop-off at step 3 of onboarding might lead to a hypothesis that the step asks for too much information, with an experiment to test a simplified version. Unexpected detours to the help page might indicate that a tooltip or inline explanation is needed. Track the experiment results back to the path analysis to close the feedback loop.