Loading tutorials…
Loading tutorials…
Funnels show you the path you expected. Pathfinder shows you the paths users actually take. The gap is where 60% of product insight lives — and where most teams never look.
Who this is forProduct, design, and growth teams ready to move beyond hypothesis-driven funnels into exploratory path analysis. Especially relevant when you suspect users are navigating your product differently from how it was designed.
What you'll need
Step 1
Amplitude → Charts → "New Chart" → "Pathfinder." Choose whether you want to see paths LEADING TO an event or paths FROM an event.
Open Amplitude → Charts → "New Chart" → "Pathfinder Analysis."
You'll see a choice: "Paths from event" (what do users do AFTER doing X?) or "Paths to event" (what did users do BEFORE doing X?).
For activation investigation: pick "Paths from Signup Submitted" — what happens in the next 5-10 events after signup?
For conversion investigation: pick "Paths to Subscription Started" — what did paying users do in the 5-10 events before paying?
For churn investigation: pick "Paths from [last event seen]" for users who stopped showing up — though this is harder to construct cleanly.
Step 2
Apply a cohort filter or event-property filter before running Pathfinder. Unfiltered, the chart shows every path of every user and becomes unreadable.
Click "Filters" panel → add a cohort filter or a property filter.
Common filters: cohort "Active Users L30," `plan_tier = pro`, `signup_source = organic_search`. Each narrows the user base to a meaningful slice.
For investigation pattern: filter to ONE persona/cohort at a time. Compare paths between cohorts side-by-side (Pro vs Free) to see behavior differences.
Set the time window — usually 14-30 days post the start event. Longer windows include too much noise; shorter windows miss the journey.
Without filters, expect to see 200+ unique paths and learn nothing. Filtered to a 5K-user cohort, you usually see 10-20 dominant paths.
Step 3
Default is 5 steps. Increase to 10 for deep-journey analysis, decrease to 3 for quick-wins. More steps = more noise.
Find "Path Depth" or "Number of Steps" in the chart settings.
Depth 3: shows immediate next-actions. Best for "what do users do right after onboarding?"
Depth 5 (default): the sweet spot for most analysis. Shows the dominant 3-5 sessions of behavior.
Depth 10+: useful for long-cycle B2B where the conversion journey spans weeks. Expect more noise.
You can also set "Max time between events" — events more than X hours apart aren't treated as the same path. Default 30 minutes; raise to 7 days for long-cycle analysis.
Step 4
Pathfinder renders as a Sankey diagram. Width of each flow = how many users took that path. Wider = more common.
Each node = an event. Each ribbon = a transition between events. Width = unique users on that path.
Look first at the WIDEST ribbons leaving your start event. These are the dominant paths — usually 3-5 paths account for 60-80% of behavior.
Look next at the THIN ribbons. These are the weird paths — small in volume but often revealing. Example: "10% of users who signed up immediately viewed the Settings page" — that's a signal.
Hover over any flow to see the user count + percentage. Click any node → "Users on this path" to drill into 5-10 specific users in User Lookup.
Compare to your funnel: in the funnel you defined Signup → Activated → Subscribed. Pathfinder might show users actually do Signup → Settings → Activated → Cancelled. The unexpected step is the insight.
Step 5
A dead-end event = a step after which users disproportionately drop off. A golden path = a sequence that disproportionately leads to conversion. Both are gold.
Dead-end signal: a node where the outgoing ribbon to "Session Ended" or "No further event" is unusually wide compared to other nodes.
When you find a dead-end, ask: "Is this page broken? Confusing? Missing a clear next-step CTA?" Frequently a CSS/UX fix recovers 10-30% of the drop-off.
Golden path signal: a sequence of 3-5 events that disproportionately appears before your conversion event. Filter Pathfinder to "Paths to Subscription Started" and look for the wide ribbons.
Once you find a golden path (e.g., "Signed Up → Viewed Templates → Created First Project → Invited Teammate → Subscribed"), redesign onboarding to push users into that sequence.
Golden path optimization typically lifts conversion 15-40% — among the highest-leverage product work you can do.
Step 6
Build two Pathfinder charts side-by-side — one filtered to converters, one to non-converters. The diff is the answer.
Create chart 1: "Paths from Signup, cohort = Converted within 30 days."
Create chart 2: "Paths from Signup, cohort = Did NOT convert within 30 days."
Drop both into the same Notebook. Compare visually.
Look for: events that appear in chart 1 but not chart 2 (these are the converter behaviors), and events that appear in chart 2 but not chart 1 (these are the friction points).
Common finding: converters viewed pricing TWICE before converting. Non-converters viewed it once and bounced. Hypothesis: pricing page needs a stronger second-visit CTA. Test and measure.
Step 7
Save Pathfinder charts to a Notebook with a written narrative. Convert findings to specific product hypotheses you can test.
Pathfinder findings are easy to forget if not written down. The chart is overwhelming visual; the narrative is what makes the insight portable.
Write each finding as: (1) what we observed, (2) hypothesis for WHY, (3) one experiment to test the hypothesis, (4) expected lift if hypothesis is right.
Push the experiment into Amplitude Experiments (separate feature — see our `set-up-amplitude-experiments` tutorial) or your existing A/B test tool.
After 30 days, return to the same Pathfinder analysis. Did paths shift? If yes, hypothesis was right. If no, revise.
Pathfinder pays off through iteration, not one-shot insight. Treat it as an ongoing investigation surface, not a one-time report.
Common mistakes
Running Pathfinder without filters
What goes wrong: Chart shows 200+ paths. The analyst gives up after 10 minutes and concludes "Pathfinder is too noisy." The team writes off the most powerful chart Amplitude has. Lost opportunity: 3-6 months of un-found product insights worth probably $50K-300K in retention/conversion impact.
How to avoid: Always filter to a specific cohort or property segment. Pathfinder is a comparison tool, not a snapshot tool.
Comparing path widths without normalizing for cohort size
What goes wrong: Cohort A has 10,000 users; Cohort B has 1,000. Path "X → Y" is 30% in cohort A and 50% in cohort B. Raw widths in Pathfinder make cohort A look dominant. Strategy decisions drift toward the larger cohort's behavior even when the smaller cohort is the more interesting signal.
How to avoid: Always look at PERCENTAGES, not raw widths. Amplitude shows both on hover; click the % toggle to make percentage the default display.
Picking a path depth too shallow or too deep
What goes wrong: Depth 2 misses the journey. Depth 15 is noise. Either way, the analysis produces no insight and the analyst concludes the data is bad. $5K-15K of analyst time spent on a misconfigured chart.
How to avoid: Default to depth 5. Increase to 8-10 only for long-cycle B2B. Decrease to 3 only for immediate-next-action questions.
Trusting Pathfinder without checking the underlying events
What goes wrong: A "weird path" shows up. The team panics. Turns out the underlying event is misnamed or fires twice. The investigation was on phantom data. Wasted: 20-40 analyst hours plus a potential bad product change.
How to avoid: Whenever Pathfinder surfaces a counterintuitive path, drop into User Lookup for 5-10 specific users on that path. Verify the events are real and meaningful before drawing conclusions.
One-shot Pathfinder analysis instead of iterative
What goes wrong: Team runs Pathfinder once, picks an insight, ships a change, and never returns. The change might have helped or might not have — they don't check. Pattern of "open Pathfinder, get insight, move on" produces 1-2 wins but misses the 80% of value that comes from iteration.
How to avoid: Treat Pathfinder as an ongoing investigation. Re-run the same analysis monthly. Track changes in path widths. Connect to specific experiments.
Recap
Done — what's next
How to build Amplitude funnels that actually answer business questions
Read the next tutorial
Hand it off
Pathfinder is the chart most teams own but never master. A specialist who runs Pathfinder analyses weekly across multiple SaaS clients will spot patterns you'd miss for months. Initial Pathfinder workshop + 3 chart builds typically $300-700; ongoing analysis $400-900/mo at $14-16/hr.
See specialist rates
Pathfinder requires Plus plan or higher. On the free Starter plan, you'll see Pathfinder mentioned in marketing but not available as a chart type. Upgrade plan, or use Funnels + Cohorts as a partial substitute.
A funnel tests a pre-defined sequence. Pathfinder discovers all sequences. Funnels are good for "did this expected flow happen?"; Pathfinder is good for "what actually happens?" Most teams use both — Pathfinder for discovery, Funnels for measurement.
Yes — Amplitude exports path data as CSV (right-click chart → Export). For deeper analysis, use Amplitude's SQL feature (Data Tables) or BigQuery export. Pathfinder visualization is for exploration; analytical work usually happens in SQL after the discovery phase.
Minimum 1,000 users in your filtered cohort. Below that, paths are too sparse — you see 50 unique paths each with 5-10 users and no clear dominant flow. For confident insights, you want 5K+ users in the cohort.
Most teams build funnels first (hypothesis-driven), then use Pathfinder when funnels show unexpected drop-offs (exploration). Pathfinder explains the "why" behind funnel drop. Use them together rather than choosing between.
Amplitude
Funnels are the most-used chart type in Amplitude and the most-misused. The defaults assume an attribution window that doesn't match most SaaS sales cycles. Here's how to build funnels that match reality.
Amplitude
Cohorts and Personas are where Amplitude beats GA4 by a mile. But most teams build sloppy cohorts and end up with overlapping segments that conflict. Here's the discipline that keeps them clean.
Amplitude
A/B testing inside Amplitude pairs experiments with the analytics that measure them. But running an experiment correctly is harder than launching one. Here's the framework that keeps you out of false-positive trouble.
Amplitude
DIY Amplitude is a great idea — until your taxonomy gets out of control or your charts disagree with reality. This is the honest framework for when the math flips toward hiring.