Most products have multiple funnels. Most analytics tools assume one.
Pendo, Mixpanel, Amplitude, Heap — every analytics tool you’ve used assumes you already know which funnel matters. You define stages by hand. You pick events by hand. You name the cohort by hand. And you get exactly one funnel back: the one you already had a mental model of.
That’s a problem. Because most products have three or four funnels operating in parallel. A marketplace has demand-side and supply-side journeys. A SaaS has signup, activation, and upgrade. A content product has anonymous browse and authenticated read. The funnel you defined is rarely the one with the biggest leak.
The Onboardics founder learned this on his own product. He thought the Hodego pilot funnel was Landing → Explore → Expert Profile → Join Hodego → Login → Booking Confirmed. Six stages. He’d been telling everyone the leak was at “Join Hodego.”
Then he ran the SQL. Zero of the four converters had ever visited /join-hodego. /join-hodego is the supply-side path — experts applying to be on the platform. The demand-side funnel is five stages, not six. The real leak was at Expert Profile → Login, where 84% of users dropped off.
Founders have confident-but-wrong mental models of their own funnels. Onboardics is built to catch that.
How auto-detection works
Click “Suggest funnels with AI”
In Settings → Funnels (Beta), one click starts the analysis. AI inspects your real path data, custom event_types, and conversion signals. No configuration.
AI proposes multiple funnels
Demand-side, supply-side, content, upgrade — whatever exists in your data. Each funnel comes with a name, ordered stages, terminal conversion event, confidence score (0.0–1.0), and rationale.
Accept or dismiss
Accept the funnels that match your mental model. Dismiss the rest. Each accepted funnel becomes its own dashboard view, drop-off chart, and AI Briefing — tracked independently.
What you see when AI runs
From a real run on a Calendly-like booking project (130 sessions, 12% conversion):
The first two are obvious accept. The third is borderline — low confidence because the AI saw the path but couldn’t verify it as a primary journey. Dismiss it; if it’s wrong, the next month’s data will surface it again with higher confidence.
The AI also emits a data_constraints field listing what was missing from the input (e.g. “identify() not called on booking-confirmed users”). When the AI passes on something, it tells you why — not silent failure.
Same pattern for audience segments
At /dashboard/segments, click “Suggest segments with AI” and the AI surfaces meaningful audience cohorts from real engagement signals: visit_count distributions, identify() property splits, page-set affinity. Examples it might propose:
- Visited pricing 3+ times — high-intent prospects who haven’t signed up yet
- Identified users who never published — activation drop-offs worth re-engaging
- High-engagement returning users — visit_count ≥ 5 with custom event activity
It refuses to propose segments it can’t verify (mobile vs desktop, returning vs new, identified vs anonymous — those are heuristics already built in or signals not present in current data). And every refusal is logged in data_constraints with what was missing.
Eval-validated against four product archetypes
Both suggesters are tested against an internal eval harness covering four synthetic project shapes:
- Calendly-like booking (parameterized expert routes, single canonical funnel)
- Notion-like collaborative SaaS (mixed event_types: workspace_created, doc_created)
- Stripe-like e-commerce (parameterized product routes, classic browse → cart → checkout)
- Multi-funnel SaaS (demand + supply + upgrade journeys in one project)
Acceptance threshold: ≥0.80 precision, ≥0.70 recall, on every archetype. The full eval runs as a regression on every prompt change.
Multiple funnels per project, mixed stage types
- Define unlimited funnels per project — switch between them with the FunnelSwitcher dropdown
- Each funnel renders independently: drop-off chart, AI Briefing, segment overlays
- Stages can mix path-based steps (page_view at /pricing) AND custom event_types (signup_complete, workspace_created) in the same funnel
- Per-funnel AI Briefing — cache key includes funnel_id, so each funnel + segment combination gets its own analysis
- Segment-aware: pick a segment from the global selector and every analytics page re-scopes to that cohort. Per-(view × segment) cache keys mean each cohort+view gets its own AI briefing
Why we call this the discovery half
Every analytics tool can show you a funnel after you tell it which one matters. That’s the execution half: charts, drill-downs, exports. We do that too — AI Diagnosis ships the prescription, the flow builder ships the fix on Deploy and above.
The discovery half is the part nobody else does: tell the operator which funnel matters before they ask. Surface the demand-side leak when they were focused on supply. Find the segment buried in visit_count distributions they never sliced for. Catch the confident-but-wrong mental model before it becomes a launch-day claim that doesn’t hold up to SQL.
Onboardics finds the leak. Onboardics tells you what to change. The first half is what nobody else does.
Pricing
Auto-suggested funnels: available on every tier, including Free. Multiple funnels per project: every tier.
Auto-suggested segments + segment-aware dashboards/briefings: available on Diagnose ($149/mo) and above.