Pick the steps
Choose two to ten events from your catalog and drag them into order. Any event you send can be a step — there is nothing to model, define or pre-aggregate first.
signup_completedproject_createdactivated
// funnels
Find the step where people give up
Pick the events, set a window, and get the drop-off, the median time to convert and the segment behind every loss. Three clicks, no SQL, no pre-built schema.
Signup → Activation · 30dUpdated just nowsignup_completed18,420100%↳ 3,520 dropped · 19.1% · 3m median
email_verified14,90080.9%↳ 3,660 dropped · 24.6% · 1h 12m median
project_created11,24061.0%↳ 5,060 dropped · 45.0% · 4h 30m median
teammate_invited6,18033.5%↳ 1,050 dropped · 17.0% · 1d 2h median
activated5,13027.9%Build a funnel in three steps
Funnels are computed from raw events at query time, so there is nothing to define in advance and nothing to rebuild when you change your mind.
Set the window and the order
Give users a conversion window — seven days by default — and decide whether the steps have to happen in sequence or merely all happen. Both are one click, and the chart redraws immediately.
7-day windowstrict order
Break it down
Split the funnel by any property on the entry event: plan, source, country, app version. Then save it as a report so the number arrives every Monday without anyone opening the tab.
by sourceby plansaved report
Change a step, a window or a date range and the funnel recomputes against the full event history — not a snapshot taken when you first saved it.
What the numbers actually mean
Funnel maths is easy to get subtly wrong, and a conversion rate you cannot explain is worse than no conversion rate. Here is exactly how Alitycs counts.
- Conversion window
- Each user gets a fixed amount of time from their first step to their last — seven days by default, anything from one hour to ninety days. Someone who signs up on Monday and activates three weeks later does not convert in a seven-day window; they show as dropped at the deepest step they reached.
- Strict order
- The listed steps must happen in the order you listed them. Other events in between are fine and ignored. Use this for onboarding, where the sequence is the product.
- Any order
- All steps must happen inside the window, in any sequence. Use this for feature adoption, where the real question is whether someone used three of these five things at all.
- Repeat steps
- Only the first occurrence of each step counts. Completing step three twice does not convert a user twice, and a user is counted once per funnel at the deepest step they reached.
- Counting basis
- Unique users by default. Switch to sessions when you want to know whether a single visit converts rather than whether a person eventually did.
- Late and out-of-order events
- Events are placed by their own timestamp, not by when they arrived — so an offline mobile client or a server retry lands in the right place. Funnels recompute as those arrive, which is why a fresh number can move slightly for a few hours.
Where the drop-off is, and how long it takes
A conversion rate tells you that something is wrong. The step chart, the timings and the segment split tell you where to look first.
Read it from the widest gap down
Each step shows the users who reached it, the share of the previous step they represent, and the number lost between the two. The largest single loss is called out, because that is nearly always the cheapest thing to fix.
Alongside the counts you get the median time between steps. A step with a good conversion rate and a four-hour median is a different problem from one with the same rate and a four-minute median — the first is friction, the second is confusion.
Click any step to see the users who dropped there, the last event they sent, and the properties they had in common.
From first step to activation. The p90 is 6d 2h, which is why a seven-day window is the sensible default for this funnel.
5,060 users created a project and never invited anyone — comfortably the biggest single loss in the funnel, and the one worth fixing first.
Web SDK signups activate at 31.2% against 24.8% on iOS. Same product, same funnel, six and a half points of difference.
Compare two cohorts on the same steps
Any funnel can be split by a property on the entry event, or by a saved cohort. The steps stay identical, so the difference between two audiences is the only thing that moves.
| Step | Web SDK | iOS SDK | Gap |
|---|---|---|---|
signup_completed | 100% | 100% | — |
email_verified | 84.1% | 79.6% | −4.5pp |
project_created | 62.0% | 51.3% | −10.7pp |
teammate_invited | 38.4% | 29.7% | −8.7pp |
activated | 31.2% | 24.8% | −6.4pp |
Rates are share of the cohort that signed up, over the last 30 days with a seven-day conversion window. The gap opens up at project_created, which points at the mobile project-creation flow rather than at onboarding as a whole.
Every funnel is also a question you can just ask
A saved funnel and a question to the AI agent run against the same events and return the same numbers. Build the funnel when you want to keep it; ask when you want to know right now.
Where do users drop off in onboarding?
Returns this exact funnel, with the invite step flagged.Did activation improve after the new empty state?
Splits the same funnel before and after the release date.Which source has the worst verification rate?
Breaks step two down by first-touch source and ranks it.See how the AI agent answers questions — including the events, filters and row counts behind every number.
Build your first funnel this afternoon
If your product already sends events, the funnel is three clicks away. If it does not, the quickstart takes about five minutes.
Saved funnels and cohorts are included from the Pro plan up.