Funnels and drop-off

A funnel is the simplest useful question in analytics: of the people who started this, how many finished? Everything interesting is in the gap. A checkout that converts at 8% is not a mystery to be pondered — it is four specific steps, one of which is doing most of the damage.

What it is

You define an ordered list of two or more event names. Talaria walks each person's events in time order and records how far along the list they got. Someone who fires step one, then step two, then stops, reached level two. Someone who fires step three without ever firing step one reached level zero — order matters, and gaps are not forgiven.

Three settings shape the answer:

• Steps — two or more event names, in the order you expect them. Each step can also require a property value, so checkout_started where method = paypal is its own step.
• Scope — count conversions per session, or per person. Session is the default.
• Window — how long someone has to complete the whole sequence, in seconds. The default is 86,400, one day.

Step property filters match scalar values, because nested objects are stored as strings.

Choosing scope and window

Scope is the question of what you are counting. Session scope asks “in one visit, how far do people get?” — right for checkout, onboarding, or anything you expect someone to finish in one sitting. User scope asks “over this period, how many people eventually got there?” — right for trials, invitations, and anything that plays out over days. Under user scope the identity is the userId when there is one, and the anonymousId otherwise, exactly as in unique counting.

The window is the clock on the whole sequence, not on each hop. A 24 hour window means the last step must happen within 24 hours of the first. Set it to the time you would genuinely expect the journey to take: hours for a checkout, days for a trial. Too short and you invent drop-off that did not happen; too long and unrelated visits get glued into one apparent journey.

How to read the result

Each step shows the number of people who reached at least that far, so the counts only ever go down. The conversion at each step is that count as a share of the first step.

• One step with a much bigger fall than its neighbours is the one to work on. Broad, even decline usually means the funnel is measuring intent, not friction.
• A collapse at step one often means the step-one event is firing more widely than you intended — a pageview where you meant a deliberate action.
• Conversion that jumps when you widen the window means people are finishing, just later than you assumed.

The part other tools cannot do

Underneath the chart, Talaria lists the people who entered the funnel and never reached step two — up to 50 of them, most recent first. Each row carries the visitor id, the session, the user id if they were identified, and the last time they were seen.

The list is scoped to the first hop: entrants who matched step one and went no further. Read it as the population that stalled at the opening move, and use the step counts above for the later hops.

Each row also carries a replay id and a trace id, so the row is a set of links. Open the replay and watch the visit. Open the errors for that session and see whether something threw. Open the trace and see whether the page was simply too slow. This is the whole argument for keeping analytics beside your telemetry: the question “why did they leave?” has an answer three clicks away instead of a research project.