Product analytics over behaviour, with delivery reported alongside it, so the two are read together.
The model proposes a plan, the plan is validated, deterministic code compiles and runs it. A bad plan fails loudly. It never returns a plausible wrong number.
Conversations are kept, so a line of enquiry can be resumed and audited.
Eight kinds of event tile, segment overlap and leaderboards, and channel reachability: not how many people are in a segment, but how many you can contact on each channel. Shared boards are cached and pre-warmed.
Per-contact, in-order steps inside a window you set. Segment by channel, country or property at the entry step.
It separates what is derived by formula from what merely moves together, and shows the sample size behind every correlation. Open it and every node carries its current value.
Eight standing views, including delivery performance with campaigns excluded, savings from routing in your own currency, and the fallback paths actually taken.
Define which events count and what each is worth. Campaign lift, funnels and journey attribution all measure against it.
Sessions are tracked automatically on web and app. Open one and see the entry point, the exit point, and how long the person spent on every step. Thirty seconds on a product page and four minutes on a form is a form with a problem.
Start a journey when a session ends that way. Frequency caps, fail-closed evaluation, one trigger per rule per session. A misconfigured rule sends nothing, not everything.