§ 01

One scenario, three answers

Drag the touches. Change filters. Numbers update live.

30-day timeline · drag any marker, edit filters below

Day 1 ←———→ Day 30
March 1March 30
151015202530
Meta Ads Google Ads Campaign A Direct Purchase conversion Newsletter signup

Model A — Session-scoped

Atom: session (1-day window)
0counted

Model B — User-journey

Atom: user, window-bound
0counted

Model C — Segment-first

Atom: segment cohort
0counted
§ 02

The questions, mapped to all three models

Read horizontally — same question, three different answers.
Question Model A · Session Model B · Journey Model C · Segment
Source touch before the report range — counts?User came from Meta on March 1, converted March 15, range = March 10–20. NoThe conversion session must itself match the source filter. Yes, if within windowSource can pre-date the range; only the conversion must fall inside. YesMembership is sticky — once in the segment, always in.
No conversion filter selected — what gets reported?"All traffic from Meta in March" Sessions from Meta in MarchThe atom is the session; conversions are an optional metric on top. Users whose last source in window = MetaSample = users; conversion column may be empty. Users in "Meta segment", active in MarchStats describe the segment regardless of conversion.
Are filters a segment definition or an event filter?Does "Meta" mean: narrow down events from Meta, or find users who came from Meta and report everything they do? Event filterOnly events/sessions that match the source filter are counted. The user outside that session is invisible. Event filter with user scopeThe last matching touch defines attribution; other activity by the same user is not pulled in. Segment definitionFilters define who belongs to the cohort. All their activity in the period is then reported — including sessions from other sources.
How is conversion rate computed?What is the denominator? conversions / sessionsPer-session ratio, easy to explain. conversions / unique attributed usersPer-user, attribution-aware. conversions / segment users active in periodBehavioural rate inside the cohort.
Region: event-level or user-level?IP geolocation vs. user's home zipcode Event-levelWhere they were when the session happened. User-levelFrom first session, or "dominant" location. User-levelRegion is part of segment definition, must be persistent.
Conversion window — how long?How far back can a touch be credited? 1 daySource touch and conversion must occur on the same calendar day. Configurable, 7 / 14 / 30 / 90 daysPick from data: what % of conversions each window captures. Effectively infiniteOnce segmented, always segmented (until rules change).
Multi-touch: 3 campaigns, then conversion — who gets credit?"Meta → Google → Campaign A → Convert" The session that convertedIf conversion happened in a Campaign A session, A wins. Others = 0. Last-touch in window (default)Optionally first / linear / time-decay — separate decision. All three countUser belongs to Meta + Google + Campaign A segments simultaneously.
Multiple conversion points — count once or per point?"Purchase" and "newsletter signup" both matched by same user Once per session per pointDedupe on (session, conversion_point). Two distinct points = up to 2. Once per user per point in windowDedupe on (user, point). Same user, two distinct points = 2. Per definition of segment metricMost often: once per user per point per period. Same user = 2 if both fire.
Conversion value when there are several touches?e.g. 1200 PLN purchase, three campaigns touched the user 100% to the closing session's source 100% to last-touch (or split if model = linear / time-decay). 100% to every segment the user is in — sums across reports exceed total revenue, by design.
New vs returning user — based on what?What makes someone "new"? First session everPer session — straightforward boolean. First conversion in window"New" relative to attribution lifecycle. First entry into the segmentAttribute of segment membership.
Repeat conversions — same user, same conversion point, twice in the periodUser purchases on March 5 and again on March 18. Report range = March 1–31. Count bothEach session is independent. Two purchase sessions = 2 conversions. Count once per windowDedupe on (user, conversion_point) within the window. Second purchase only counts if it falls outside the first window. ConfigurableTypically: unique converters (1 user = 1), but can be set to count all events. Must be explicitly decided.
One user session qualifies under two source filters simultaneouslyURL has both utm_source=google and utm_campaign=campaign-a. Report filtered by "Google" AND by "Campaign A" separately. Each report picks one winnerSession has one canonical source (last UTM wins). It appears in one report, not both — unless source hierarchy is defined. Same — one last-touch sourceAttribution collapses to a single source per conversion. The session cannot be attributed to two sources simultaneously. User can belong to both segmentsIf the user ever came from Google AND ever came from Campaign A, they appear in both reports. Double-counting is intentional.