Attribution

A customer arrives through an ad, comes back through the newsletter, then buys after typing the address. Which visit “made” the sale? Each attribution model answers differently; eodia analytics computes all six, and insights dashboards compare them.

Touchpoints

The analytics.touchpoints view gives each person’s visits, in order: their rank (rank), their channel, source, medium and campaign. The person is the matched identity if the visitor signed in, otherwise the visitor identifier (see the signed-in visitor).

To follow a person from one visit to the next, they must be recognized: measurement cookie, consent given, or signed-in visitor. In cookieless mode, the hash changes every day, and attribution is limited to the day; the view’s comment says so.

The attribution window

Only visits in the 30 days before the conversion count, by default. The window is set per site, from 1 to 90 days (Site settings › Attribution window).

The six models

model	Distribution
first_click	everything to the first visit
last_click	everything to the last visit
last_non_direct	everything to the last non-direct visit; to the direct visit only if it is the only one
linear	equal shares across all visits
position_based	40% to the first, 40% to the last, 20% spread across the others
time_decay	a share that halves every 7 days before the conversion

last_non_direct is Google Analytics’ default model apart from the “data-driven” model. There is no “data-driven” model: it requires volumes and training beyond the reach of a view.

The attribution view

One row per conversion, visit and model. Conversions are goals reached (conversion_kind = 'goal') and orders (purchase).

Column	Contents
model	the model
conversion_kind, conversion, conversion_id, converted_at	the conversion: goal or order, its name, its identifier, its instant
person	the person who converted
session_id, started_at	the credited visit
touchpoint, touchpoints	the rank of the visit in the journey, and the number of visits in the journey
channel, source, medium, campaign	the origin of the credited visit
credit	the share of the conversion attributed to the visit, from 0 to 1; the sum is 1 per conversion and per model
value	the attributed value: credit × conversion value
days_before	the time between the visit and the conversion, in days

Filter on the model

Each conversion appears once per model: filter on model before adding up credit or value, otherwise revenue is counted six times.

Comparing models

-- Revenue attributed by channel, according to each model (last 30 days)
SELECT model, channel, round(sum(value), 2) AS revenue
FROM analytics.attribution
WHERE conversion_kind = 'purchase'
  AND day >= current_date - INTERVAL '30' DAY
GROUP BY model, channel
ORDER BY model, revenue DESC

In the demo, Maison Arvor customers often come back three times before buying: the “first visit” model credits paid social, “last visit” credits direct, “last non-direct visit” credits the newsletter. None of them is wrong; they do not tell the same story.

-- The average time between the first visit and the order
SELECT avg(days_before) AS days
FROM analytics.attribution
WHERE model = 'first_click' AND conversion_kind = 'purchase'

Goals

A goal is declared in the site’s Metadata tab: a page reached (/contact/merci, * for any text) or an event (optionally with a property to check), and a value: none, fixed, or the event’s value property. The analytics.goals view lists them, and analytics.conversions gives each goal reached.