Which metrics exist and how are they calculated?

The conversion rate per channel divides the attributed shares that fall on visits from this channel by the measured visits. Shares on clicks, impressions and post-view do not count, because they have no counterpart among the visits. It is above 100% only if visits carry several conversions. It mixes two layers and is therefore called “attributed”: switch the model and it changes, the visits do not. A visit starts with the first event of a session, after a 30-minute pause or with a new campaign signal, so its count differs from other tools’ sessions. The new customer share counts only orders whose customer type is known. Leads do not count, even if your shop sends a customer type with them; if it is known for no order, no value is shown.

CPA and CAC are not the same. CPA divides ad spend by all attributed conversions. That gives a price per completion, and a completion can come from a returning customer. CAC divides only by conversions of new customers. That needs the customer type: either your shop sends it with the purchase, or it is derived from our own measurement, which only knows the time since measurement began. The tile therefore states for what share of conversions the customer type is known: the figure stands on those only.

If you filter the reports to one conversion type, say “purchase”, the cost stays the total of the period: ad spend is incurred for campaigns, not for a type of conversion. The numerator is then only the attributed revenue of that type. That is the purchase ROAS (purchase revenue against all cost), and the tile says what it stands on. A CPA per purchase arises the same way.

Funnel steps count like any conversion, with one difference in value. If the event brings an amount, that is revenue and part of ROAS. The acquisition value from the portal is set by you: it has columns of its own and never appears in revenue or ROAS. Narrowing to a single step works the same way as narrowing to “purchase”.

Attributed conversions can carry decimals. If a model spreads one conversion across three touchpoints, each takes a third. In tables they therefore always carry exactly one decimal place (265.0 above 295.5). Key figure tiles show a whole value without decimals and a fraction with two.

Monetary values are computed as decimals, not floating point numbers. That sounds technical but has a practical effect: totals are correct to the cent, even across many rows.

Is my advertising worth it?

ROAS tells you how much attributed revenue comes per euro of ad spend. Whether you make money on it depends on your margin: a ROAS of 3 is a loss at a 20 per cent margin and a profit at 60 per cent. If your shop sends the cost of goods, “Commerce & products” shows the margin ROAS, the attributed contribution margin divided by ad spend. Per channel it appears in the channel table of the overview, per channel and per campaign in the breakdown under “Attribution”. Below 1.0, the attributed orders leave less after cost of goods than the ads cost. If only some of a row’s orders carry the cost of goods, the share appears below the figure: the costs are those of the whole row, the margin only that of these orders. If the cost of goods is missing entirely or costs are missing, no value is shown, only the reason. Whether a campaign caused the orders is answered by none of these figures; only an experiment measures that (see “Which campaign actually delivered?”).