MarketingEvidence Brief (Crabstone)
A compositor lifts the 'all customers' line out of a typeset fraction beside a printed trade page headlined '44% more new customers', tagged 'Anzeige' and stamped with the Google G, next to a bonprix shopping bag.

bonprix's 44% Lost Its Denominator in an Advertorial

Google's case study said bonprix's new-customer rate rose 44% in AI Max test markets. Sponsored pages in OMR and TextilWirtschaft turned that into 44% more new customers, with no count, baseline or holdout behind it.

Sir John Crabstone

On 12 October TextilWirtschaft, Germany’s fashion trade weekly, ran a DMEXCO recap headlined “44 Prozent mehr Neukunden”, or 44% more new customers, for bonprix’s work with Google AI. The case study behind that figure reports a 44% rise in a new-customer rate, which is a different thing. The recap is tagged “Anzeige”, advertisement, and its footer says the advertising partner, not the editors, answers for its content. The partner goes unnamed; every outside link in its text leads to a Google page.

The case study, written in January by two Google employees, describes bonprix’s beta test of AI Max, which reads a query’s intent and picks the landing page. The “Neukundenquote”, or new-customer rate, as measured by Google’s new-customer acquisition (NCA) function, rose 44% “in den Testmärkten”, in the test markets. Cost per new customer fell 22% and the conversion rate rose 15.3%. All three are ratios, and the page never defines the first; the share-of-buyers reading follows Google’s help-centre term “new customer ratio”. The case study credits AI Max alone, while the recap credits “AI Max und Live-Katalog”.

The page is loose in two places. It files the figures under campaign KPIs from Italy, then captions the chart by calling Italy’s results “besonders stark”, particularly strong, as if other markets were included. The chart’s lettering says the rate rose. Its alt text, written for screen readers and search engines, reads “+44 % Neukunden” and “-22 % Kosten”.

A ratio can rise while the count beneath it falls. Google’s help centre tells advertisers to judge the new-customer setting by this ratio, before and after: customers from the segment who converted, divided by all customers who converted. Suppose a campaign wins 100 buyers, 25 of them new, and later wins 50, of whom 18 are new. Its new-customer rate has risen 44%. It has seven fewer new customers. The same page defines acquisition cost as spend on new customers divided by a distinct count of them, so the example squares with a 22% fall only if that spend fell by more than 22%. Google Ads reports the counts; bonprix holds them. The page’s remedy for a falling ratio is to bid more for new customers, and a higher ratio can be bought.

Who counts as new is, by default, Google’s call. Google builds its list of existing customers from “up to the last 540 days of campaign activity and tracked purchases”. Its help centre calls the method “less comprehensive because users can remove cookies and opt out”. Advertisers can add their own customer lists; no page says whether bonprix did. A regular who clears their cookies may count as new.

Sponsored copy made the count explicit. In January the marketing publisher OMR ran a piece marked “Sponsored”, at a web address ending “google-sponsored”. It said bonprix, testing AI Max since April 2025, had raised “die Anzahl der Neukund*innen”, the number of new customers, by 44% in the Italian test phase. The October recap dropped Italy and the test.

The recap does footnote Google’s own claims. Its line that each euro invested in search returns advertisers €6 on average cites “143 Conversion-Lift-Studien”. The bonprix figures have no footnote. They are credited to the brand “nach eigenen Angaben”, by its own account, and linked to an account written by Google.

The new-customer setting cannot show whether its customers would have come anyway. “Incrementality testing of lifecycle goal settings isn’t supported within Google Ads,” Google’s help centre says, pointing advertisers to its Conversion Lift studies instead. The case study and OMR both assert incrementality: the first says “einige der Tests gaben Entwarnung” on cannibalisation, and the second says growth came “nicht, indem einfach mehr Budget in die Hand genommen wird”. None of the three pages cites a Conversion Lift study for bonprix, and none gives a baseline, a spend level, a holdout or a count.

Percentages travel well because they leave their denominators at home.

A board handed this recap would do well to ask for the count, the period and the holdout. The recap sets the 44% straight after a quote from bonprix’s CMO, Dr. Nicolai Johannsen, on the DMEXCO stage, without saying what he presented. It closes by promising that AI Max’s new reports now let marketers prove this value to management. The link goes to Google’s 23 September announcement: a single view of search terms, ad assets and landing pages, with availability to be detailed later in the year. The retailer can count its own customers; none of these pages prints the count.