AI & Strategy, , 9 min read
Almost every marketing organisation now uses AI. Almost none runs it in regular operations. That gap is well measured by now, and the explanations are more uncomfortable than the usual answers about missing tools or lacking acceptance. This article sorts through the 2026 data and describes where pilots actually get stuck in practice.
The research is unusually consistent on this point. For its 2025 State of AI survey, McKinsey questioned close to 2,000 executives across 105 countries. 88 percent of organisations use AI in at least one function, up from 78 percent the year before. So the entry step is done.
The second number from the same survey is the interesting one: only 33 percent scale AI across the enterprise. Two thirds remain in experimentation or pilot mode. Only 39 percent report any effect on operating results at all, and for most of them it stays below five percent. Six percent of respondents qualify as high performers with an EBIT effect of at least five percent.
For marketing specifically, the Gartner CMO Spend Survey 2026, based on 401 CMOs in North America, the United Kingdom and Europe, shows a matching picture: 15.3 percent of marketing budgets go into AI, but only 30 percent of marketing organisations have mature AI readiness. At the same time, 70 percent of CMOs name AI leadership as a goal for 2026.
Since 2025 the claim has circulated that 95 percent of all AI pilots fail. It originates from work by the MIT initiative NANDA, which evaluated around 300 deployments. The figure is usually reproduced in shortened form, and a clarification is worthwhile.
What is documented is not that 95 percent fail technically. What is documented is that they show no measurable effect on the profit and loss statement. That is a difference. A pilot can work, save time and still never appear in a metric, because nobody defined which metric it should appear in.
Two secondary findings from the same work are practically more valuable than the headline. First, more than half of AI budgets flow into sales and marketing, while the demonstrable return tends to arise in the back office. Second, purchased solutions are roughly three times as successful as in-house builds. The work is not peer-reviewed, and that belongs to the assessment.
If you sort the blockers across the studies, one topic sits at the top throughout, and it is not the model.
For its State of Marketing 2026, Salesforce surveyed around 4,450 marketing decision makers. 75 percent have adopted AI. 84 percent still run generic campaigns. 98 percent run into personalisation obstacles. Data problems are named as the main blocker for effective AI use. Only about half have full access to data from service, sales and commerce.
Gartner names immature processes, insufficient data foundations, missing governance and talent gaps as causes for the lack of scalability. In Germany, according to a Bitkom survey from March 2026 covering 604 companies with 20 or more employees, another factor comes in: data protection requirements are seen as the biggest hurdle at 77 percent, followed by the shortage of skilled staff at 70 percent.
This explains why switching tools so rarely helps. A better model on a thin data base produces the same mediocrity faster. In family marketing this is sharpened further, because reliable first-party data about children and parents rarely sits in-house and, for data protection reasons, cannot simply be created either.
For consumer goods there is a dedicated measurement. For its State of Consumer Products 2026, EY surveyed over 850 executives across 24 markets. The results are more sobering than the industry average.
Only seven percent reach a maturity level that EY classifies as AI-capable. Only 15 percent have fully integrated commercial data. Only 21 percent believe they can influence algorithmic recommendations. And 63 percent see little improvement at the point of purchase despite ongoing transformation programmes.
One limitation belongs here: for children's, toy and family brands specifically, no reliable survey exists. The EY figures are the closest approximation, not the exact finding.
From consulting practice and from the studies, four patterns can be derived that separate the scaling organisations from the experimenting ones. None of them is technical in nature.
The first step is not a tool selection but a stocktaking along the marketing value chain. Three questions are enough to begin: where is time burned on routine today? Where are decisions made without data? Where does missing speed block innovation?
The answers produce a prioritisation. In practice, two or three use cases remain that can show measurable impact within 90 days. Only then is it worth discussing tools, licences and training.
The advantage of this order is not methodological tidiness. It is political: an initiative with a documented number after 90 days gets budget for the next one. An initiative without a number gets another pilot phase.
The scaling gap in AI marketing is not a technology problem. It arises from a missing data base, unclear target metrics and unassigned responsibilities. All three causes can be addressed without additional budget, but not without decisions.
Anyone investing in AI in 2026 should put less money into tools and more attention into the question of which number should move in the end. That is the more uncomfortable exercise, and it is the difference between the six percent and the rest.