AI Is Everywhere in Marketing. Almost Nobody Has It Working.
Industry data shows marketing leads on AI adoption but trails on implementation — and the gap between the two is getting harder to explain away.
Marketing’s reputation as an AI-forward function is not bad. Adoption rates across the sector outpace most other departments. But the tools being used are not the same as the workflows being changed, and the data is already reflecting that distinction.
McKinsey’s research into European marketing organisations found that 94% have not moved beyond the pilot phase in their generative AI maturity. Alongside that sits a finding that should give AI vendors pause: generative AI ranked 17th out of 20 on the average CMO’s priority list. The gap between how much the industry talks about AI and how senior marketers actually prioritise it is considerable, and the pilot phase figure helps explain why.
Pilots, by design, are bounded. They run in controlled conditions, with defined success criteria, a small team, and an exit point. What most marketing organisations have not worked out is what comes after a successful pilot, how to take something that worked in a limited context and make it a reliable part of how work actually gets done. That transition requires process design, training, governance, and someone with clear accountability for the outcome. Most marketing teams have none of those things in place, which is why 94% are still running variations of the same pilot rather than building on it.
More tools, same problems
Zilvinas Girenas, head of product at nexos.ai — a Lithuanian-founded AI platform backed by Index Ventures — describes what fragmented adoption looks like in practice. “In larger companies, there is typically a dedicated team handling operations, analytics, or martech to support the adoption of AI,” he says. “In SMBs, that responsibility often falls on one marketing lead’s shoulders. This person is left to piece together separate tools with no clear system behind them.”
The pattern is recognisable to anyone who works with mid-sized marketing teams. Copy drafted in one AI tool, visuals in another, performance data pulled from a third, manually assembled for a leadership update. Each individual step may be faster. The process as a whole has not become more manageable, and in some cases it has become less so, because the number of tools requiring oversight has increased without a corresponding increase in the capacity to oversee them.
For enterprise marketing teams, that problem is at least partially absorbed by dedicated operations and analytics functions. For SMBs, it lands directly on whoever is running the day-to-day. The result is not failed adoption so much as adoption that never quite coheres into something that can be measured, repeated, or improved.
The ROI problem
HubSpot’s research found that 52% of marketers say AI has made their content less effective. For a capability whose primary pitch in the content space is volume and efficiency, that outcome warrants examination rather than dismissal.
The most plausible explanation is not that the tools are failing, but that volume without process produces content that is harder to evaluate and harder to stand behind. When a team is producing more material across more channels using more tools, and no one has a complete picture of what was made or why, quality control becomes an editorial judgment call made under pressure rather than a systematic check. The output reflects that.
Girenas puts the problem on visibility rather than capability. “Teams are producing more content than ever, but when you ask how that actually affected the numbers last quarter, no one can clearly identify the process that made a difference. When each team member uses AI in their own way, through different tools and without a cohesive workflow, it creates confusion.”
There is also a brand dimension that tends to get less attention in coverage focused on productivity metrics. Gartner found that half of US consumers would prefer to do business with brands that do not use generative AI in consumer-facing content. Consumer attitudes toward AI-generated material are still forming, and that preference may not persist as familiarity increases. But it represents a meaningful portion of the market right now — and for brands publishing at volume without a shared review process or clear brand guidelines, the risk of erosion is real and not easily reversed once it takes hold.
Who owns the process?
What nexos.ai is selling, a unified platform that brings model access, agent automation, and governance together, is a direct commercial response to the fragmentation problem. Their interest in that narrative is transparent. But the underlying issue does not depend on a vendor to substantiate it. Fragmented AI tooling across marketing functions is a documented pattern, and the consequences difficulty attributing results, inconsistent brand output, limited accountability follow from fragmentation regardless of which solution a team reaches for.
The more durable problem is ownership. Most marketing teams have someone responsible for the content calendar. Very few have someone responsible for how content moves through the process — which tool produced which asset, where approval sat, what changed between draft and publication, and whether any of it connected to a measurable outcome. In a manual workflow, that gap rarely becomes critical because the pace of output keeps it manageable. In an AI-augmented one, where output volume increases and steps multiply, the absence of process ownership surfaces quickly and compounds.
That points to something the pilot-phase data reflects but does not state directly. The organisations stuck at pilot are not necessarily stuck because the technology disappointed them. Many are stuck because moving beyond a pilot requires organisational decisions about ownership, accountability, and infrastructure investment, which are harder and slower than buying a new tool. Until those decisions get made, the pilot continues, the tools proliferate, and the gap between what marketing teams are doing with AI and what they can demonstrate from it keeps widening.
