StackAdapt built its new report around a 28-point drop. Comfort with AI holds at 78% among marketers when the software acts inside rules a human has set, then falls to 50% when it acts on its own, even after performance has been proven. The Toronto advertising platform published the finding on 19 August 2026 and named the break the AI delegation gap.

That framing puts the constraint on marketers’ willingness to hand over authority. The report’s own figures put it somewhere less flattering to the companies selling the tools. Only 6% of respondents said they act on in-platform AI recommendations almost always, 42% of those who ignore a recommendation said it felt generic or irrelevant to the campaign in front of them, and just 19% said their AI tools are fully integrated into marketing and advertising workflows. Marketers declining to act on advice assembled from partial data are exercising judgement, not dragging their feet.

Where does marketers’ comfort with AI break down?

StackAdapt’s four-level ladder puts the break at the line between bounded and unbounded action. Comfort ran at 90% for AI recommending actions while a human decides, 89% for AI preparing actions for human approval and 78% for AI acting within rules set by humans. It then fell to 50% for AI operating autonomously once performance had been proven. The first three levels sit within nine points of one another. The fourth sits 28 points below the third.

The condition attached to that last question is what gives the number its weight. Respondents were not asked whether they would trust an untested system. They were asked about autonomous operation after performance had been demonstrated, and half still said no.

The evidence is a vendor-commissioned survey fielded by an independent research firm. NewtonX surveyed 500 marketing and advertising professionals at mid-sized and enterprise organisations between 19 May and 8 June 2026, across the United States, Canada, the United Kingdom, Germany, Australia and Singapore, with StackAdapt running a parallel survey of 187 of its own customers. Respondents work in programmatic advertising strategy, execution or platform use, so the findings describe media buying rather than B2B marketing in general, and every figure is self-reported perception rather than behaviour observed in platform logs.

Why do marketers ignore AI recommendations inside the platform?

The recommendations themselves are the weak point, on StackAdapt’s own diagnostic. Across five dimensions of the recommendation experience, covering relevance, timing, clarity, manageability and ease of action, no dimension drew more than 40% positive ratings, according to trade title PPC Land, which reviewed the full report. Explanation and reasoning ranked lowest at 31%.

The stated reasons for ignoring a suggestion follow the same line. Advice that felt generic or irrelevant accounted for 42% of dismissals, misalignment with strategy for 22%, a lack of explanation or transparency for 17%, and a sense that the recommendation was timed to drive spend rather than performance for 12%. That last figure is a verdict on the commercial motive of platform-generated advice, delivered by the buyers receiving it.

What moves a marketer to act is narrow. A clear explanation or rationale was named by 33% and a clear tie to a KPI the team already cares about by 31%. Nothing else came close.

Regional behaviour splits from the global average. Advertisers in EMEA act on platform recommendations often or almost always at 59%, against 47% in North America, yet only 22% of EMEA respondents say they are very confident judging whether AI is making the right campaign decision. With 72 of the 100 EMEA respondents based in the United Kingdom, that reading is substantially a British one.

Behind the relevance complaint sits a plumbing problem. Fragmented data pipelines were cited as a barrier to delegation by 49% of respondents, and CRM or first-party data disconnected from buying platforms by 41%. The 19% integration figure and the 42% relevance figure describe one failure from two ends.

What has AI actually improved so far?

AI has delivered speed rather than measured business return in StackAdapt’s data. Some form of AI-driven performance improvement was reported by 88% of respondents. Broken down, the gains concentrate in workflow: reduced manual optimisation time at 62%, faster campaign setup or launch at 56%, faster optimisation cycles at 53% and better audience targeting at 47%. The pattern matches earlier research The Helm has covered, including McKinsey’s finding that marketers have adopted AI widely while under a tenth of firms capture the value.

Commercial metrics sit far lower. Improved return on ad spend was cited by 27% and lower cost per click by 22%. That gap lands at a point when finance teams are auditing what marketing’s AI spend returns. Nate Elliott, principal analyst for AI at EMARKETER, who is quoted in the report, framed the distinction as “efficiency wins fans, but effectiveness wins funds”.

StackAdapt disclosed a caveat unusual in vendor research. Clients using AI regularly were more likely to report performance improvements than occasional users, and the report states it cannot establish whether deeper use produces better outcomes or whether stronger teams simply adopt more deeply.

Who answers when an AI decision costs money?

StackAdapt’s respondents do not agree, and 17% said there is no clear accountability at all. Collective team responsibility for a poor AI-driven decision was named by 31%, the leader who approved the recommendation by 31%, the campaign manager by 28% and the individual operator by 26%. No option reached a third of the sample.

That fragmentation lines up with where the comfort curve breaks. Autonomous operation transfers a decision without transferring responsibility for it, and the closer a respondent sat to the consequence, the more cautious they were. Strategic decision-makers reported roughly 58% comfort with autonomous AI against roughly 34% among hands-on practitioners, a spread of 24 points.

A quieter finding sits underneath. Automation nobody had explicitly switched on was running for 78% of respondents, and 11% had effectively no visibility into which features were active by default. A stated 50% ceiling on autonomy sitting alongside unreviewed automation is a governance gap rather than a preference.

StackAdapt’s answer is six conditions for expanding AI authority: context, transparency, control, testability, accountability and connected infrastructure. Ryan Nelsen, its chief marketing officer, said the organisations that succeed with AI “won’t be the ones that automate the most”.

The report closes by positioning Ivy Studio, the AI hub StackAdapt launched on 28 July 2026, as the route to those six conditions, according to PPC Land’s analysis. StackAdapt has not published the underlying response tables, and the full report sits behind a registration form.