Ninety percent of marketers surveyed accept artificial intelligence that recommends. Only half accept artificial intelligence that acts alone, even after performance has been demonstrated. The distance between those two numbers is the finding at the center of research published on August 19, 2026.

StackAdapt released The AI Delegation Gap on Wednesday, August 19, 2026, a global study of how advertising teams divide decision-making authority between software and staff. The Toronto company issued the report at 9:48 a.m. Eastern Daylight Time. It rests on two separate data collections: an external survey of 500 marketing and advertising professionals conducted by business-to-business research firm NewtonX between May 19, 2026 and June 8, 2026, and a parallel survey of 187 StackAdapt clients.

The headline finding is not adoption. Adoption is settled. Ninety-one percent of respondents said their organization currently uses AI tools in marketing or advertising, and 86% said they use those tools regularly or for most tasks. Eighty-eight percent reported some form of AI-driven performance improvement. Regional variation is negligible: 91% in North America, 90% in EMEA, 92% in APAC.

What the study measures instead is the point at which willingness stops.

The report maps four levels of AI involvement and asked respondents how comfortable they were with each. The pattern is close to linear until the final step, where it collapses.

Ninety percent agreed with AI recommending actions while humans decide. Eighty-nine percent agreed with AI preparing actions for human approval. Seventy-eight percent agreed with AI taking action within rules set by humans. Fifty percent agreed with AI operating autonomously when performance has been proven.

The first three levels sit within nine percentage points of one another. The fourth drops twenty-eight points below the third. That break is what StackAdapt has named the delegation gap.

"The conversation around AI is shifting from how marketers use it to how much decision-making authority they're willing to give it," said Ryan Nelsen, Chief Marketing Officer at StackAdapt. "That's the AI Delegation Gap. As marketers build confidence in AI, they're gaining a clearer understanding of where it creates value and where human judgment remains essential."

The framing matters because the autonomy question was posed with a condition attached. Respondents were not asked whether they would trust an untested system. They were asked about autonomous operation once performance is proven. Half still declined. That result suggests the reservation is not primarily about accuracy. It is about who answers for the outcome.

Comfort in principle and behaviour in practice are different measurements, and the report separates them.

Only 6% of marketers said they act on in-platform AI recommendations almost always. The report describes this as a say-versus-do gap: advertisers are open to AI recommendations in the abstract, and selective when the recommendation arrives inside a live campaign.

The stated reasons are concentrated. Forty-two percent said they ignore recommendations because the suggestion feels generic or irrelevant to the campaign in front of them. Twenty-two percent said it does not align with strategy. Seventeen percent cited a lack of explanation or transparency. Twelve percent said the recommendation feels timed to drive spend rather than performance. Six percent said there are simply too many recommendations to evaluate, and 1% selected other.

That fourth category deserves attention. Twelve percent of respondents at mid-sized and enterprise organizations reported reading in-platform AI suggestions as commercially motivated rather than performance motivated. The figure is not large, but it is a direct statement about the credibility of platform-generated guidance from the buyers who receive it.

On the other side of the ledger, the report asked what makes advertisers act. Thirty-three percent named a clear explanation or rationale. Thirty-one percent named a clear tie to a key performance indicator they care about. No other driver came close.

The broader diagnostic is harsher. Across five dimensions of the recommendation experience - relevance, timing, clarity, manageability, and ease of action - no dimension received more than 40% positive ratings. Explanation and reasoning ranked lowest of all, at 31%. The single largest shortfall advertisers identified was knowing when to trust AI over their own judgment.

Nate Elliott, Principal Analyst for AI at EMARKETER, is quoted in the report on what separates a useful recommendation from noise. "Advertisers don't care if a suggestion is created by a human or an AI, they just want good advice that helps them succeed," Elliott said. "A recommendation that's based on a brand's own data and that shows an understanding of their campaign history has a better chance of being useful."

Where AI is actually used tracks the trust curve closely. Reporting and summaries lead at 77%. Performance analysis and insights follow at 74%. Creative development sits at 69% and research and audience discovery at 68%. Then the numbers fall: optimization and in-flight adjustments at 51%, campaign setup at 40%, budgeting and bidding at 40%.

The split is clean. Interpretive and generative work has high adoption. Work that moves money has roughly half that adoption. Fifty-three percent said they were comfortable allowing automation to operate in reporting and analysis, the highest comfort level recorded for any workflow.

EMEA leans harder into this pattern than the global average, with 80% using AI for reporting and summaries and 74% for performance analysis and insights. Among StackAdapt's own clients, 88% expect AI involvement in reporting and insights to increase, which the report identifies as the strongest future-growth area in its client data.

The report argues that reporting is shifting from a retrospective deliverable into what it calls a decision layer. The evidence for that claim comes from what advertisers say they want next. Forty-one percent want AI to flag risks before they escalate. Forty-one percent want AI to summarize insights. Thirty-nine percent want recommendations tied to campaign goals. Thirty-five percent want AI to explain clearly why performance changed.

No single capability dominates that list, which the report reads as evidence that advertisers are asking for a bundle rather than a feature.

There is a counterweight in the same section. Thirty-four percent of respondents said AI recommendations do not reflect their campaign goals or context, and 37% cited limited ability to explain why a recommendation was made. An analysis layer that cannot state its own reasoning is a weak foundation for a decision layer, and the report acknowledges as much.

Elliott framed reporting as the logical first deployment. "Reporting is a natural starting point for AI because it combines high value with relatively low risk," he said. "And when AI-generated insights miss the mark, as they inevitably will sometimes, marketers can validate them against other data sources before acting."

One of the sharper cuts in the report divides respondents by role rather than region.

Strategic decision-makers reported roughly 78% comfort with AI acting within set rules and roughly 58% comfort with autonomous AI. Hybrid operators, defined as respondents who both contribute to strategy and manage or execute campaigns, reported roughly 81% and roughly 47%. Hands-on practitioners reported roughly 59% and roughly 34%.

The autonomy spread between strategic decision-makers and hands-on practitioners is twenty-four percentage points. The rules-based spread is nineteen points. Practitioners are consistently the most cautious group, and the caution widens as authority increases.

The report is careful about what does not differ. Across role groups, respondents act on platform recommendations at similar rates. The divergence appears when a recommendation becomes an action. Practitioners place more weight on limited-budget testing, brand safety constraints, and the ability to pause or reverse AI-driven changes.

The interpretation offered is that practitioners sit closer to the operational consequence. A budget shift that underperforms lands on the person managing the account before it lands on the person who approved the strategy. That structural asymmetry, not a difference in enthusiasm for the technology, is what the numbers appear to describe.

This has a practical implication for anyone building an internal automation roadmap. The 78% figure for rules-based action is a blended average across role groups. Inside a team where hands-on practitioners hold operational veto, the applicable figure is closer to 59%.

A short section of the report addresses something rarely quantified: how much automation is actively chosen versus inherited from platform defaults.

Twenty-two percent of respondents said they intentionally chose most platform automation features. Thirty-eight percent said most were deliberate, with some defaults accepted without review. Twenty-nine percent said active automation was about half deliberate and half defaults. Eleven percent said most active features were platform defaults, or that they were unclear which features were on by default.

Combining the last three categories, 78% of respondents reported at least some automation running that nobody explicitly turned on. Eleven percent reported effectively no visibility into the configuration at all.

The report frames this as delegation occurring without a delegation decision. When automation is accepted without review, teams have less visibility into what context the system is using, what assumptions it is making, and how its outputs affect campaign results. That mechanism sits uncomfortably alongside the 50% ceiling on autonomous AI: a stated limit on autonomy coexisting with unreviewed automation is a governance gap rather than a preference.

Barriers cluster around consequence rather than capability.

Brand risk was cited by 63% of respondents and was both the most widely selected hesitation factor and the top first-ranked concern. Data quality concerns followed at 56%. Lack of transparency and performance volatility tied at 36%. Poor integration between systems came in at 34%.

The unlock side is flatter. Confidence level or predicted impact before applying a change was named by 37%. Ability to test on a limited budget before scaling reached 36%. Clear explanation of why a change was made reached 35%. Full audit trail and change log and easy undo or revert tied at 31%.

No unlock exceeds 37%, which means no single control resolves the hesitation for a majority. The report reads this as advertisers requiring a bundle of trust builders rather than one feature. StackAdapt clients produced a similar ordering when asked about general hesitation toward delegating more control, with brand risk, data quality, and lack of transparency again ranking highest.

Elliott's comment on this point is the most pointed in the document. "Trust, not technology, will be the main obstacle to advertisers' AI adoption," he said. "The minute automation scales past the point where someone can answer for it, advertisers will pull back."

The report separates brand-side advertisers from agency professionals and finds the same delegation gap expressed through different pressures.

Agencies are more likely than brands to act on AI recommendations often or almost always, at 56% against 47%. Agencies are also more likely to report that leadership expectations run ahead of readiness, at 63% against 55%.

Those two findings together describe an agency segment that is both faster to act and more aware of the gap between mandate and capability. The report attributes the difference to structure. Agency teams manage AI across multiple clients, campaigns, and platforms, which makes consistency and workflow fit critical, and shared accountability across client approvals adds friction that brand teams do not face in the same form. Brand-side teams more often own the final business outcome, which pushes the requirement toward connecting recommendations to internal justification.

The performance data contains the report's most significant internal tension, and the document does not hide it.

Eighty-eight percent of respondents reported some form of AI-driven performance improvement. When that figure is decomposed, the gains concentrate in workflow rather than in media outcomes.

Reduced manual optimization time was cited by 62%. Faster campaign setup or launch reached 56%. Faster optimization cycles reached 53%. Better audience targeting reached 47%.

Direct business metrics sit far lower. Twenty-seven percent cited improved return on ad spend. Twenty-two percent cited lower cost per click.

The distance between 62% and 27% is the substance of the finding. AI is measurably removing manual work from campaign operations. It has not yet demonstrated, at comparable scale, that removing that work produces better media outcomes.

Client data carried a caveat the report states plainly. StackAdapt clients who use AI as a regular part of their workflow were more likely to report performance improvements than clients who use it occasionally. The report notes that the data cannot establish whether deeper AI usage creates better outcomes, or whether higher-performing teams are simply more likely to adopt AI deeply. That is a correlation disclosed as a correlation, which is unusual in vendor research and worth recording.

Elliott put the measurement problem in commercial terms. "Efficiency wins fans, but effectiveness wins funds," he said, arguing that speed and hours saved are the easiest metrics to move and the least conclusive.

The pattern is consistent with earlier industry data. Research from TransUnion covered by PPC Land found that only 53% of marketers report meaningful return on investment from AI, with data and process readiness rated high by 36% of respondents. Typeface research released in June 2026 went further, finding that enterprise campaign timelines grew longer rather than shorter despite rising AI adoption.

Seventy-nine percent of respondents said they feel strong or moderate pressure to increase AI usage. Fifty-nine percent said leadership expectations are ambitious but may be ahead of current readiness.

The readiness numbers explain why. Nineteen percent said their AI tools are fully integrated into marketing and advertising workflows. Forty-nine percent cited fragmented data pipelines as a barrier to AI delegation. Forty-one percent cited customer relationship management or first-party data that is not integrated with buying platforms. Eighty-three percent said varying AI capabilities across tools disrupt efficiency at least somewhat.

That 19% integration figure is the constraint underneath every other number in the report. A recommendation engine operating on partial signal produces recommendations that feel generic, which returns the analysis to the 42% who ignore suggestions for exactly that reason. The relevance complaint and the integration complaint are the same problem observed from two ends.

Pressure sources were also measured. Executive and C-suite leadership was rated an extreme source of pressure by 35% of respondents. Industry trends and fear of falling behind competitors followed at 28%.

The readiness picture matches what other surveys have recorded through 2026. Mediaocean's H2 2026 Market Report, drawn from 312 marketing professionals surveyed in May 2026, found AI media the fastest-growing investment category at 60% while orchestration held at 86% importance against roughly one in ten organizations achieving it. Brandwatch research covering 1,028 marketers found that AI tools are embedded in daily workflows without resolving the underlying audience understanding deficit.

The accountability data is the most fragmented set in the report, and multiple responses were permitted, so percentages exceed 100%.

Thirty-one percent said the team collectively would be accountable for a poor AI-driven decision. Thirty-one percent named leadership or the decision-maker who approved the recommendation. Twenty-eight percent named the manager overseeing the campaign. Twenty-six percent named the individual operator or campaign owner. Seventeen percent said there is no clear accountability at all.

No option reaches a third of respondents. In practical terms, the industry has not converged on who answers when an automated decision goes wrong.

Brands and agencies split here as well. Brands are more likely to hold the individual operator or campaign owner responsible. Agencies are more likely to distribute accountability across team, manager, and platform or vendor, which reflects the more complex mix of clients and tools in agency workflows.

That 17% figure - respondents reporting no clear accountability - is the number most directly connected to the 50% autonomy ceiling. Autonomous operation transfers a decision. It does not transfer responsibility, and roughly one in six organizations surveyed has not assigned that responsibility to anyone.

Nelsen addressed the point directly. "Organizations that succeed with AI won't be the ones that automate the most," he said. "They'll be the ones that define clear decision boundaries, governance and accountability as AI takes on greater authority across advertising workflows."

Adoption is between 90% and 92% across every region surveyed. Behaviour is not.

North American respondents were the least likely to act on platform recommendations often or almost always, at 47%. EMEA reached 59% and APAC 60%. Broken into components, NORAM recorded 6% acting almost always and 41% acting often. EMEA recorded 10% and 49%. APAC recorded 4% and 56%.

The APAC composition is worth noting. It has the highest combined action rate but the lowest almost always rate of the three regions, meaning its receptiveness is broad rather than deep.

The report characterizes NORAM as trusting through control, with the path to delegation running through approval workflows, audit trails, reversibility, and testing before scaling. EMEA is characterized as trusting through strategic relevance, receptive to recommendations that connect to campaign strategy and named KPIs. APAC is characterized as trusting through readiness, with the highest usage and the highest pressure.

EMEA's openness has a limit the report states explicitly: only 22% of EMEA respondents say they are very confident evaluating whether AI is making the right campaign decisions. Eighty percent feel moderate or strong pressure to increase AI usage and 53% say leadership expectations may be ahead of readiness.

APAC shows the widest gap between appetite and infrastructure. Eighty-one percent feel at least moderate pressure to increase AI usage and 67% say leadership expectations may be ahead of readiness, the highest figure recorded for that measure in any region.

Liam McCarten, VP of Sales for APAC at StackAdapt, described the regional dynamic in the report. "In APAC, we have a clear split: AI adoption is racing ahead, but our operational readiness is lagging behind," McCarten said. "The issue isn't a lack of appetite for AI - it's having the right infrastructure to back it up."

Alongside its data, the report embeds a sequence it labels the path to delegated AI, and each stage attaches to a different section of the findings. Relevance comes before trust. Insight comes before action. Rules come before autonomy. Risk containment comes before delegation. Proof comes before expansion. Readiness comes before confidence.

Read against the survey numbers, that sequence is a reordering of the barrier data into an implementation order. Relevance sits first because 42% of recommendations are dismissed for feeling generic. Rules precede autonomy because the drop from 78% to 50% occurs exactly at the boundary between bounded and unbounded action. Risk containment precedes delegation because brand risk at 63% is the single most cited hesitation. Readiness sits last because the 19% integration figure conditions everything above it.

Elliott's distinction on that final stage is the observation least dependent on vendor framing. "The difference between AI experimentation and AI transformation isn't how much AI a company uses, it's how seamlessly the technology fits into the ways people do their work," he said. Isolated productivity gains, on that reading, are the signature of experimentation rather than of a changed operating model - which is a reasonable description of a market where 62% report reduced manual optimization time and 27% report improved return on ad spend.

StackAdapt closes with a framework of six elements it identifies as prerequisites for expanding AI authority: context, transparency, control, testability, accountability, and connected infrastructure.

Context covers whether recommendations reflect the campaign, audience, KPI, brand nuance, business priority, and regional market. Transparency covers whether teams can understand why a recommendation was made, what data informed it, and what outcome is expected. Control covers approval paths, budget caps, brand safety constraints, escalation triggers, and override rights. Testability covers the ability to trial recommendations at limited scale before expanding automation. Accountability covers clarity on who owns decisions, outcomes, and failures. Connected infrastructure covers integrated tools, connected data, reliable attribution, and clear ownership across media, data, and analytics teams.

Five of the six map onto barriers or unlocks already measured in the survey. The sixth, connected infrastructure, maps onto the 19% integration figure. As a framework it is coherent, and it is also a description of a product roadmap: the report concludes by positioning Ivy Studio, the AI-first hub StackAdapt introduced on July 28, 2026, as the vehicle for delivering those conditions, and closes with an invitation to book a meeting.

The industry survey was fielded by NewtonX between May 19, 2026 and June 8, 2026 among 500 marketing and advertising professionals at mid-sized and enterprise organizations, spanning both senior leaders and hands-on practitioners involved in programmatic advertising strategy, execution, or platform use.

The sample covers six markets: the United States with 258 respondents, Canada with 42, the United Kingdom with 72, Germany with 28, Australia with 84, and Singapore with 16.

That distribution has consequences for how the regional cuts read. North America accounts for 300 of 500 respondents, or 60% of the sample. EMEA totals 100 respondents, of which 72% are British. APAC totals 100 respondents, of which 84% are Australian. The report acknowledges the last point in a footnote, stating that APAC results are based on an Australia-heavy sample with Singapore also represented.

Regional findings presented as EMEA are therefore substantially UK findings, and findings presented as APAC are substantially Australian findings. The German cell at 28 respondents and the Singaporean cell at 16 are small enough that percentages derived from them alone would carry wide margins.

The client survey of 187 StackAdapt customers is a separate instrument. The report states that where internal and external results are compared, it treats them as directional comparisons because question wording, response formats, and sample composition may differ. That disclosure is more explicit than most vendor research provides.

One further qualification applies throughout. The research was commissioned and published by a company that sells an AI advertising and orchestration platform, and the findings are self-reported by respondents rather than observed in platform logs. Figures such as the 88% reporting AI-driven performance improvement describe perception, not audited outcomes.

StackAdapt is not a neutral observer of the question it surveyed. The company sells a demand-side platform spanning CTV, digital out-of-home, display, native, audio and email, and it has spent the past thirteen months building the product category the report identifies as unmet demand.

Ivy arrived as an in-platform assistant in July 2025. Ivy Studio followed on July 28, 2026 as an agent-driven hub covering planning, forecasting, analysis, optimization and execution, with the company disclosing more than 15,000 AI-assisted internal workflows a week. StackAdapt joined OpenAI's ChatGPT advertising pilot as one of five technology partners on May 5, 2026, and introduced in-platform economic impact measurement on July 30, 2026.

A vendor documenting a ceiling on autonomous AI while selling a co-pilot model that stops short of autonomy is describing a market it is positioned to serve. That does not invalidate the survey, which was fielded by an independent research firm across a sample the company did not select. It does mean the ordering of the six conditions and the emphasis placed on connected infrastructure carry a commercial interest that the underlying percentages do not.

The practical value of the report is not the delegation gap as a concept. Vendors have been describing a trust deficit around autonomous advertising for eighteen months. The value is the specific location of the break in the authority ladder, because that location is where product requirements and procurement questions actually live.

Between 89% comfort with prepare-for-approval and 78% comfort with act-within-rules, the requirement is a defined rule set and an approval path. Between 78% and 50%, the requirement is accountability for outcomes, and the survey shows that roughly one in six organizations has not established it.

That maps onto a pattern PPC Land has tracked across the year. IAB research published in July 2026 found CTV buyers evenly split on whether agentic AI optimises the auction or replaces it, with visibility and explainability ranked above the guardrails and audit trails vendors have marketed. TripleLift research from May 2026 found 73% of advertising professionals using AI for optimisation against 25% for creative production. Taboola's advertiser platform reported in May 2026 that 76% of senior performance marketers saw meaningful gains from agentic tools, concentrated almost entirely inside search and social. Earlier this month, Fluency disclosed that it restricts AI agents from executing against live budgets across roughly 3 billion dollars in managed spend, citing Deloitte research finding that only one in five companies has a mature governance model for autonomous agents.

Four separate instruments, four different sponsors, one consistent shape. Advertisers use AI for analysis and preparation at high rates. They restrict it at the point of execution. The restriction tracks accountability structures rather than model quality.

The supply side has been building for the autonomous case regardless. IAB Tech Lab published an agentic roadmap extending OpenRTB, AdCOM, and VAST in January 2026 and formally named the initiative AAMP on February 26, 2026. Google has moved advisory AI into default homepage placement across Ads and Analytics, adding prompt-generated dashboards and benchmarking against anonymized peer averages. The StackAdapt data suggests the constraint on adoption of those systems will not be their capability. It will be whether the buyer can name who is answerable when a delegated decision costs money.

For media buyers evaluating agentic products, the survey supplies a usable checklist drawn from what respondents said would change their behaviour: predicted impact before a change is applied, limited-budget testing before scaling, a stated rationale for each recommendation, an audit trail, and a revert path. None of those five exceeded 37% individually. All five together describe the conditions under which the other half of the market might move.

• July 9, 2025 - StackAdapt launches Ivy, an in-platform AI assistant, reporting more than 1,700 messages during 30 days of pre-launch testing
• December 29, 2025 - IAB Tech Lab chief executive Anthony Katsur warns the industry is chasing agentic AI before resolving transparency and measurement fundamentals
• January 6, 2026 - IAB Tech Lab publishes an agentic roadmap extending OpenRTB, AdCOM and VAST
• January 28, 2026 - IAB forecasts 9.5% US ad spend growth for 2026 with two-thirds of advertisers focused on agentic systems
• February 26, 2026 - IAB Tech Lab formally names its agentic initiative AAMP
• March 16, 2026 - Brandwatch publishes research on 1,028 marketers showing AI embedded in workflows without resolving audience understanding
• April 20, 2026 - Basis reports that 87.3% of agency professionals believe the traditional agency model is broken or will be
• May 5, 2026 - StackAdapt joins OpenAI's ChatGPT advertising pilot as a technology partner
• May 19, 2026 - NewtonX begins fielding the industry survey underlying The AI Delegation Gap
• May 19, 2026 - TripleLift publishes research finding 73% AI use in optimisation against 25% in creative production
• June 8, 2026 - NewtonX fieldwork closes with 500 completed responses across six markets
• June 17, 2026 - Mediaocean's H2 2026 Market Report records AI media as the fastest-growing investment category at 60%
• June 18, 2026 - IAB Australia publishes its report on agentic AI as an execution engine rather than a response engine
• June 22, 2026 - Typeface research finds enterprise campaign timelines lengthening despite AI adoption
• July 7, 2026 - DigiCert research finds 78% of organizations have experienced an AI-related security incident or vulnerability
• July 14, 2026 - IAB research finds CTV buyers split on whether agentic AI optimises or replaces the auction
• July 28, 2026 - StackAdapt introduces Ivy Studio, disclosing more than 15,000 weekly internal AI-assisted workflows
• July 30, 2026 - StackAdapt launches its Economic Impact Report measurement tool
• August 2026 - Fluency discloses that it blocks AI agents from executing against live budgets across roughly 3 billion dollars in managed spend
• August 19, 2026 - StackAdapt publishes The AI Delegation Gap at 9:48 a.m. Eastern Daylight Time

• StackAdapt logs 15,000 weekly workflows as Ivy Studio cuts hours to minutes - Details the July 28, 2026 launch of the AI-first hub the delegation report positions as its answer to the six conditions.
• StackAdapt launches AI assistant Ivy for programmatic advertising - Covers the July 2025 debut of the assistant underneath Ivy Studio, including early message volume.
• Most advertisers use AI for optimization but not creative, TripleLift finds - Independent survey evidence of the same adoption split between analysis and execution.
• Only 53% of marketers get meaningful ROI from AI, TransUnion finds - Quantifies the readiness gap between AI investment and measured return.
• IAB: 43% of CTV buyers doubt where their ads actually ran - Buyer-side data ranking visibility and explainability above guardrails and audit trails.
• Fluency blocks AI from touching live ad spend across $3 billion in budgets - A vendor implementing the recommendation-not-execution boundary the survey describes.
• AI media leads H2 ad investment plans as implementation gap widens - Mediaocean data on orchestration importance against achieved orchestration.
• Most advertisers see AI gains only in walled gardens, Taboola study finds - Where agentic performance gains have and have not materialised.
• AI made campaigns slower, not faster, new Typeface data shows - Counter-evidence on whether AI efficiency claims translate into shipped work.
• How IAB Tech Lab plans to prevent agentic AI's fragmentation problem - The standards work being built for the autonomous scenario half of respondents decline.
• IAB Australia maps the agentic AI shift search marketers can't ignore - Framework treating trust rather than capability as the constraint on agentic adoption.
• DigiCert: 78% of firms face AI incidents, half lack governance - Security-side evidence of the governance deficit behind delegation hesitancy.

Who: StackAdapt, a Toronto-based AI advertising and orchestration platform, published the research. Ryan Nelsen, Chief Marketing Officer at StackAdapt, and Liam McCarten, VP of Sales for APAC at StackAdapt, provided company commentary. Nate Elliott, Principal Analyst for AI at EMARKETER, contributed external analysis. NewtonX conducted the industry survey.

What: The AI Delegation Gap, a report finding that comfort with AI falls from 90% when it recommends actions to 50% when it operates autonomously with proven performance, that only 6% of marketers act on in-platform AI recommendations almost always, that 42% ignore recommendations for being generic or irrelevant, that 19% have AI tools fully integrated into workflows, and that 17% report no clear accountability for poor AI-driven decisions.

When: Published on Wednesday, August 19, 2026 at 9:48 a.m. Eastern Daylight Time. The underlying industry survey was fielded between May 19, 2026 and June 8, 2026.

Where: The external sample covers six markets - the United States, Canada, the United Kingdom, Germany, Australia, and Singapore - with regional cuts reported for NORAM, EMEA, and APAC. A separate global client survey covered 187 StackAdapt customers.

Why: Advertising platforms across the sector are shipping agentic products faster than buyers are authorising them to act. The survey locates the break precisely: not at capability, and not at recommendation, but at the transfer of responsibility for outcomes. That places accountability structures, audit trails, reversibility, and data integration ahead of model performance as the determinants of how far automation spreads through campaign workflows.