Bench Media has published research showing that senior Australian marketers are becoming more confident in AI-assisted decisions even as human review remains inconsistent. The findings cover strategic recommendations, budget choices, creative concepts and calculated risks, placing AI closer to the center of marketing leadership rather than at the edge of production.
The tension is not that marketers lack confidence. It is that confidence is arriving before many organizations have defined who must challenge a recommendation, what evidence should support it and who owns the final call. That makes governance a decision-design problem, not simply a compliance exercise.
Table of contents
Jump to each section:
- Confidence and control are separating
- AI advice gets riskier as decisions get bigger
- Veto literacy needs an operating model
- What marketing leaders should change next
Confidence and control are separating
The Confidence Equation surveyed senior client-side marketing leaders in Australia about the inputs shaping decisions on brands, budgets and growth. The overall picture is optimistic: AI appears to be reducing some of the friction around complex choices rather than adding another layer of hesitation.
90% of respondents said decision-making had become easier over the previous year.
That result sits beside a more complicated signal.
59% named the volume of AI and new technology tools as the biggest factor still making decisions more complex.
The apparent contradiction is useful. More tools can increase the number of options while AI helps leaders compress those options into a recommendation. Ease, in other words, may come from delegation rather than from a simpler market.
Confidence is not proof that uncertainty disappeared. It may be evidence that the uncertainty has been packaged more persuasively.
The research also indicates that AI is shaping how leaders feel about their own judgment.
48% said AI had made them more confident in their decisions.
That is strategically important because confidence changes behavior. A system that merely produces an answer is a tool. A system that makes an executive more willing to commit budget or defend a direction is participating in organizational authority.
AI advice gets riskier as decisions get bigger
The strongest governance concern appears where recommendations become consequential. Bench Media found that some marketers accept AI output without an explicit human challenge, including in areas that affect strategic direction, spend and creative expression.
38% said they accept AI-generated strategic recommendations without human review.45% reported accepting AI-generated budget recommendations and creative concepts without human review.
These findings do not establish that the decisions were poor. They expose a missing distinction between trusting a recommendation and being able to account for it. A useful output can still be based on incomplete context, questionable assumptions or evidence that no one has inspected closely enough.
The common assumption is that the safest organizations will keep a person in every workflow. The contrasting reality is that universal review can become ceremonial, especially when senior teams face more outputs than they can examine properly. The strategic implication is to concentrate human judgment at the decisions where money, brand meaning and accountability can materially change.
The survey’s budget comparison makes that need more urgent.
51% versus 24% of leaders managing budgets above and below $20 million respectively said they lean on AI when considering a calculated risk.
Larger decisions appear to attract more AI reliance, not less. That can be rational if the systems help leaders synthesize more signals than a team could process manually. It also means governance should strengthen with the consequence of the decision, because the cost of a fluent but weak recommendation rises with the size of the commitment.
The larger the bet, the less useful a generic approval step becomes. Leaders need to know which assumption they are approving.
For budget decisions, that could mean testing whether a recommendation is sensitive to attribution choices, margin constraints or an unusually short performance window. For creative concepts, it could mean asking whether the model has recognized cultural context and brand history, rather than merely reproducing patterns that scored well elsewhere.
Veto literacy needs an operating model
Bench Media uses the phrase “veto literacy” for the ability to override AI and defend that choice. The phrase captures an emerging leadership skill, but the skill should not depend on executive instinct alone. An override becomes more credible when the organization has already defined the evidence, thresholds and decision rights that make intervention legitimate.
The research suggests instinct still matters when recommendations conflict.
59% said they would back their own judgment when an AI recommendation disagreed with it.
That should not be read as a contest between humans and machines. The more productive question is what each side contributes. AI can surface patterns, compare scenarios and pressure-test options. Experienced leaders can interpret commercial context, recognize when the framing is incomplete and accept responsibility for a decision that may need to be explained later.
A veto is only governance when it is connected to evidence and authority. Otherwise it is preference with seniority behind it.
This is where fast, targeted primary research can become a practical counterweight to machine confidence. If an AI system recommends a new audience, offer or creative direction, a focused customer test can examine the specific assumption most likely to change the decision. The purpose is not to ask research to certify every output. It is to introduce fresh evidence where the model’s inherited patterns may be least reliable.
Michelle Wee, Hisense’s general manager of marketing, frames the balance around technology plus an understanding of consumers, culture, brands and business objectives. Emma Fogarty, head of marketing and category at nudie Foods, similarly argues that outputs should be challenged, validated and considered in context. Both perspectives place the human role beyond proofreading. The task is to supply meaning the system cannot reliably infer on its own.
The report is directional rather than universal. Its Australian, client-side sample can reveal how a group of senior marketers is thinking, but it does not prove that the same confidence or review habits exist across every market or organization. The value lies in the operating questions the findings expose.
What marketing leaders should change next
Marketing leaders do not need to choose between trusting AI and slowing every decision with extra approval. They need a clearer architecture for deciding when confidence is sufficient and when a recommendation must be challenged.
Separate advice from authority. Teams should know whether an AI system is generating options, ranking them, recommending one or triggering action. Each step carries a different level of consequence and should have a different control.
Define evidence before the meeting. A board or executive team should not discover the limits of an AI recommendation only after someone objects. Decision owners can specify in advance which customer evidence, financial assumptions and market context must accompany a consequential recommendation.
Assign the right challenger. Human review is not automatically valuable because a person performed it. Budget, creative, legal and cultural decisions require different expertise, so the reviewer should match the risk rather than simply hold the most senior title.
Record why the final call changed. When a leader accepts or rejects AI advice, the organization should retain the decisive assumptions and evidence. That record turns isolated judgment into institutional learning and makes future automation more accountable.
The deeper shift is that marketing leadership is becoming partly a discipline of governing recommendations. As AI absorbs more research, synthesis and scenario work, executives will spend less time assembling inputs and more time deciding which inputs deserve authority.
That change raises the standard for both machines and people. AI recommendations must become easier to interrogate. Human overrides must become easier to explain. Confidence without either capability is fragile, however decisive it appears in the room.
The organizations that develop this discipline will not be the ones that reject automation most often. They will be the ones that can move quickly because everyone understands where the machine can advise, where a person must decide and what evidence allows the decision to stand.