AI is very good at giving us something marketing rarely offers: a quick answer with a convincing explanation. Put a business problem into a prompt and, within seconds, you can have a recommendation, a rationale and a list of reasons to back it. When budgets are under scrutiny and everyone wants a decision yesterday, the appeal is obvious.
There is real value in that. AI can help us analyse information, explore options and challenge our thinking. But as AI becomes more involved in marketing decisions, we need to ask whether the confidence it gives us is always matched by the quality of the advice.
Recent Bench research, The Confidence Equation, surveyed 71 senior Australian client-side marketing leaders. Almost half, 48%, said AI had made them more confident in their decisions. That is a positive finding, particularly in a role where leaders are expected to make commercial calls with incomplete information.
Alongside it, however, was a finding that gave me pause: 38 per cent said they accept AI-generated strategic recommendations without human review. For budget recommendations, the figure was 45 per cet. The research did not measure whether those decisions delivered better results which leave us with a question worth asking: what is giving us that confidence, and how thoroughly are we checking it?
A convincing answer can still be the wrong answer
One of the challenges with AI is that the quality of the presentation can disguise weaknesses in the reasoning. A recommendation arrives in clear language, neatly structured and supported by an explanation that sounds sensible, making it easier to accept, particularly when it confirms something we already suspected.
However, as every senior marketer knows, the information missing from a recommendation can matter as much as the information included. A plan might draw on strong historical performance while overlooking a change in pricing, a distribution problem or a competitor’s recent move. It might recommend targeting an audience that converts efficiently without asking whether those people were likely to buy anyway.
None of these problems began with AI. We have been mistaking correlation for causation and building persuasive stories around selective data for years. AI gives us the ability to produce those stories much faster, which makes the discipline of questioning them more valuable.
There is also a difference between using AI to explore a problem and using it to justify a decision. If we ask for a rationale to support a preferred channel, audience or budget allocation, a plausible rationale is exactly what we may receive. The more useful exercise is to ask where the argument is weak, what alternatives deserve consideration and what evidence would change the recommendation.
Of course, that requires a little more effort than accepting the first answer, but it may also be where much of the value lies. A tool that helps us examine a difficult possibility can improve our thinking. A tool we use to make our existing position sound more authoritative, may simply make us harder to persuade.
The same applies to human judgement. The research revealed that 59% of leaders said they trust their own judgement when AI’s recommendation conflicts with theirs. Experience can bring context a system does not have, but it can also bring entrenched preferences. “I’ve been doing this for 20 years” deserves scrutiny too. Whichever recommendation wins, we should be able to explain why.
Confidence must survive contact with results
For media agencies, this puts a sharper focus on what clients are paying us to contribute. A platform’s recommendation may be useful, but we need to understand the objective behind it, the information informing it and the options it can consider. A system can be effective at allocating money within a platform while leaving unanswered the broader question of whether that platform deserves the investment.
If an agency’s role becomes packaging automated recommendations into a presentation, clients will reasonably question its value. An agency’s contribution must include checking assumptions, bringing in evidence the system lacks and considering the commercial context. Agency partners must also be willing to recommend a different course when the available evidence supports it.
Otherwise, there is a risk that competing brands, using similar tools and asking similar questions, become increasingly confident in similar answers. AI can help explore unfamiliar opportunities, but that requires us to look beyond the most readily available recommendation. Competitive advantage still depends on finding something others have overlooked.
The practical response is to make scrutiny part of the process. Before committing to a significant recommendation, we should understand its assumptions, identify the gaps and decide how success will be assessed. Where uncertainty remains, a focused test may be more useful than another round of increasingly polished explanations.
This is also how agencies can turn time saved into something clients value. Faster analysis and reporting should create room to understand customers better, examine a brief more closely or test a different approach. If efficiency simply allows us to process more work, we should be honest about who receives the benefit.
AI can improve marketing decisions, and there are good reasons for the confidence reflected in the research. It provides access to analysis and alternative perspectives that might previously have taken considerable time and resources to assemble. We should make use of that capability while remaining curious about where it falls short.
Feeling more certain is useful when it helps us make a considered decision and act upon it. It becomes a problem when it removes the urge to check. As AI becomes a more familiar presence in marketing, we should judge its contribution by the decisions we can substantiate and the results we can demonstrate.
What matters is whether we’re making decisions we can defend once the results are in, not just when the recommendation is on screen.
Shai Luft is the managing partner of Bench Media.
