Marketing teams everywhere face a tempting proposition: pair up your best people with the latest artificial intelligence tools, and watch the results soar. The logic seems obvious. Humans bring creativity, context, and ethical judgment; AI brings speed, pattern recognition, and tireless analysis. Put them together, and you should get the best of both.
But does that actually happen? A new analysis published in Marketing Letters suggests the answer is more complicated than the hype implies. While adding AI to a human’s workflow reliably improves performance, the combined human-AI team rarely beats what a capable AI system can do on its own.
The question behind the research
Vu Minh Ngo of the University of Economics Ho Chi Minh City set out to systematically answer two questions that have been nagging marketing researchers and practitioners alike. First: do humans and AI genuinely complement each other in marketing decisions? Second: when human-AI teams do outperform solo agents, by how much, and under what conditions?
These questions matter because the existing evidence has been contradictory. Some studies show that AI recommendations lift human judgment substantially. Others find that people either lean on algorithms too heavily or dismiss them without good reason. The marketing field, which has been rapidly adopting AI for everything from customer segmentation to content creation, has lacked a clear synthesis of when collaboration pays off.
How the analysis was built
Ngo pulled together empirical studies that met a strict set of criteria: each had to directly compare the performance of humans alone, AI alone, and human-AI teams on the same marketing task, with enough statistical detail to compare results across studies. After searching databases including the ACM Digital Library and Web of Science for papers published between January 2020 and September 2024, the final sample included 27 studies yielding 132 separate experiments.
The tasks in these experiments spanned common marketing activities: price forecasting, demand prediction, customer service inquiries, sentiment analysis of reviews, and social listening. For each experiment, Ngo calculated standardized effect sizes comparing the human-AI team’s performance against three different baselines: humans alone, AI alone, and whichever of the two performed better on its own.
To make results more intuitive, the analysis also reported simple percentage differences alongside the formal statistics.
What the numbers showed
The headline findings fall into three parts. When human-AI teams were compared against humans working alone, the collaboration produced a clear improvement: performance rose by roughly 21 percent on average, a statistically significant gain. This is the “AI as helper” result that most people expect.
When the same teams were compared against AI working alone, however, the advantage essentially disappeared. The combined performance was statistically indistinguishable from what the AI could achieve by itself.
And when the team’s performance was benchmarked against the stronger of the two solo performers, the human-AI combination actually came up short by about 11 percent. In other words, teaming a human with an AI often produced worse results than simply letting whichever agent was better handle the task alone.
Ngo interprets this pattern as evidence that, under current conditions, most of the value in human-AI collaboration comes from the AI component. Human input, on average, is adding little extra and sometimes introducing friction.
When collaboration actually works
The average figures mask significant variation, and the more interesting findings emerge from looking at what moderates the results.
Task type mattered a great deal. Structured analytical jobs like demand forecasting and review analysis, where AI systems are already strong, showed limited or even negative synergy when humans were added. Tasks involving ambiguity or judgment, particularly customer service and social listening, produced more meaningful collaborative gains. Ngo interprets this as reflecting the nature of the work itself: when problems require contextual interpretation, human intuition adds something AI can’t easily replicate. When problems are clean and data-rich, human involvement mostly introduces opportunities for error.
AI sophistication was another key factor. Human-AI teams paired with basic, rule-based, or simulated AI systems beat humans alone but lost to fully automated AI. The picture changed when advanced deep-learning systems were involved: in those cases, human-AI teams frequently outperformed even the AI on its own. Ngo suggests that more capable AI gives experienced marketers richer material to work with, rather than outputs they simply need to override or ignore.
User expertise reinforced this pattern. Experienced marketers were better able to calibrate their trust in AI recommendations, knowing when to lean on the algorithm and when to push back. Less experienced users tended either to accept AI suggestions uncritically or to dismiss them without good reason.
A surprising finding about transparency
One result ran against a popular assumption in the AI-design community. Providing users with explanations of how the AI reached its recommendations did not reliably improve collaborative performance. Explanations alone, it seems, are not enough to fix miscalibrated trust or to change how people integrate algorithmic advice into their decisions.
AI confidence indicators, which tell users how certain the system is about a given recommendation, performed somewhat better. When these were included, human-AI teams showed meaningful improvements over AI alone. Ngo interprets this as a sign that confidence scores help users distinguish between reliable and uncertain AI outputs in a way that narrative explanations often do not.
A pattern rooted in who is better at what
Digging deeper, Ngo found that human-AI collaboration produced its biggest wins in experiments where humans were already outperforming AI on the task. In those settings, AI served as a useful second opinion that selectively sharpened human decisions. In roughly two-thirds of the experiments, though, AI outperformed humans on its own, and in those cases adding human input typically dragged performance down.
The author draws on cognitive load theory to explain part of this: when marketing decisions come fast and involve many moving pieces, human mental capacity is already stretched thin, and integrating AI recommendations on top becomes an additional cognitive burden rather than a help.
What this means for marketing practice
The practical takeaways split along task lines. For structured, data-intensive jobs like pricing optimization, customer segmentation, and demand forecasting, Ngo argues that organizations should let AI systems run with minimal human intervention, reserving human roles for quality checks, ethical review, and interpretation of unusual results.
For creative, ambiguous, or context-heavy work, such as campaign development, brand management, and customer engagement, the collaborative model has more to offer. Here, pairing experienced marketers with sophisticated AI tools produces the kind of synergy that advocates of human-AI teamwork often assume is universal.
Training also emerges as a practical priority. Because transparency features alone don’t fix trust calibration, Ngo suggests that organizations need to invest in giving marketers hands-on experience with AI systems, so they develop a realistic sense of when to rely on the machine and when to override it.
Limitations worth noting
The analysis carries some standard caveats. Publication bias is a concern, since studies showing dramatic human-AI synergy may be more likely to see print than those showing null results. The experiments included varied widely in design, task type, and performance metrics, which limits the precision of any single pooled estimate. Ngo points to the heterogeneity itself as a reason for future work using standardized protocols across different marketing contexts.
