Last week, we attended the Market Research Society’s ‘AI: Powering Up Insights 2025’ conference, highlighting how generative AI is shaping the sector and where the next year may take us.
In line with our own AI exploration at Obsurvant, the technology is undoubtedly offering new capabilities, unlocking new levels of efficiency, creativity, and analysis. However a consistent theme at the conference was that AI capabilities are still very much developing and that it’s not yet a complete solution. This is particularly evident in delivering consistent and accurate synthetic results without extensive training and validation. Something that has hindered our AI implementation within many of our projects.
Here are some of our key takeaways from the conference:
The Potential of Generative AI
Several sessions explored how Gen AI is enhancing market research through innovative applications. For instance, Dentsu showcased its use of AI in creating synthetic personas for workshops, concept testing, and even avatars representing dataset segments in client calls. While promising, they highlighted some limitations, including struggles with brand-related questions, scale responses, and complex survey routing. In fact, their testing revealed significant discrepancies between AI and real-world datasets.
Ipsos also shared its six-month journey of training AI to replicate real respondents in conjoint and MaxDiff analysis. Despite testing over 70 models, including GPT-4, Claude 2, and Gemini Pro, results were underwhelming. Even with model training, responses lacked the natural variation found in human data, reinforcing that synthetic respondents still have a long way to go.
The Importance of Prompts
A recurring theme throughout the conference was the pivotal role of prompts in generating meaningful AI outputs. Companies like Glaut have developed extensive libraries of prompts for analyzing open-ended responses. Similarly, Ipsos demonstrated how adjusting “temperature” settings in AI systems can influence response variability. This is something that can drastically improve accurate reporting on sentiment analysis within open end survey response.
Qualitative Research: A New Frontier
AI is proving particularly effective in qualitative research. Several speakers highlighted the depth and richness of responses gathered through AI-powered chatbots, which consistently produced higher word counts compared to traditional written open-ends. This approach also offers cost and efficiency advantages, though concerns were raised about the lack of a “human touch” and of “bots talking to bots” as the researcher becomes automated and the bots entering surveys become more sophisticated.
Emerging Tools and Applications
Beyond research methodologies, some sessions focused on new tools and platforms. Inspirient, for example, demonstrated its AI-driven reporting and dashboarding platform, which offers customizable data outputs for consumer research. Meanwhile, Viasat shared how it built an internal AI-powered search engine for employees to access insights across market research, sales data, and news. These examples illustrate the diverse ways AI is being integrated into business processes.
Challenges and Limitations
Despite these advancements, the conference underscored the limitations of current AI capabilities. Issues of bias, both in AI models and the humans creating them, remained a hot topic. Moreover, the backward-looking nature of AI models was critiqued, particularly in concept testing, where predictive accuracy is critical. The notion of “proper” synthetic respondents feels distant, but the ability to use AI for querying datasets and data libraries is within reach.
Looking Ahead
Perhaps the most thought-provoking discussion came from Microsoft and Space Doctors, who explored how AI is influencing global culture and communication. Their session underscored the importance of understanding cultural nuances when deploying AI-driven insights, highlighting the varying levels of adoption and perception across regions.
AI is undeniably transforming market research, offering incredibly exciting possibilities for our insight and analysis. However, as highlighted at the conference, its capabilities are still evolving. For now, human expertise remains essential—not just to validate AI outputs but to ensure that insights are meaningful, accurate, and actionable.
As we continue exploring how AI fits into our own processes, it’s clear that collaboration, creativity, and caution will be key to unlocking its full potential.