Many people think AI will put them out of a job. But we have good news: while it has been helpful in understanding what kind of food consumers want, there is still time before AI eats our lunch. That much has become apparent in our latest research.
AI has been touted as a helpful tool in overcoming the challenges in conducting conventional market research, especially for niche products. Increasingly, marketers are looking at the use of synthetic personas. By generating personas that mimic humans, researchers can explore diverse consumer behaviours, preferences and trends at scale, without compromising sensitive information or facing ethical constraints.
Recently we studied views on a niche food product – for reasons of confidentiality, let’s call it tacos – among customers of a major British retailer, and compared it to synthetic personas. To build realistic synthetic personas, we tapped our library of consumer segments, which have endearing names like ‘curious foodie’ and ‘social statement maker’. Overall, the results were promising and helpful – but not perfect.
The responses from the synthetic personas were closest to the real-world data on factual questions, and questions that offered a list of options. For example, 44% of humans said they were generally interested in tacos versus 42% in the synthetic sample.
However, the synthetic personas had a tendency to expand on the listed responses and give more opinion than requested. This synthetic extemporising was reflecting the attributes of the consumer segments we had built into the system – perhaps trying to show it understood the roles it was told to play. Our team resolved this issue as soon as they finished chuckling at responses such as: “[I am] very interested in tacos. Tacos are a trendsetter, visually appealing, perfect for the ’gram. Plus, it’s delicious.”
On more complex and emotion-based questions, the synthetic sample had an admirable stab at displaying imagination. When we asked curious foodie about tacos, the reply was: “Love it! Always up for trying different tacos. The mix of flavours, it’s a whole new experience each time. Can’t resist a good taco!” Who can, curious foodie, who can?
While we gave our synthetic personas an A for effort, there is room for improvement, especially with open-ended questions. The synthetic sample sometimes emphasised things that did not reflect the real-world data.
When asked how much price affects the decision to buy tacos, it mirrored the human answers in stating that quality and freshness were important, but it tended to place a lot more significance on authenticity.
As machine learning and AI improve, so will the quality and realism of synthetic personas. Eventually it will be a powerful tool when considering new products or entry into niche markets and, with more granular data, businesses can develop highly personalised marketing strategies, building customer engagement and driving better business outcomes.
But it will take time before that’s a reality. Luckily, if the machines do rise up, we will at least have established a healthy working relationship.
No comments yet