AI in market research
Since the launch of ChatGPT 3, the first generative AI models with human-like skills for language and analysis, market researchers have been building AI systems into their research tools. We can offer AI-built questionnaires, AI-simulated data, AI generated reporting and summaries, plus transcripts, translations, coding and aggregation. AI provides lots of useful tools to pre-testing ideas and concepts, but reality remains the ultimate test for success.
AI, or LLM (large language models), is revolutionizing the practice of market research, offering the promise of research findings without actually doing any primary data collection. Simply ask the AI system to play a role (personas), or to behave like someone (digital twins), or to create simulated data.
On top of this, it also provides tools for generating questionnaires and content, creating outputs, summarizing finds, and doing translations and transcriptions.
There is no getting away from the fact that AI is transforming the world of research
AI versus Reality
It is fair to say that everyone should be using AI to at least act as a preliminary sense check on new ideas or new content. Even if you disagree with the AI output, it will make you think, often provides valid criticism and makes good suggestions (though often in a similar AI-based style).
The same is true for any market research even before AI. It is worth running ideas past experts and influencers to see what they think.
However, the lesson for large FMCG companies was that even the PhDs in the product development teams could miss things, or have assumptions that weren't quite right. The only way to really test and check is to go to real customers and real potential shoppers, because reality is always a bit more complex than it first appears.
The arrival of AI takes this up a couple of notches. AI has data about what people have said about things in the past, and it can assimilate and randomise this data to create simulated respondents or personas.
It would be mad not to do use AI in this way, given the low costs of this type of simulation. However, it still leaves a nagging doubt. AI, for all its parameters, is still not reality. Real people have a kitchen with cream-coloured walls that they want their kitchen roll to match to. They have a dog that has learnt how to open the fridge door. They have children who won't eat sausages because they saw a documentry on how pork mince was made.
Current experimentation with AI has simulated data reaching about 63% replicability of survey data. Not bad, given surveys with real human beings are only about 85% replicable. But the evidence is pointing to AI being good in the averages, but less good in the 'edge-cases' - the more niche areas, and often the areas marketers are trying to find and understand.
They also struggle with new subjects, like new products, because there is no prior history. And where a market is changing rapidly they can get caught relying on old or out-dated information.
Since most market research is about testing new ideas and innovations, AI will give answers and help steer those ideas. But, when all said and done, reality is the ultimate test.
