Generative AI is undoubtedly one of the top 5 terms most frequently used in discussions between market research players. But what does it actually mean? What are its benefits, particularly in terms of listening to consumers on social networks? And what are the possible avenues of development? These are the questions we put to Mathilde Guinaudeau, head of Social Intelligence & Analytics Expertise at Ipsos, and François Thiebaut, expert in the Ipsos Synthesio solution, which has just integrated this brick into its platform.
MRNews: How would you define « generative » AI in a few words?
Mathilde Guinaudeau (Ipsos): This is an artificial intelligence capable of generating text from data corpora based on language models (LLM). It can generate not only textual elements, but also images, sounds and videos. Of course, the generated elements could be elaborated « by hand », by humans, but that would take years! The strength of generative AI is that it can do this in a matter of seconds, even if the corpus is gigantic. The integration of generative AI therefore represents a decisive leap forward for market research.
François Thiebaut (Ipsos Synthesio): Applied to social listening, and therefore to unstructured data, generative AI is a radical game-changer, thanks to its ability to compile information simply and efficiently in record time.
Ipsos recently announced the integration of generative AI into its Synthesio social listening platform. How is it used?
FT: You have to think of the Ipsos Synthesio platform as a kind of super « vacuum cleaner » for social media data, found in particular on social networks. All in all, there are over 600 million online sources whose data we vacuum up and then process using « classic » artificial intelligence. The difficulty lies in navigating such an immense ocean, and detecting what we call the « Golden Nuggets », the most enlightening data to respond to a precise business challenge within the framework of a study. It is on this platform that we have deployed Generative AI, which considerably enhances the power of the tool for our users. In concrete terms, we have integrated a library of data observation angles. Facially, these are « buttons », behind which instructions, prompts, have been developed. Rather than querying ChatGPT directly, and asking 150 questions to produce a potentially satisfactory but uncontrolled output, users can now make hyper-efficient use of a database created and sourced by Ipsos.
Rather than querying ChatGPT directly, and asking 150 questions to produce a potentially satisfactory but uncontrolled output, users can now make hyper-efficient use of a database created and sourced by Ipsos.François Thiebaut
One of the key parameters in the relevance of AI responses is the quality of the prompts. Developing them is becoming a skill and a profession in its own right. Is that really what you’ve integrated?
MG: Absolutely! Once users have defined the theme of their choice – a particular product category or brand – simply clicking on a button enables them to access an analysis based on Ipsos proprietary frameworks, written by our experts, with every guarantee of effective insight. They can both capitalize on the expertise of our specialists – writing relevant prompts is part of their job – and exploit a sourced, mastered database, thus surpassing the usual limits of ChatGPT.
Once users have defined the theme of their choice – a particular product category or brand – simply clicking a button takes them to an analysis based on Ipsos’ proprietary frameworks, written by our experts, with every guarantee of effective insight.Mathilde Guinaudeau
More specifically, what kind of lighting do these « buttons » activate?
FT: We have defined around 30 buttons, grouped around 5 major verticals. These include market understanding, brand understanding, innovation and trends. We also use Ipsos’ own frameworks, such as the Censydiam motivational grid, or a prism designed to understand CSR issues. In concrete terms, for example, it’s extremely easy to identify a brand’s distinctive features in relation to its competitors, in just a few seconds.
Behind these buttons, the « prompts » aren’t necessarily very long. But they are precise, designed to be highly efficient within a marketing research logic. When it comes to trends, for example, we at Ipsos are used to distinguishing between nano-trends, macro-trends and micro-trends. The buttons we have defined incorporate this level of subtlety.
How far can these tools take marketing research? How do they fit in with human intelligence?
MG: Generative AI, as we have been able to implement it, provides invaluable assistance in drawing up marketing recommendations. In particular, it enables us to synthesize unaddressed needs and consumer expectations in a given market. The idea is to go even further, by asking Artificial Intelligence to propose new product ideas to meet these expectations, specifying the key benefits to be highlighted.
But – and we feel it’s really important to stress this – the outputs of artificial intelligence remain materials to nourish and inspire our thinking. They are aids. When Ipsos experts use them, it is inconceivable that they would simply copy and paste these elements into reports without adding their own value. It’s our job to make sure that these materials make sense in relation to our clients’ challenges, and to associate our thoughts with these insights.
Generative AI, as we have been able to implement it, provides invaluable assistance in drawing up marketing recommendations (…). But – and we feel it’s really important to stress this – the outputs of artificial intelligence remain materials to nourish and inspire our thinking. They are aids.Mathilde Guinaudeau
The launch is very recent. What is the initial feedback?
MG: These features have only been available to our customers since mid-September. They are now being tested on a massive scale, with over 4,000 queries to the platform in the space of a few days. Initial feedback has been extremely positive, particularly in terms of the time savings these tools can deliver. One of our interviewees told us that in a matter of seconds, she was able to obtain an insight that had previously taken her 15 days. But here again, I’d like to make it clear that, while the tool’s contribution is exciting, it’s by no means a magic wand.
Does AI have other uses at Ipsos?
MG: Yes, absolutely. In particular, we carried out a major project to meet our internal needs. We identified all the tasks carried out by our employees worldwide, however diverse, which could be facilitated by generative AI. The first list included 200 tasks, which we gradually reduced to 60. For example, writing e-mails, summarizing a business meeting or a research report, translating a questionnaire… From there, we were able to invest in an internal generative AI solution that we called Ipsos Facto, to which all Ipsos employees have access, whether they have marketing, administrative or production functions… We use this tool on a daily basis, just like Google or office software, it simply enables us to be more efficient.
We (also) carried out a major project to meet our internal needs. We identified all the tasks performed by our employees worldwide, however diverse, which could be facilitated by generative AI (…). And so we were able to invest in an in-house generative AI solution that we called Ipsos Facto (which) is used on a daily basis, and quite simply enables us to be more efficient.Mathilde Guinaudeau
FT: It’s important to note that Ipsos Facto guarantees data confidentiality, since it’s built in a closed, dedicated Ipsos environment. The data is therefore not released into the public domain.
Are these solutions unique in the world of market research?
FT: To my knowledge, the project Mathilde has just described has no equivalent in our world. When it comes to integrating generative AI to meet business needs, we’re all hands on deck! But our great strength lies in the fact that we have already set up a generative AI tool that is accessible to all Ipsos employees (Ipsos Facto), as well as having integrated generative AI into solutions for our customers, such as our Ipsos Synthesio platform. I’d also like to add that our market position has enabled us to build first-rate partnerships with OpenAi and Google. That’s a real big advantage…
Do you foresee any developments in these uses of generative AI?
MG: Today, generative AI is mainly used to search for productivity gains or to synthesize observations on markets or brands, for example. But these models offer much more. I’m thinking in particular of creativity, such as the design of new products or services.
FT: Indeed, our models are currently parameterized to answer precise questions, as Mathilde mentioned, but we can also parameterize them to be more creative and less precise to serve innovation. We could certainly imagine using several models, depending on the degree of creativity required. What’s clear is that the more we want generative AI to provide creative answers, the more human intervention will be needed to make the most of them.
Today, our models are parameterized to answer precise questions, as Mathilde mentioned, but we can also parameterize them to be more creative and less precise to serve innovation. We could even use several models, depending on the degree of creativity required.François Thiebaut
Is there anything else you’d like to add?
MG: Regulation is still uncertain. Take Italy, for example, which has restricted the use of ChatGPT, or Europe, which is in the process of developing its « AI Act ». This is bound to have an impact on the use of generative AI models.
FT: The second parameter is, of course, that this technology is new and constantly evolving. We can imagine the arrival of Chinese models turning the current market upside down. And the answer will have to be agility. That’s why our data scientists are constantly testing new models, such as open source, to make sure we’re on the right wave. So, here again, things can move very quickly!
FOR ACTION
– Talk to the interviewees:@ Mathilde Guinaudeau @ François Thiebaut
