Responsible AI adoption in marketing research  

Editor’s note: Finn Raben is the founder of Amplifi Consulting. The following article was first published in Dutch by the Data & Insight Network in Holland and translated to English exclusively for Quirk’s Media. 

Full disclosure: I am a fan of AI, but I am not a fan of how slowly we are addressing the multiple issues associated with it!

Members of the market research/insights/business intelligence profession have always been early adopters of new technologies and developments, demonstrating a keen desire to keep up with commercial evolution and new practices. However, in adopting such practices, our sector (with a few exceptions) has also been quite slow – and sometimes too late – to defend some of the founding principles of our profession. 

An example of this is representative sampling. As an industry we shifted very quickly from probability sampling to quota sampling to online sampling as the internet was – in the early 2000s – deemed to be an essential offering for all businesses. Now we have, with distressing regularity, debates about the quality, representation, churn and recruitment issues associated with online samples. ESOMAR did eventually publish the (well received) 28 Questions, but the horse had already left the stable and, as clearly demonstrated at the 2023 ASC conference held on May 25, the issue of quality has become so widespread that a global initiative has been established to try and minimize the effects … 20 years on!

Data quality and the adoption of AI 

The successful adoption and integration of AI in marketing research is dependent on two crucial elements: curation of training data and the productivity paradox.

  • Curation of training data. There is an assumption that all AI systems are “intelligent,” but they are learning from humans, and humans are not infallible. This is not necessarily always a bad thing (as it teaches the concept of trial and error), but it does mean that the curation of the training data is essential to producing a competent model. 
    • A recent article published by the Research Society Australia provides a very interesting perspective on this. (There is an associated debate about how, where and when learning systems can access training data, such as the recent Zoom case, but that is for another day).
  • The productivity paradox. The second element, referred to as the productivity paradox, shows that the expectations of efficiency increases associated with inward (technological) investment are never fulfilled. This is because the value of new technology is directly and proportionally linked to the business’s ability to invent new structures, processes and procedures to leverage the new capability.

    As a result, implementation lag (and business frustration!) becomes evident, which has led to the development of the sociotechnical school of implementation, which strongly advocates away from technology determinism and the primacy of getting the technology installed, and instead underlines the critical importance of including culture, strategy and people in all new technology introductions.

The need to better curate training data – and to provide greater transparency of algorithms – has been previously underlined by Michael Campbell in an article in Research World, while the imperative for our sector to be on top of all the implications has also been widely commented upon (including my own article, published earlier this year. 

However, our sector still lags in issuing any form of guidance on this topic, despite the incredible efforts of people like Judith Passingham and Mike Cooke, who called our sector to action more than a year ago. For instance, the Marketing AI Institute has already published a guide for CMOs that also includes a manifesto for “responsible” AI (p 17). So, where is ours?

ESOMAR has now begun an initiative to collect people’s opinions on AI that has produced suggestions for actions/initiatives to be taken, but these will likely take several months to produce an output, by which time it will likely have been overtaken by further iterations of already published documents. 

The time for guidance and leadership in this debate is now – time to wake up, people!