BLOG
August 1st 2022

Herberts World #4

Very nice, already the fourth issue of Herbert's World! Thematically all as diverse as our industry is. The column appeared on 28 June - as always on marktforschung.de - or here:

Artificial intelligence as the answer to everything?

42! That was the answer to all the questions! At least that's what Douglas Adams revealed in "The Hitchhiker's Guide to the Galaxy".

What may be true for Mr. Adams, however, does not necessarily apply to everyone - and certainly not to the market research industry! At least it feels like all questions between panel composition, survey methods, analyses and recommendations for action boil down to the following answer: artificial intelligence (AI)!

Just a fortnight ago, the BVM Congress in Frankfurt and the IIEX Europe in Amsterdam took place more or less in parallel and wherever you looked or listened, there was only one topic in the cosmos of researchers: AI, AI, AI!

In fact, virtually no paper, presentation, keynote or table conversation over pretzels and coffee was without these two letters, which for many seem to have almost religious significance. So, are we witnessing the rebirth of the egg-laying willy-nilly that will solve all our problems for today and the future? Seriously now?

AI definitely has downsides too

Sorry, but I see it differently! Admittedly, AI promises huge potential - for a very long time, by the way - but it also has a downside that brings quite a few new challenges and critical questions with it. So it's really like every time an innovative technology comes around the corner.

Without wishing to belittle the importance of artificial intelligence (God forbid!), I personally see three other major issues that we market researchers will have to address and solve in the coming years. AI plays a role almost everywhere, but mostly as a means to an end or as a point of discussion about when technology can actually replace the human mind, reason and, above all, experience, or should "only" support it.

1.) The accessibility and willingness to participate of our target groups Only in the last mafo.de interview with Ulrich Buchholz did I name the accessibility and participation of test persons as a weighty, current problem. A problem that will become bigger rather than smaller in the future.

Sufficiently high participation rates have been the Achilles' heel of us market researchers since before yesterday. Today, we can only dream of values that were achieved in the 90s. Response rates of over 50%, as we know them from the past, now seem like pure utopia. To make matters worse, participants now seem to take their role less seriously, i.e. they take less time on average and presumably also invest less thinking power in a questionnaire.

Of course, this is also due to the omnipresence of market and opinion research - whether online, by phone or live on the street, companies, institutions and authorities always and everywhere want to know something. In addition to this oversaturation, the objects of the surveys, i.e. products and services, are also more individualised, more complex and thus less clear than before. A complexity that makes every survey situation less easy and relaxed.

Using Survey Experience (SX) as a lever

Improving the survey experience (SX) is just one strategy to motivate participants to participate again in larger numbers and to engage more. This includes, for example, modern possibilities to be able to answer questions later online from a telephone survey if desired. Flexible method switching is the keyword here. More variety during the survey, surprising "surveytainment" elements, gamification, monetary rewards or the prospect of exclusive market research results are further potentials. In any case, the following applies: test persons want to be taken seriously and actually valued. And this should be tangible from the start, not just lip service.

And what role does the use of AI systems play here? From internet-based observation of consumer behaviour to voice and interaction analyses on social media channels to automated survey methods online or by telephone, a wide range of innovative possibilities are already in use today to survey and analyse significantly larger numbers of test subjects. But how (highly) valuable and valid are these methods in the end? What are the weak points and where are the limits of meaningful use?

Questions that I will answer at least in rudimentary form in the second point.

2.) The sensible use of technology

AI already makes it easy to find out which people are most likely to take part in surveys, e.g. because of their basic affinity and sympathy for market research or because they want to earn money with it. For fulfilling a high response rate, this way is of course bomb! But what are the answers really worth in the end if, to exaggerate, the same type of person always answers my questions? That's not the representativeness we as an industry always talk about.

Another case: if you automate the collection of consumer data, it leads to a disproportionately larger amount of data than before. For today's IT infrastructure, the sheer quantity is not a problem, but what about the quality? What do I need, what do I want to glean from the data? Which patterns are interesting. Which results may only stop at a certain point?

AI-based social media monitoring: a treasure trove of data

AI can solve clearly defined tasks incredibly well, quickly and thus cheaply. If the algorithm is trained for a clearly defined problem, we are talking about an unbeatable technology that could not be mastered manually, i.e. by humans. If a branded company wants to know in what kind of social context its products appear online - e.g. in which places, at what time, how often and with which emotional words - an AI-based social media monitoring is able to do this. What a treasure trove of data!

But one must never forget: Such algorithms are only as good and as smart as their development teams and the data with which they are fed. If the data is incomplete, incorrect or biased, these weaknesses will ultimately be reflected in the results. But the AI operators are not always even aware of this!

Human emotions or linguistic subtleties such as irony and sarcasm are also still hurdles that AI systems regularly fail to overcome. In any case, a constant and critical review of the data feed and, of course, the algorithm itself is urgently called for in order to effectively counter skewed, erroneous or even dangerous result dynamics. Why dangerous? Here is a nice overview article on this.

Avatars as survey conductors?

Yes, having online market research conducted by avatars or interview bots instead of tedious, static questionnaires sounds extremely tempting with an eye on the SX. But only for relatively simple questions. Real, subtle, emotional and thus truly insightful dialogues between humans (interviewees) and machines (AIs) are probably still a long way off - if at all.

At least one thing, however, cannot be accused of an AI-based solution: That it does exactly what it is told or programmed to do. It does not intentionally influence a test person, does not miscount or commit other human errors. This leads me to the third point that market research must master today in order to retain its trustworthy role in the economy and society.

3.) Quality. Quality. Quality.

The quality criteria of market research (objectivity, reliability, validity and representativeness) refer to the survey design, to the persons interviewed as well as to the survey leader or a questioning authority. The latter could indeed also be an algorithm. But thinking further: would we also put the analysis and interpretation of complex questions in the hands of a computer? If, for example, our client demands strategic recommendations for his company on how to proceed with its brands, employees and customers in the future?

Tact, empathy, expertise and intuition

All three entities - brands, employees and customers - are not only loaded and charged with a wide variety of emotions, they can also be determined and controlled by them. At this point, doesn't data-based market research definitely need sensitivity, empathy, expertise and intuition? These are all qualities that an AI unfortunately lacks, even if a programmer were to stuff it with as much emotional material as possible.

Human judgement is also required, not least at the origin of any market research: What should be surveyed in the first place? And why? What for? What is already known and what can the survey be based on? Which basic knowledge is still sufficiently valid, where does it need to be sharpened? Which old and new knowledge should be linked in a meaningful way? Which data silos do we dissolve in order to create a data lake rich in insights?

Not a replacement, but a supplement

So that I am not misunderstood here: I have an open and positive attitude towards all technical innovations, including AI algorithms. It's not for nothing that almost 20 years ago I was one of the first to back the idea of digital market research, which still seemed crazy and revolutionary.

Online research has since turned our industry upside down - but of course it hasn't completely replaced it. After all, (almost) every old and new method has its own advantages in terms of target groups, accessibility and questions.

And we at moweb have even developed our own AI in recent years, MindsetMatching®. A tool with which advertisers can find and evaluate influencers according to qualitative criteria. And yes, through the use of artificial intelligence, among other things.

AI or not AI? My opinion on this is clear: In certain areas of our industry, AI is already producing impressive results and will definitely displace some traditional tools. However, the more complex the questions and not least the answers become, the more any algorithm will reach its effective limits. Quite simply because it will then only fulfil the above-mentioned quality criteria to a limited extent. And because it cannot replace the decisive authority at the end of the process: The human being in his function as market researcher!

---