In many areas, personalisation represents great progress, economically and socially. It has complex causes and is, among other things, a result of scientific progress (which gradually overcame average thinking), as well as the crisis of industrial society (which led to more modern, customer-oriented marketing approaches), growing value orientation (which divides society today) and, last but not least, the digital transformation (which led to information being derived from data for which it was not originally created). The goal of data-based personalisation is – in most cases – to maximise the value of products and services for customers and/or to maximise the revenues generated by them. Depending on the context, we also have to deal with variations of this goal, for example when the revenue perspective changes the customer benefit perspective. Social networks use usage data to try to keep customers on the network as long as possible (and thus steal as much time from them as possible) in order to show them as much advertising as possible that is paid for by third parties. In some parts of the US, judges also personalise sentences and police interventions against residents based on a combination of data and interpretations (logical shortcuts) in order to supposedly construct the greatest possible benefit for society. This is often the “other side of logic”, because this personalisation of policing and punishment is usually based on gross generalisations that are either completely wrong or pay homage to average thinking that ignores the context of action.

The double added value of digital personalisation

Let’s look away from dubious forms of use for the time being and concentrate on personalisation that actually creates value: at its core, this is about optimising conditional expected values – mostly (as outlined above) in relation to customer benefit and in relation to price. Customers should receive the variation of a product or service that is probably most valuable to them individually and pay a price for it that exploits their individual willingness to pay as much as possible. This optimises the added value for the customers on the one hand and the added value of the customers for the provider on the other. What is new is that one does not start from average values – neither from average values for the whole society nor from average values of the targeted customer segment – but from conditional expected values. Their estimation is derived from digital behavioural data, so-called digital twins. The optimisation task is as follows: Given the data on a customer (= condition for the formation of the expected value), which product variation would bring her the greatest benefit and what is the maximum amount she is willing to pay for it? To solve this task, the data of many other customers are used. The results are then used to calculate the actual product variation offered and the actual price offered with the aim of ensuring that the company performs best in the sum of all offers. This is a technical process that also uses data, but also depends on the company’s strategy, because any optimisation in a complex context is based on an arbitrary simplification. Conditional expected values can be thought of as averages of very small sub-segments that are just large enough to have statically relevant data about them. In practice, however, they are also calculated from estimates and are not based on data about a concrete sub-segment. The latter are modelling details that have little relevance for understanding the consequences of digital personalisation. In any case, the imagined sub-segments are much smaller than the segment of all targeted customers, which is why the product variations and prices offered can vary greatly in the case of personalisation and thus generate double added value as described.

A special area of application: working with people

Personalisation is conceivable in almost all areas of the economy – from personalised mass-produced products to highly specialised services. It has a particularly large social benefit in the health sector. There, it is possible in many cases even without conventional data analysis – with markers, for example, the effectiveness of therapies can be assessed in advance, and with genetic material analysis, individually tailored medicines can actually be designed. In other areas, however, the conditional expected values described above are applied mutatis mutandis. Depending on the perspective, however, the specialists in the case do not speak of personalised medicine, but of precision medicine. This conceptual preference does not change anything. The specialists are primarily concerned with being correctly understood by unsuspecting laypeople. Another special area of application for digital personalisation is education. It is also about “working with people” and about “co-production” between service provider and customer. Currently, there are interesting considerations and pilot tests – especially at university level – to offer data-based individual support for learners or even to propose individual learning paths. Although these projects are still much less advanced than digital personalisation in cutting-edge medicine, there is a lot of evidence for the potential of such digitally personalised education.

Different forms of cognitive intelligence

We know from research on infant development that not all children follow identical developmental paths. Learning research has impressively shown us that many “natural assumptions” are wrong. Being able to learn faster is useful, but neither sufficient nor necessary for cognitive excellence. Cognitive excellence itself is neither necessary nor sufficient for success in life. Moreover, there is not one cognitive intelligence, but very different forms of cognitive intelligence, which do not necessarily exist to the same extent in the individual. Incidentally, healthy people have learning abilities up to a ripe old age, including the associated remodelling of the brain, but they change in the course of life. The ability to learn something completely new disappears, the ability to network knowledge increases, the view of substantial patterns often only emerges at an older age. Accordingly, universally applicable training concepts that are not adapted to the specific current skills make little sense. The goal must be an appropriate personalisation of teaching. To understand this, we need to clear out our preconceptions. Playing football is an illustrative example of how specialised cognitive skills can shape elite sport without turning most elite athletes into intellectuals. The connection between motor and specialised cognitive abilities that pops up in the example of football reminds us that before the triumph of machine learning, the traditional New Artificial Intelligence was concerned with the emergence of intelligence from sensorimotor interaction – and renowned top researchers are still doing this. Meanwhile, there is also interesting research-based speculation that cognitive intelligence may be the developmental consequence of emotional intelligence. And, to complete the picture, psychology has shown that due to different parenting practices, young people in the West and young people in the East have very different cognitive abilities. But instead of lumping everyone together and classifying them one-dimensionally, it makes much more sense to personalise education more on the basis of data. This conclusion seems compelling. There is also a lot of lip service paid to it. However, the exact opposite can also be observed in the higher education market, the emerging generalist Bachelor’s degree for all. In fact, the situation is more complex – as with very many forms of digital personalisation. Let’s look at the obvious risks first, before exploring the issues in detail in Part 2 of this essay.

The triple ambivalence

As we saw at the very top, personalisation, despite its simple principles, is a wide country. Often enough, this vast land is acted upon contrary to data evidence and the findings of contextual analysis. This does not change the fact that sensible, competent and information-based personalisation can be of great benefit. But the perception of such perceived “real” benefits can also be very different. Because from a value-neutral point of view, personalisation is always discrimination – a conscious distinction of the counterpart that is reflected in practical action – which is why we actually perceive personalisation as discriminatory in many contexts, even when it is handled in a factually competent manner. Personalisation therefore comes with two obvious problems in practice: if it is practised reasonably and competently, it can still be perceived as unfair, and if it is handled unreasonably or incompetently, it represents gross mischief that may even harm everyone. So far, so obvious. But there is a third, less obvious, problem. Consumption and consumption opportunities change people – whether it is a matter of products, services or merit goods. Those who get their needs met to the maximum follow a different path in life that may be much less abundant, lead to much lower individual capabilities and offer much less benefit to others. The same applies if the needs of others, in the field of higher education for example the employer, are fulfilled to the maximum. The best is indeed often the greatest enemy of the good. The question is – to use the iceberg metaphor once again – is the visible benefit offset by greater, often barely discernible harm? Or is the damage a manageable one? Or also: Is the damage great, but caused by a few harmful patterns that can be specifically eradicated? In order to be able to decide this three-fold question on a case-by-case basis – both as a society and as consumers – it is important that we anticipate the consequences of personalisation. Or that we, if we do not anticipate them – we are usually pretty bad at that – at least recognise and reflect on these consequences.

Part 2 will be published shortly. It will present some basic considerations on the impact assessment of digital personalisation.