This post wants to show you how artificial intelligence can personalize webshops autonomously to improve customer experience and ultimately increase sales. Special focus is on the simultaneous personalization of several elements of the webshop for a coherent overall experience.
Personalization allows you to tailor websites to meet the current, individual needs of your visitors.
As so often, an example helps: One visitor to a web store may be a “bargain hunter”, another may prefer high-priced items. If in the case of the first customer relevant, i.e. cheaper goods are displayed than in the case of the second, we speak of a (partially) personalized web store.
Why the effort of personalization?
Quite simply, customizing the user experience has been proven to have a positive impact on buying behavior.
Personalization is not limited to only displaying specific goods, as in the example above. Theoretically, every element of a web store can be customized to individual needs. Other examples of this are:
As defined above, personalization is the situational adaptation of a website. A situation is thereby characterized by features and the website is adapted according to the characteristics of these features. These characteristics define the customer’s profile in his or her current situation. The website should be personalized according to this customer profile.
Features that can be used as a basis for personalization can be divided into the following three categories.
The more characteristics are used, the more precise personalization can be achieved with sufficient amounts of customer data – from websites personalized on the basis of simple customer segments to a truly individual user experience.
The difference between the first of these categories and the other two lies in the time when they become available:
While personal interests only become apparent in the course of the customer’s interaction with the website, the information listed under 2. and 3. is available in the first milliseconds of a website visit and can be used for personalization.
If it concerns a returning visitor, past personal interests are also available in real time as a data source.
The large number of elements that can be personalized and the multitude of different customer needs result in an almost unmanageable number of possible assignments (cf. Figure 1). To personalize means to know which combinations are the right ones. A further complication arises from the fact that users change their behavior. Rules according to which web stores are personalized therefore lose their validity over time.
People quickly reach their limits when it comes to finding and maintaining this set of rules. Personalization therefore absolutely needs some form of automation. This calls for artificial intelligence, a special form of artificial intelligence: so-called self-learning software, i.e. software that can independently figure out how the elements of a website need to be adapted according to user needs.
This requires a precise definition of the goals of personalization. The user selects the KPIs that are to be optimized by intelligently adapting the elements of the web store. This adaptation is usually subject to business logic, i.e., boundary conditions that define the scope of action of the self-learning software.
Self-learning software involves a so-called artificial agent learning from its own interaction. In the context of personalization, this means the following. The artificial agent adapts an element of the web store for a user and registers the user’s resulting behavior. If the purchase of a product follows, for example, a product description that has been changed in some way, this action was probably not completely wrong. Consequently, this action in this situation is “upgraded” to some extent.
This means that corresponding parameters of the algorithm are adjusted. The artificial agent has then learned the following: This visitor or a visitor with comparable characteristics responds positively to the changed product description. If this visitor returns later, or if it is a comparable visitor, this measure should be repeated in most cases.
But why only in most cases, why not always?
User behavior is changing, and with it the necessary personalization. Users also seem to be changing their behavior more and more quickly. This may be due to the increasing transparency made possible by the high availability of comparison portals. For example, if you are not able to adjust prices quickly enough, you will lose customers who have found the product they are looking for cheaper elsewhere.
An artificial intelligence worthy of its name must paradoxically act less intelligently from time to time. In other words, it has to try something suboptimal, like adapting a product description that has worked less well than another one so far. But why does it have to do this? The answer is simple: it could be that this product description works better. Possibly because it has “always” done so, but perhaps also because visitors have imperceptibly changed their behavior in the meantime.
An artificial agent should therefore always invest a little in the exploration of supposedly negative experience in addition to exploiting existing, positive experience. Like humans, artificial intelligence can learn from trial and error and in this way adapt. Humans, for example, are bound by the social order, while artificial intelligence is normally subject to business logic. Accordingly, not every trial is permitted.
In both cases (exploitation or exploration), however, it is important not to make the same or comparable mistakes unnecessarily often. So having a notion of comparability of situations is desirable.
In the previous section, I wrote about comparable situations or customer behavior. As mentioned, having a concept of comparability is an advantage. The reason for this is that with the multitude of possible situations and the diversity of customer behavior, even with a huge wealth of experience, there are always situations that are different from those seen before. Those who can then draw comparisons with what they have seen before, who are able to generalize previous experience, as it were, will be able to make a well-founded decision despite the new situation.
Formally, this means identifying invariant features of situations that require the same optimal decision. In this context, we also speak of pattern recognition (see Figure 2). In the example of the customer, it may be that the effectiveness of a particular discount campaign, for example, does not depend on the day of the week, but on whether a holiday is coming up. The proximity to a holiday would therefore be one of these sought-after invariant characteristics.
Deep learning, another method of artificial intelligence, has the outstanding property of independently identifying those features in data that allow the best possible generalization. This has made it possible to achieve performance on a par with human ability in some fields of application, such as the diagnoses of leukemias. These algorithms were often able to detect and exploit features that were hidden from humans.
Besides complex strategic tasks, humans will probably still be superior to artificial intelligence in the long term in the following task: Knowledge transfer. Humans are outstandingly good at transferring knowledge acquired in one context to another. Software has had a hard time at this task so far. And yet, algorithms can be extremely useful in clearly defined problems.
For instance, deep learning finds application in drug design, language recognition, or image restoration. The combination with a self-learning component allows users to enjoy the advantages of both worlds. Corresponding algorithms make it possible to make independent decisions even in unknown situations by means of generalization. However, deep learning is not always the best way to generalize.
As a matter of fact, in our experience, a lot of money has been sunk in recent years into projects that would have been better implemented with less complex algorithms. Deep learning methods all require large amounts of data to create value. Not every use case meets this requirement. That’s why you should never rely on a single algorithm, but always on a series of complementary algorithms that can take the helm depending on the data situation.
To summarize, effective website personalization requires two basic skills. 1. to generalize and 2. to be able to adapt independently. There are software services that implement this more or less well.
All these solutions, however, have one crucial flaw in common: Elements of web pages are personalized independently. Unknowingly or against one’s better judgment, it is neglected that all personalized elements add up to an overall user experience. This is often sold as an advantage: a long list of personalizable elements is presented as a complementary range of services. Product recommendations, for example, are used independently of personalized discounts.
It is, however, axiomatic that a customer must find the right product at the right price, otherwise he will not buy it.
So the left hand literally doesn’t know what the right hand is doing. The user experience becomes a patchwork of separate optimization attempts, similar to a poorly managed company whose departments optimize according to their own metrics and there is no coordination with regard to the overall corporate goals.
As a result, potentially significant portions of profits are not realized. And one more example should help: In soccer, a well-rehearsed team of merely “serviceable” players can beat a team of gifted but egotistical soloists.
In e-commerce, as it were, a harmonious interplay of personalization of all elements is required.
Since now there are hopefully convincing arguments for having the holistic personalization of several elements implemented by a single AI, the following will show how this can be implemented in practice.
This practically means that in each situation, all the respective customizable elements in all their variations are presented to the agent for selection. The crucial point here is that this list is not limited to individual elements of the web store, but encompasses the totality of all combinations of possible variations of the elements.
Consequently, it contains, for example, options for redesigning the layout and design, the range of possible product recommendations, and possible discount variants. With two possible products (blue and red) and two possible discounts (5% and 10%), this implies four different combinations (blue-5%, blue-10%, red-5% and red-10%) that the artificial agent has to use situationally.
Through the evaluation of the resulting user behavior, the artificial agent can then identify particularly lucrative combinations in a user-dependent, i.e. personalized, manner. However, with all these combinations, it becomes even more difficult to solve the problem of optimally matching user experience and user behavior.
The following video shows how artificial intelligence learns to use tools cooperatively in the game of hide-and-seek through trial and error.
An appropriate artificial agent is then not subject to any systemic barriers. If a sufficient number of interactions and resulting user behavior are available to it, it will always be able to derive effective recommendations for action.
Because of their historical development, existing web stores in their current implementations are typically static in nature. With our service, we offer the possibility of gradually integrating personalization into these systems in a controlled environment, in order to keep expenses for the new development of IT infrastructure low on the one hand, but on the other hand to be able to profit from new technologies already now.
Over the long term, the industry is facing a paradigm shift that will make dynamically generated content the core of the next generation of store systems. The trend is therefore away from a one-size-fits-all solution and towards a fully personalized shopping experience to meet the demanding customer of tomorrow.
As a Data Scientist, I have been working for DAX companies and startups for many years.
With my Munich-based company, res mechanica GmbH, we launched a personalization service ourselves in 2020.
The technology behind it achieved the highest score ever awarded in the prestigious “EXIST” grant from the German Federal Ministry for Economic Affairs and Energy.
Do you have any questions or comments? Then I am looking forward to hearing from you.
Workshop B2B Market Positioning
6 vital tools to differentiate and convert more leads with Laura Schulte (former Managing Director Windeln.de SE) together with André Wehr from tractionwise. Become the Survivor of the Digital Fittest.
100% free of charge | 25 min. individual consultation
This service is not only available because of the current corona situation.
But right now it is more important than ever, together and for each other.