Predictive shopping
Or, When E-commerce Knows What the Customer Wants
A kiss on his Qualitypad had been enough to sign up for TheShop’s premium service OneKiss. Now Peter Unemployed gets the company’s products sent to him automatically, without having to order them. This sounds like a work of fiction, right? In Marc-Uwe Kling’s bestseller Qualityland, a system calculates what the protagonist consciously or unconsciously wants, without them intervening in the process. When Peter Unemployed tries to return a pink dolphin vibrator, the android at the service center says: “I’m afraid that’s not possible”. Peter Unemployed replies that he doesn’t want the vibrator. “But you do want it,” Kling has the android of the online retailer TheShop say, which advertises with the slogan: “We know what you want”.
Does it have to end like this? Is this our future? No, I don’t think so. In fact, predictive shopping based on artificial intelligence can fundamentally change commerce if companies can predict with a high degree of certainty what their customers need — and therefore send them these products without ordering them. This could not only enhance the customer experience, but also help to improve production flows, make processes in companies more efficient, and optimize supply chains by aligning each of these points with the calculated wishes of the customer.
Everything is networked
One pillar here is the Internet of Things. In an increasingly networked world, more and more data is available which can calculate future purchasing decisions. A simple example of this is a coffee machine. Sensors can check fill levels, record consumption, and derive preferences and habits from this to deliver the appropriate refill products directly to the customer’s home. Other more complex examples are conceivable. Weather and movement data, for example, could be of interest to textile manufacturers in order to deliver clothing that fits perfectly. Fitness data, as well as surveys about screen time and general smartphone use, could also be used by producers of sports shoes and manufacturers of fitness equipment.
The crucial point, however, is this: do these eCommerce merchants have enough data themselves? And if so, do they have the right data models? Are they reading enough of their data? Everyone is familiar with the product suggestions of large eCommerce sites that draw on previous purchases, search behavior, or the purchases of other customers to point to other offers that might also be of interest. Often, these suggestions are inappropriate. If you want to dive deeper, I’ve written about that here, too. After all, just because I was interested in a new garden table once doesn’t mean I want to buy several of them right away.
However, predictions for children’s clothing sizes, for example, can be very accurately calculated. This is because, on average, size development quite clearly follows comprehensible rules and therefore the most suitable sizes can be suggested to the customer for the next purchase of children’s clothing. And even if it is not immediately compulsory set to default, the customer experience would be enhanced with a simple function such as the following: “With your last order you had ordered the girl’s dress in size 122. Are you sure you want to order this one in 110?”
Data and free shopping
A second pillar is networking. Predictive shopping uses artificial intelligence and machine learning to draw conclusions from the multitude of data. The more high-quality data is available, the more precise the predictions can be made. For eCommerce, however, this means: which offline data can be combined with my online data? Which data sources can be additionally tapped?
The example of the so-called “zero hit” analysis shows how valuable the collection of data is. Here, an online store collects and evaluates the search terms that have led to an empty hit list in a customer search query. The analysis may also reveal that users make many typing errors. However, additional insights can be gained, such as the following:
a) Customers regularly enter a difficult brand or product name (example: “Aerpods”) incorrectly. Here, a synonym dictionary can then be created within the search, so the customer can be automatically guided to their correct target (“Airpods”) in the future.
b) Customers enter search terms that belong to products that are not in stock. Here, the retailer can then decide whether demand is high enough to expand the range to include the in-demand products.
Some will counter that all this data and the use of artificial intelligence cannot definitively predict human decisions. However, it is also true that large parts of nature are controlled by biological, chemical and physical processes. The principle of cause and effect prevails. To this day, Libet’s 1984 experiment remains a subject of controversy. Authors like philosopher Sam Harris even go so far as to dismiss free will entirely as an illusion. It is certainly not necessary to take such a radical standpoint, but one can at least acknowledge that far more of our desires and preferences can be calculated, computed and predicted than we sometimes admit to ourselves.
And now what about the dolphin vibrator?
By the way, TheShop from the novel Qualityland is prone to delivering the odd unsuitable product to customers from time to time. Out of calculation. The accuracy of the predictive shopping system was scary for the customers. The dolphin vibrator is not such a product, as its delivery is based on missing or insufficient data. This remains the challenge par excellence for any predictive shopping system — even in dystopian, but entertaining, novels.
