Achieve more sales in e-commerce with artificial intelligence: An expert discussion with axytos CRO Matthias Schubert [guest contribution]

axytos CRO Matthias Schubert: “An AI must be trained by an expert on the basis of as many and correct data sets as possible, which must also be formulated correctly.”

A small advertising block is allowed. But what is the basic idea behind this novel approach?

It is basically about the acceptance rate, i.e. the decision to whom I as an online retailer even offer a purchase on account in the checkout of my shop. As a merchant or manufacturer who wants to implement this on his own, I need reliable information about the buyer so that the risk of default remains manageable.

Very “good” and very “bad buyers” can still be easily identified. But there is usually very little information about the 15 to 20 percent in the grey area, who would like to pay by purchase on account. However, these buyers are about an immense, valuable sales potential that you cannot simply give away as a merchant.

As far as understood. And when does artificial intelligence come into play here? Finally, the title of your presentation is ‘Insert AI – achieve more valuable sales’.

To explain the AI methods used on the axytos platform, I would like to use an image: For the first step, imagine a fairly rough sieve that uses classic methods to determine which buyers the merchant offers purchase on account and which does not. In the first AI analysis, a finer sieve is then used to determine from the buyers in the remaining grey area that the merchant can also enable a purchase on account.

We repeat the whole thing in further iterations with ever finer sevens, whereby the remaining grey area is repeatedly divided into green, gray or red. Practically, this is an infinite loop, as new data is always imported into the system.

Is that an appropriate task for AI?

Absolutely! The strengths of artificial intelligence lie in pattern recognition based on pre-defined properties and constant recalibration based on experience, i.e. fresh input. This is a typical machine learning application, and in this firmly defined area of responsibility, self-learning algorithms also work very well.

The ability of an AI stands or falls with the quantity and quality of the data material. Where does the data come from?

The information on feeding AI comes from several data sources. Internal data comes directly from the shop, from the merchant, from the customer and his shopping cart: Which device and at what time is he travelling in the online shop and what does his shopping cart look like in terms of size and composition?

The whole thing is enriched with reliable information via the axytos platform by numerous data sources, for example about the creditworthiness and identity of the buyer: Can his account number be verified from free sources, is his e-mail address genuine? In addition, there is a fraud filter that detects multiple orders and stolen identities, among other things, and establishes a correlation between creditworthiness and shopping cart content.

They say an AI is only as good as its programmer. Is that true?

Yes, that is absolutely right and, in my opinion, will remain the case for even longer. An AI must be trained by an expert on the basis of as many and correct data sets as possible, which must also be formulated correctly. This is precisely why the axytos team also includes human analysts. They form an integral part of the platform and also ensure that a single ‘black swan’ does not confuse the whole algorithm.

Can you explain this as an example?

The so-called black swans are unlikely events that neither humans nor AI systems can predict. Self-learning AI systems have difficulty dealing with it in principle, since swans are predominantly white (grinning). A somewhat exaggerated example: A fully legally competent retired official moves to a city with high unemployment and a low average income. The so-called geoscoring, i.e. the combination of place of residence and creditworthiness data, now means that he can no longer buy his beloved shoes on account.

The algorithm could interpret this in such a way that all ex-officials in this city are excluded from buying on account, i.e. get the red flag. Here the axytos experts intervene and recalibrate the algorithm, our official gets a green flag and can buy his shoes as usual. AI has therefore learned not to make a decision solely because of the changed place of residence, but to take more account of other parameters.

Good and beautiful. But what do online retailers get out of it?

To put it briefly: By using AI methods from the axytos platform, the acceptance rate for merchants for the payment-guaranteed purchase on account is significantly higher than that of market competitors. As a result, this ensures higher sales and better customer development – with maximum transparency and controllability of the processes.

Mr. Schubert, thank you very much for the conversation!