AI-Commerce - Are You Thinking or Are You Already Generating? A review at the neuland Fachtag 2023

My mind is still buzzing—just like the beehive—with all the impressions and conversations. A personal review at the neuland Fachtag 2023 by Ralf Zarsteck.

The neuland Symposium is my favorite event: Clients, partners, friends—everyone gathers for a day at our offices, enjoys engaging presentations and participates in workshops, talks with one another and listens to each other, catches up with old acquaintances and meets new ones, explores topics beyond the everyday, and immediately tries to integrate the new insights into their own work. And at the end, everyone looks forward to Pecha Kucha presentations covering both serious and lighter topics.

Maybe it’s a bit like a Swedish furniture store, but it aptly describes the current situation: a general sense that “something big is happening,” some uncertainty about where to start, and a certain urgency to find the right answers for using AI in one’s own business. There was a palpable sense of openness and a thirst for knowledge. Add to that some witty and functional artwork by my colleagues Marco Pantaleo and Fabricius Seifert. Event technology, ambiance, catering, help whenever you need it: neuland at its best. One minor downside: a very image-heavy layout for name tags with tiny text in longer lines. (But we were able to quickly turn that into a feature that fostered connections: put on your glasses, lean in close, “Can I take a look… it’s so small. Ah, now I see it.”)

Depending on the viewer’s level of knowledge, our topic of AI is positioned very differently on the hype scale. The very different perceptions that coexist within the company make it challenging for our customers to adopt AI.

The democratization of AI usage (driven in part by the ChatGPT UI) contrasts with the still-lacking user-friendly language (and the associated technical jargon), while at the same time the technical frontier of “Applied AI Science” is advancing at breakneck speed.

Beyond the technical aspects, the buzzword “artificial intelligence” is used, for example, used to advertise a coffee machine that automatically sorts the most frequently brewed specialty coffees just as much as for the (long-established) analysis of tabular data and modeling—even beyond BI—or the new, and often somewhat overused and nonspecific, “all-purpose weapon” of the neural network.

For many of our customers, AI is still in its “early stages.” Many seedlings are allowed to grow, and we’ll soon see—very soon, in fact—which ones turn out to be fruitful, which are just pretty to look at, and which fall somewhere between companion plants and weeds. And this brings us to the key question of the technical challenges that need to be solved when seeking to apply AI solutions to drive the core business. Or, to stay with the metaphor: gardening is the method of choice. The “container plants” of IT (aka: off-the-shelf products) tend to remain foreign elements; they don’t fit into the ecosystem and thus yield little return. In many cases, they merely fill the gaps in the corporate gardens—temporarily—that obstruct the view of upper management from the front steps.

A familiar pattern keeps emerging from the presentations and workshops: Technical innovations (including AI) make large amounts of data manageable for people. This isn’t a capability exclusive to fancy multidimensional data models; there are also many tried-and-true methods from statistics and mathematics that can be applied to add value.

Handling data requires understanding, and the crucial question at the start of any solution development begins with “What for” and “Why” AI should be used. Only then can the question of the tool (i.e., the “How”) be meaningfully answered.

When searching for solutions, it’s advisable to look into freely available components. Pre-trained models can be fine-tuned for your own use case with minimal effort. Even if you don’t need to understand the models in detail, you should still test different solutions and make a decision based on their effectiveness. At the very least, this approach minimizes the impact of potentially serious problems such as data bias or AI hallucinations.

AI is automatically gaining ground wherever integration into existing workflows and value chains is possible by replacing previous human labor with automation. Product tagging from unstructured data, translations into various foreign and technical languages, and consulting and search assistance are already standard today or will soon become so. And the use of ChatGPT as a personal assistant for handling inquiries and large volumes of text has become the norm much more quickly than some people thought, hoped, or feared—the user numbers and usage times speak for themselves.

Training, expertise, and robust checks and balances must now become standard when using AI in e-commerce. And this applies not only to keeping potentially “hallucinating” algorithms in check, but also to the everyday work of real people. However, the “human-in-the-loop” approach is not a universal or permanent model, nor is it a solution to all problems. The narrower the subject matter and the more specific the task, the faster human approval will become unnecessary—or at least less significant.

Just because a tool exists doesn’t mean you should necessarily let yourself be convinced that there’s a problem that supposedly goes with it. Problems are solved at the operational level; there should be no room for tool fetishism there. For the conference topic, this means: AI can be, but doesn’t have to be, a meaningful contribution to the solution. A careful examination of speed, effort, and costs is absolutely essential to avoid making hasty or even wrong decisions. Using AI without measuring the effort and effects involved is a waste. Often, a combination of proven solutions is just as effective.

We often discuss software development from the perspective of efficiency and subsidiarity—and this applies here as well: AI can do things that we cannot achieve with existing solutions. However, the decision to use it should only be made after exploring existing, cost-effective, and understandable solutions. The ability to get things done better, faster, and more cost-effectively positions AI as a productivity-boosting tool. In this context, measuring efficiency is indispensable.

Or, in short: “AI where it makes sense, not because someone wants to sell a product.”

The workshop “The E-Commerce Store of the Future” by Anita Schüttler and Therese Flämig was structured as a simulation game on the topic of the circular economy in e-commerce.

It began with a concise introduction to the circular economy and its significance for future business models. A lot of content covered in just a few minutes.

In the workshop itself, eight groups of three to six people each were then asked to found their own imaginary company. The participants were tasked with selecting five actions for more sustainable business practices from a set of cards we had created and then combining them into a suitable business model. They were also asked to consider the necessary partners and software. At the end, three groups presented their results and had to answer questions from investors, customers, and employees in the audience.

It became clear that this format is ideal for introducing the topic, sparking conversations, and conveying knowledge in a relatively short amount of time.

My takeaway: We’re definitely doing this more often!

I’m looking forward to the 2024 symposium and hope we can continue these discussions in the meantime.