3 Learnings From 3 Years In Applied AI

4 min readDec 15, 2020

Since 2017, we have been working in the field of applied Artificial Intelligence. Our projects ranged across various industries and across all sizes of corporations. But still, there have been recurring patterns that transformed into condensed learnings. Those learnings happened to be very helpful for early client discussions. That is why we would like to share them.

Learning 1: Expectation management is crucial

Before you start an AI initiative, please consider what your expectation is. By doing this pre-assessment, you make sure not to fall into the “magic-trap”. Don´t get me wrong, AI is a powerful hammer. But not every business challenge is a nail. Be sure, on which of the following three dimensions you start.

Many decision makers want to start at the biggest piece of the pie (Things you don´t know that you don´t know). The notion of “we give you all our data and you tell us our new strategy” seems seductive, but does not work. In this case, an explorative data analysis does the job. But AI initiatives work best, when you put them in context with your strategic objectives, test your hypothesis and build a solution around the problem you solved with AI.

Learning 2: Everyone wants to know the future

At one point, we sat down and analysed all the 100+ use cases we had assessed with clients so far. Those included not only the ones we actually did, but also the ones that clients put on the table as desired future functionalities, products or services.

By far, prediction use cases are the most requested category. Predicting customer churn, predicting logistics capacities or predicting demand peaks in a given period. This showed us that apparently, we all are fascinated by knowing the future. But on a second look, we also observed a big overweight on data-driven use cases. This may be due to the fact, that Machine Learning was (and still is) by far the most prominent representative of the AI technology family. Often times, ML is used synonymously with AI. Most of the other use case groups build upon the given predictions. Based on the prediction, you want your resources to be allocated correctly. This includes scheduling of tasks, planning and allocating space in a delivery truck or dynamically schedule machine utilisation in your factory.

Based on your resource allocation, you want to quickly identify and anticipate exceptions. And down the line, the actual work also needs to be done in terms of running a process on an IT system, adjusting master data or generating and triggering an email workflow. Here, we are more on the Automation side. Less data-heavy, less looked at but not less important if you plan to create a powerful AI solution.

Learning 3: Powerful AI solutions consist of two sides: A data side and an automation side.

What is it worth knowing that your customers might churn or that your machine is about to overheat…if you do not act upon it?

Right — almost nothing. That is why you need both a data side and an automation side. Detection and Action.

Our third key learning was the symbiosis of detection and action. Mixing the data world with the process automation world. Similar like Kahneman´s System 1 and System 2, a holistic AI solution also needs one part that quickly grasps information and detects a scheme, classifies it accordingly and hands it over to the reasoning side, where rule-based decision-making takes over. This approach does not only match the human decision-making, it also creates an explainable AI.

BONUS: Learning 3.5: For scalable (!) AI use cases, solid data integration is key and serves as a foundation for a companies data- and AI-strategy

By far the biggest time eater in most AI projects is data integration. This includes sourcing, cleaning and organizing of appropriate data sets. Depending on a company´s degree of data savviness, this number can go way up than 61%. That is why it makes sense to invest some time to be clear about what you want to achieve with your AI project. Starting with smart data integration and a solid data pipeline eases the way down the road significantly. Otherwise, scalability is at stake and your AI project will have a hard time leaving the PoC stage.

In summary, corporate AI frontrunners have already embraced the need for explainable AI and built powerful products, services and business models around it. With the right use case, diligently assessed on technical feasibility and strategical relevancy, every company can jumpstart their AI journey.

About Norders: NORDERS is a consulting firm specialising in the development of use case specific Artificial Intelligence solutions. Based on advanced analytics models and alternative data sources, we help companies to measure the so-far immeasurable. Combined with intelligent automation technologies, we help companies to create lasting competitive advantages.




Norders is an AI consulting firm that helps mid-sized companies to advance in AI and automation technologies.