AI is never wrong?

The term AI is used in a very inflationary way. Every second product in marketing is advertised with AI. There are various definitions of AI, of which I will use the following here.

AI solutions have at least the following three properties: They have what is known as domain knowledge, i.e., information (knowledge) about a specific subject area and associated rules. This means that the AI knows certain objects and situations, for example, and can optimize routes, detect deviations, and suggest measures.

In order to be able to take in data and make suggestions, the AI must secondly be able to act with its environment. At a minimum, this means interfaces to other systems (e.g., ERP, traffic information, inventories, etc., or even sensors that work directly for the AI). Interfaces are also needed for the dissemination of results and, increasingly, for interactive communication with users.

Autonomous but sensitive

Many AI solutions operate without communication to humans. Increasingly - and this is where it strikes us most as users - communication with us is via text or voice input. Today's systems can not only speak tinny, but also express linguistic nuances.

With the information we speak, the stress level can also be determined, for example (an angry caller can then be routed directly to call center employees). They can also correctly assign words with multiple meanings (so-called homonyms) such as "bank" as a financial institution, seat, instance in gambling, or terrain formation ("sandbank"), etc.

Third, an AI must also be capable of learning. This is where most solutions offered as AI-powered solutions fall short. One example is IBM's Supply Chain Insights, which "discusses" successful past solutions with a human team and incorporates the solution found for a supply chain disruption into its repository.

Or inbound call centers that communicate with customers via AI and voice input and output (you wouldn't believe how often this is now used) and then offer the solution accepted by the customer preferentially to the next callers. Similarly, the automated claims processing of insurance companies, which settle more and more expensive claims via AI.

Does AI increase sales?

In a study by Forrester, "Overcome Obstacles To Get To AI At Scale" (2020), customers were asked about the most important goals they expect from the introduction of AI-supported solutions. Top of the list were increases in revenue growth, employee productivity, (positive) customer experience, profitability and business process efficiency. As you can see, a broad bouquet of expectations, but understandable since these are tools that are widely applicable. 

When we look at the status of the introduction of AI solutions, we see that the will is there and that many companies have launched initiatives to this end. In practice, the data required is the biggest hurdle.

This concerns data quality, which is usually not good enough to train AI systems or for which it is not clear whether it may be used for the required purposes. Then there is the integration of the different data silos. The next hurdle is understanding which data is needed for which outcome.

The introduction of AI-driven solutions is not a sprint, but an "endurance run" that never ends. Can AI solutions also err? Unfortunately, yes. However, the better the rules, the better the system learns, and the better the data, the more optimal the results.