Different AI problem types

Ansgar Bittermann
3 min readMay 15, 2021

--

Gerd Altmann on Pixabay

In order to give you a better start for your AI brainstorming session, let us look at eight problem types which can be solved with AI solutions. This list is of course not finite or complete, but it covers big problems many companies face. Some might apply to you and some not, so pick the ones you also see in your company and develop them further with your team as potential “First AI project ideas”.

Real-Time Optimization

Real-time optimization (RTO) does not always have to be real real-time, but can also mean hours, but in general RTO tries to optimize processes for systems or machines on their own — continuously and autonomously. An example is the optimization of delivery roots for package delivery companies. But in general it is a way to enhance the performance of any system — model based and on its own.

Strategy Optimization

If you are in the banking, agriculture or media industry, you have probably heard of strategy optimization. In banking you have systematic trading or algorithmic trading, in agriculture you have optimized planting strategies and strategies for optimized watering (e.g. in rice fields) or in media you constantly optimize your schedule or articles to obtain highest views or clicks. In all three industries, optimizing your strategy is a huge part of your daily work. The amount of information is extreme high, changes constantly and the decision patterns are highly complex. Thus the human mind quickly reaches its limit by finding global maxima.

Predictive Analytics

Problems regarding precisions can be found in almost every industry. The target might be different, but the problem is the same. How can you predict important KPIs in your business using historic information? Many companies will start here as it is easy to understand and supporting machine learning models are very mature.

Predictive Maintenance

If a product or machine breaks, your customer or employee is dissatisfied and quickly — depending on the severity of your product or machine — in financial or physical danger. Are you selling system critical parts for machines? Are you responsible for developing motors or any high value, critical product for elevators, hospitals, industry? Then this is your category. Imagine you are selling elevators and you could predict when an elevator would stop functioning. This would be huge unique selling proposition.

Radical personalization

Radical personalization is driven by individualization. Customers expect that products fit to them and not they to the product. With the manifestation of industry 4.0 and smarter production systems, a radical personalization or individualization of products is possible. Many people think of personally branded sneakers, but also think of compression socks, heart valves, walking sticks and cars. Radical personalization is going to be a challenge for every customer facing industry.

Discover new anomalies

Anomaly detection is primarily a problem mass producing industries like automotive, pharma or telecommunication. If something goes wrong in the process of producing one million pills an hour or on the assembly lines of high-volume car manufacturers, the costs of failure are very quickly very high. Thus a faster process for discovering (or even predicting) anomalies can help to save a lot of money.

Forecasting

Forecasting is a sub-category of prediction and it tries to predict future events based on time series. That’s why it is called weather forecast and not weather prediction. Weather forecast always takes the course or progress of the weather and uses it to predict the future weather. If it would be weather prediction, it would just use distinct variables (pressure, temperature, humidity, wind…) to predict future weather.

Forecasting can be found in any industry.

Processing of unstructured data

The reason why data engineering is making up the most part of artificial intelligence is that the data is sadly only rarely already in the form the data scientist needs it. This has to do with the multitude of systems which are involved in creating the data. Just imagine the different kinds of data being crated in the transport and logistics industry? So many players in this process, so many different systems and technologies, so many different countries and languages. In industries like the logistics industry, processing the huge amount of unstructured data is one of the key problems.

— —

If you want to become A.I. ready, just contact me at ansgar_linkedin@goldblum-consulting.com

--

--