ideas

How do German companies start to leverage the power of Machine Learning?

circle white

80% of the German industry start their AI Journey with typical regression / classification use cases

As Kineo, we offer companies easy access to Machine Learning solutions. This means that we are the first port of call for many German enterprise companies when it comes to evaluating use cases.

Artificial intelligence (AI) is a hot topic in almost all larger companies. Every company has it on the agenda to use it, but we realised, after meeting more than 100, companies only 15–20% of companies have implemented an AI-based solution.

The primary reason for the rarity of AI solutions in companies is (lack of) data availability. The common challenges include collecting the right data long enough to build up the necessary amount, digitizing their paper-based data collections as well as making data available within the right format.

Below, we look at two statistics.

  1. What type of Machine Learning algorithms are typically #1 to start and work with?

In Kineo, we work across many industries including automotive, e-commerce, healthcare, metalworking, manufacturing, mobility, and more. Hence, the insights we provide below are not a complete market analysis, yet we believe our numbers can provide indications across the industries.

Machine Learning Types

Almost 80% of Kineo’s proof-of-concept solutions were based on a typical regression/classification algorithm whereas 12% of the use cases are NLP-based (Natural Language Processing). We saw that only about 9% of the use cases are in the computer vision field.

Why only 9% computer vision?

Especially in the area of Computer Vision the entry barrier is relatively high. The problem here is two-folds: a typical solution relies not only on software, but also on excellent hardware, such as camera systems. Therefore, computer vision is rarely suitable as the first use case, since the main objective here is to demonstrate the possibilities that Machine Learning offers on a small scale. We also speculate that high initial costs are a disruptive factor.

When we use computer vision in a first use case, the hardware is usually already installed. Concrete use cases consist of automated quality control, object recognition (mostly in the automotive sector) or predictive maintenance.

Why only 12% NLP?

We realized that many different companies have a handful of common challenges that can be solved by NLP.

Most of the time, NLP is used for the purpose of easing the labor-intensive tasks such as automated processing of fax, mail, e-mail, customer service or feasibility analyses.

In the area of NLP, our experience shows that the quality of the data is not sufficient. Moreover, as companies are in the beginning of their digitalization journey, we find that labeling the data is not a concern yet, especially if the company has no experience with AI. Lastly, most of the digital entries are mutable, leaving the data vulnerable to be manuell overwritten by humans, which makes it difficult to implement the use case immediately. Overall, we communicate intensely with our partners to document such processes to increase the quality of the data. Especially for the first use case in a company, where there is a time-pressure to show the value of the AI project, NLP use-cases are not so widely preferred because they take time.

time

Why so many regression/classification-based use cases?

Let’s take a closer look at the distribution of use cases within this category:

Use Case Distribution

Why do so many companies start with forecasting?

The reason is simple: historically, processes such as documenting the sales, recording possible opportunities and supply-chain planning had been the first to be digitized. Orders, inventory of products and sales consist of the most important data in a company and are therefore well maintained regardless of the intention to optimize these processes by the help of AI.

Since the pandemic, companies are also very interested in models where they can monitor the effect of external factors (such as weather, stock market or progression of a pandemic) in forecasts. We find this approach plausible and experience it first-hand: when a relevant external data source is used in a forecast model, evaluation metrics can improve.

Another advantage is the measurability of success. You are able to compare the AI driven forecast with the forecast in place and measure the value of the AI-based forecast solution.

Every company makes forecasts, but if the AI solution is closer to reality in the following months, the added value and potential of AI is proven and the confidence to invest in these technologies increases: an optimal first use case.


 

Email
Twitter
LinkedIn
Facebook

Read more Blog articles