Get in Touch
With Our Experts
+49 711 / 75886-600
Please Type A Message
Let us know where we can contact you
Please click to start verification
Send message
Thank You For Your Message!
We will contact you as soon as possible.
Close Window

Why AI Projects Fail – And What Organizations Can Do About It

25 April 2023
Marie Wetzel
Artificial Intelligence (AI) is an emerging technology that offers enormous opportunities for companies, but also comes with challenges. In this interview, Steven Jones explains why many AI projects fail and what companies can do to increase their chances of success.
Steven Jones

As "Director AI & Analytics", Steven Jones is responsible for the areas of data analysis and Artificial Intelligence at Parsionate. Since 2016, he has been advising and supporting his clients and accompanying them on their way to becoming a data-driven organisation.

His team enables companies to understand their own data, to collect new data and to use this in a targeted manner for the benefit of the company in order to generate measurable added value.

How do you assess the current status of AI projects in Germany?

When it comes to AI projects, people often refer to the USA, which is considered a pioneer in this field. But is that really the case? A comparison shows that many AI projects also fail in the USA or do not deliver the desired results. The difference, however, is that American companies are often quicker to learn from their mistakes and develop new approaches. German companies should sometimes take an example from this mindset.

Meanwhile, there are also many companies in Germany that invest in AI and are already successfully implementing projects. But just as many cases of AI projects that fail miserably. Above all, it is important that the goals and expectations for an AI project are defined realistically and that all departments and employees involved are well informed. Only then can an AI project be launched successfully.

Which use cases for AI are currently the most promising?

Currently, a very promising area is the automation of business processes, such as the automatic creation of invoices or the identification of fraud cases. AI is also being used successfully in the area of personnel recruitment and development to analyze applicant profiles or predict employee changes.

Another important use case is the personalization of customer interactions to ensure the best possible customer experience. Especially in this area, there is still an enormous amount of unused potential, as many companies only use a fraction of the available data or the theoretically available knowledge in a purposeful manner. At Parsionate, for example, we used machine learning algorithms to help porta, a furniture retailer, optimize its customers' buying experience across all channels.

Depending on the source, 60-80% of all AI projects fail. Why do you think that is?

I think one of the main reasons is that companies often don't exactly know what they expect from AI and what problems they want to solve.

Another reason for the failure of AI projects is also the lack of high-quality data. AI systems need large amounts of data to work effectively. If the data is not available in sufficient quantity or quality, the AI system will not deliver the desired results.

The lack of AI experts or data scientists in the company can also be an obstacle. It is important for companies to invest in talented employees to ensure they have the skills and know-how in-house to successfully implement AI projects. When I say skills and know-how, however, I don't so much mean the ability to build the various layers of a neural network, but rather soft skills such as an increased tolerance towards frustration and a willingness to really dig into a problem and not give up too easily.

In addition to a clear strategy, the careful selection of project partners is also elementary to the success of AI projects.

What else can companies do to minimize the risks in AI projects?

It's important to understand that AI is not just about the technology. Companies need to be aware that they must adapt their business processes and strategies in order to use AI successfully. It's about collecting and analyzing the right data to make informed decisions.

Furthermore, companies need to consider the ethical and legal issues surrounding AI and ensure that systems are transparent and accountable. In Europe, for example, compliance with the GDPR plays an important role when processing personal data.

I believe companies need to fundamentally adjust their mindset and method of working in order to be successful with AI.

What particular challenges do AI projects present compared to classic data management projects?

The difference lies, on the one hand, in the complexity of the projects. While classic data management projects aim to collect, store and manage data, AI projects go one step further. Their goal is to recognize patterns and trends from this data and to make predictions.

Companies must ensure that solutions developed are quickly put into operation, where they are regularly adapted and monitored. A large proportion of prototypes do not even make it into operational use because the initial results, often based on poor data quality, do not directly meet expectations and are abandoned too quickly.

Another difference: In order to generate sustainable benefits, setting up a cloud infrastructure is usually indispensable for AI projects. Today, classic data management projects can usually still be carried out on-premises without any problems. Machine learning algorithms, on the other hand, are often very resource-intensive. In order to be able to scale when calculating models and to obtain results more quickly, dynamic use of computing power is essential. And this is only possible in the cloud.

AI projects therefore require somewhat more investment than classic data management projects. But I would go out on a limb and say that the potential profit for the company is also much higher.

What approach do you follow in the Parsionate AI team to make your projects successful?

At Parsionate, we are aware that the actual code for a machine learning model represents only a fraction of the actual work in a typical project. Collecting the data and making it usable, as well as operating a permanent cloud infrastructure and monitoring the results, represent a significantly larger effort in the project.

Components of an AI project

I really like this chart, which shows how much effort is actually involved in the individual areas of an AI project. A holistic view of an AI project is elementary for us.

For the launch of a project, we have developed the AI Quick Check. We start by examining a company's infrastructure, processes, use cases and, above all, data. We help to avoid common mistakes and to get the project off to an optimal start.

Once we have worked out realistic expectations and goals together, our experience comes into play. AI projects are not about functionality, but primarily about data. That's why we help our customers with a suitable project methodology. This methodology is neither pure agile nor pure CRISP DM (Cross-industry standard process for data mining), but rather a good mix of both approaches. On the one hand, there are some steps that need to be executed in a certain order to achieve the desired results. On the other hand, however, since artificial intelligence is fast-moving and constantly changing, an agile component is essential.

Last but not least, as a team we rely on a great deal of experience with various tools and algorithms and always stay on the cutting edge of technology. And, perhaps most importantly, we have a desire to experiment and grow with new challenges.

Ihr Webbrowser ist veraltet

Aktualisieren Sie Ihren Browser damit diese Webseite richtig dargestellt werden kann.

Zur Infoseite