Consulting firms use machine learning

Research

The hype surrounding machine learning (ML) and its sister technology, artificial intelligence (AI), is gradually giving way to a pragmatic approach in German companies. IT experts and specialist departments set about developing application scenarios ("use cases") for machine learning and putting them into practice. There are almost no limits to the imagination, according to a group of experts who, at the invitation of COMPUTERWOCHE, discussed the status of ML and AI and German companies.

"At Microsoft we are currently registering a certain amount of curiosity among companies when it comes to machine learning and artificial intelligence. A number of companies have set up smaller projects in this area," says Jürgen Wirtgen, Dataplatform Lead at Microsoft Germany. In his opinion, both approaches are developing into tools that companies use just like other software tools. "To do this, however, it is necessary that the consulting firms make ML 'easy' and provide users with the right tools," says Wirtgen.

Telefónica has also registered a "positive attitude" towards machine learning among German companies. The telecommunications service provider uses machine learning itself, as Thorsten Kühlmeyer, Head of Business Analytics & Artificial Intelligence explains: "Machine learning and AI facilitate the maintenance and optimization of the mobile network, help analyze social media, accelerate the processing of service requests and network Employee."

Information on the partner packages of the machine learning study

The financial and insurance industries are pioneers in the area of ​​machine learning, "for example in the context of risk assessment when granting loans", says Kay Knoche, Principal Solution Consultant DACH Decisioning Solutions at the software house Pegasystems. "Questions are clarified, for example which up-sell and cross-sell product should be offered or whether a claim can be settled unbureaucratically or it makes sense to wait for an expert opinion first."

No need to be complacent

However, there are also critical voices about the situation of machine learning: "In Germany, the topic of the use of machine learning has not really caught on in comparison with countries like the USA. The reason is that there are German companies compared to companies in the USA there is a lack of willingness to take risks, "states Karl Schriek, Head of AI / Leading Machine Learning Engineer at the consulting company Alexander Thamm. In the case of projects in the area of ​​machine learning, he believes that German companies have a tendency towards perfectionism that hinders the marketing of solutions.

Farhad Khakzad, an expert in the field of analytics who most recently worked as Head of Risk Analytics in an international technology company, cites another weak point: "In general, it can be said that local companies have ideas for practical application examples, i.e. use Cases are missing. One reason is that companies often lack an idea of ​​the extent and type of machine learning to be used. " Another reason are partially unrealistic ideas about the possibilities that this technology offers.

And Paul-Louis Pröve, Consultant Data Analytics, Artificial Intelligence & Blockchain at Lufthansa Industry Solutions, sees the discussion about ML and AI "strongly shaped by marketing statements, according to the motto: 'We are now also doing machine learning.'" Part of the company first of all develop use cases.

The "right" use cases are required

Use cases that can be easily implemented are a good starting point for companies. "The aim here should not be to solve a specific business problem, but to get a feeling for the possibilities of this technology," emphasizes Dr. Karsten Johannsen, Business Development Executive Artificial Intelligence at Tech Data. It is important to provide the required data in advance and to check its quality. "The preparation of the data is currently the greatest challenge," confirms Christian Dyballa, Head of Sector Financial Services - Insights & Data at Capgemini in Germany.

According to Thorsten Kühlmeyer, a multi-stage process is recommended for projects to bring the hoped-for success: First, companies should identify and analyze the problem. A use case is then derived from this. "Only then do you look for the right tool," says the Telefónica expert. Classic analysis methods could be considered, but also AI approaches such as deep learning and natural language processing.

  1. Christian Dyballa, Head of Sector Financial Services - Insights & Data at Capgemini in Germany
    "The greatest challenge in machine learning is the preparation of the data. In addition, the transparency of the machine learning algorithms used and their models is of central importance."
  2. Dr. Frank M. Graeber, Manager Application Engineering and Technical Account Management at MathWorks in Ismaning near Munich
    "The cloud enables companies to use machine learning applications in a comparatively simple manner."
  3. Dr. Kay Knoche, Principal Solution Consultant DACH Decisioning Solutions at Pegasystems GmbH in Munich
    "The promise of machine learning and AI is to restore the personal relationship with the customer that has been lost through the use of call centers and digital channels - the old mom and pop principle."
  4. Farhad Khakzad, expert in the field of analytics and most recently in the role of Head of Risk Analytics in an international technology company
    "In general, it can be said that there is a lack of ideas for practical application examples, that is, use cases, in local companies."
  5. Dr. Dieter Mayr, Digital Services -Vertical Market Solutions at A1 Digital
    "With the help of the Low-Code Platform, companies can quickly develop and try out AI and ML applications. This makes it easy for them to gain initial experience with machine learning."
  6. Paul-Louis Pröve, Consultant Data Analytics, Artificial Intelligence & Blockchain at Lufthansa Industry Solutions AS GmbH in Norderstedt
    "The discussion about ML and AI is currently heavily influenced by marketing statements, according to the motto 'We do machine learning'. But a considerable number of companies must first develop appropriate fields of application."
  7. Dr. Karsten Johannsen, Business Development Executive Artificial Intelligence at Tech Data GmbH & Co. OHG in Munich
    "A good starting point for the topic of AI and machine learning in a company is an easily implementable, exemplary use case. The goal here should not be to solve a specific business problem, but to get a feel for the possibilities of this technology."
  8. Karl Schriek, Head of AI / Leading Machine Learning Engineer at Alexander Thamm GmbH in Munich
    "Like any innovative approach, complex use cases or new AI-based business models require the willingness to take risks - and possibly fail."
  9. Thorsten Kühlmeyer, Head of Business Analytics & Artificial Intelligence / Lead Analytical Insights Center at Telefónica Deutschland
    "First, companies should identify and analyze the problem. Then a use case is derived from it. Only then do you look for the right tool."
  10. Dr. Jürgen Wirtgen, Dataplatform Lead at Microsoft Deutschland GmbH in Munich
    "Artificial intelligence must be transparent. It is important to ensure the balance of the results of calculations and analyzes. This requires extensive tests of processes."

Dieter Mayr, Expert Digital Services - Vertical Market Solutions at digitization specialist A1 Digital, sees another option: "One possibility is to use a user-friendly low-code platform with which AI and ML applications can be quickly developed and tried out. It's an easy way for companies to get started with machine learning. " Ultimately, according to Mayr, it is about "pouring" expert knowledge into models and making it available for AI and ML solutions.

Controversial: the role of transparency

The experts come to partially different assessments on the topics of transparency of algorithms and traceability of the results that machine learning and AI systems determine. From the point of view of Christian Dyballa from Capgemini, it must be clear how decisions are made on the basis of machine learning. "Just think of sensitive areas such as medicine or autonomous driving."

Kay Knoche from Pegasystems gives another example: "It must be clear which predicators, such as age or gender, have an impact on lending." This is the only way to prove that factors such as foreign origin are not taken into account. "Every individual decision must remain comprehensible," said Knoche. There is also another aspect: It should be possible for people to recognize whether a machine learning or AI system has made a decision.

But that is not straightforward to implement, says Karsten Johannsen from Tech Data: "Why and how an AI makes a decision can in principle only be determined with the classic ML methods such as decision trees, other conventional classifiers or even 'flat' neural networks . " With technologies such as deep learning, this is also possible to a limited extent at best.

Therefore, from the perspective of the experts, it must be transparent on which database an AI instance was created. In addition, extensive tests are required, for example of processes and the effect of certain databases and models on results. "Negative effects should be minimized," emphasizes Jürgen Wirtgen from Microsoft.

Complete traceability remains an illusion

However, there will be no comprehensive traceability with machine learning and artificial intelligence, says Farhad Khakzad: "Even decisions made by people are not completely transparent, at least in most cases." Therefore, this could not be a requirement for the use of ML and AI. "Especially in connection with deep learning, this requirement is almost reminiscent of the search for the Holy Grail," notes Khakzad.

You can find a lot more useful information about machine learning and artificial intelligence here in our online special

Karl Schriek from Alexander Thamm comes to a similar assessment. It is becoming apparent that due to the growing complexity of ML solutions and algorithms, transparency is decreasing. His appeal: "The question that we as a society should be concerned with is not: 'How can we make AI solutions more transparent?', But rather: 'To what extent are we willing to rely on non-transparent solutions?'"

Entry into machine learning: everyone at one table

However, companies that want to use machine learning are not only confronted with questions such as traceability and the search for suitable use cases. "In order for projects to be implemented successfully, the IT department has to break away from classic tasks," demands Jürgen Wirtgen from Microsoft, for example. The internal IT specialists should, for example, ensure that the data material is of high quality.

In addition, according to Karsten Johannsen, the cooperation between IT specialists and business decision-makers often leaves something to be desired. In order to optimize the cooperation between IT, specialist departments and management in the context of AI and machine learning projects, Karl Schriek advocates the following model: "A new department will be set up to which employees from all areas belong. They work out use cases together and products based on these technologies. "

To make it easier to get started with ML, Frank M. Graebe has another tip: "It is a good idea to involve the respective domain experts in AI projects at an early stage or to enable them to train machine learning models themselves. " This can at least partially compensate for the lack of data scientists.

Cloud can be a jump starter for AI and ML

Cloud platforms that provide AI and ML services can help with such projects. Providers such as Amazon Web Services, Google and Microsoft are pushing this path for obvious reasons. "Cloud computing is an approach that makes access to AI and machine learning offers easier and thus solves many problems," states Paul-Louis Pröve from Lufthansa Industry Solutions.

AI and machine learning solutions from the cloud can also provide "start-up help" according to Dieter Mayr from A1 Digital: "It is important that companies use smaller projects as learning objects and thus gain experience in the field of ML. Here is an example: The operator of a call center can use machine learning to optimize the planning of shift operations. " Another advantage of this strategy, according to Mayr, is that no major infrastructure investments are necessary. "This aspect is especially important for those responsible for finance. Such a strategy also saves the budget and reduces the workload of the IT department." This allows the IT department to take on the role of a "business enabler" in ML projects.

Like many other participants in the panel of experts, however, Frank Graebe from MathWorks warns of excessive expectations when using machine learning and artificial intelligence from the cloud: "These can be generic approaches that enable quick initial successes, but less for building up your own know-how -how contribute. "

In addition, machine learning from the cloud quickly reaches its limits in complex use cases: "For industry-typical and business model-related requirements, for example, algorithms should cover the needs of the specific individual case," adds Farhad Khakzad. However, it should be borne in mind that developing in-house requires significantly more effort and resources. From his point of view, however, both points are critical success factors for the use of AI and ML in companies.

Don't just focus on quick wins