Technology: the Motor for the Super Power of AI

Many people associate artificial intelligence with expensive high-performance computing and complex algorithms. But when you have the right data and technologies, it's not really that hard to get started. Especially when gaining initial experience with simple use cases and platforms. 

Sabrina is a production manager and aims to prevent the frequent accidents in the aisles of her factory. To this end, she has cameras installed that show whether an object is in the way. In order to avoid personnel expenses, an automatic system is to alert the respective shift supervisor by means of an alert tone and a red warning light. 

The task sounds simple, but—as is so often the case—the devil is in the detail. Because she needs to clarify: 

  • Which technologies are best suited and which ones are available through IT and cover the needs of this use case? 
  • How can the solution be integrated to derive added value from the AI use case and link the data to the alert system? 
  • Which data are necessary for the AI system to differentiate between objects that are in the way and mobile transport vehicles or people?  

The Right Technology for the Use Case

To identify the right technology for their use case, organizations should assess the application requirements based on two categories: available expertise and customization options. Below, we present four levels of complexity, which help select the right technologies. 

Complexity level 1 – Integrated platform components

Various prebuilt and pretrained AI solutions are available for the applications and workflows, depending on the citizen development platform used. These include, for example, AI Builder, a Microsoft Power Platform capability, or Cognitive Services interfaces in Power BI. These interfaces empower users („citizen developers“), to create valuable applications including AI components with just a few clicks. They can be used, for example, for the design of form scanners that can read out information via a cell phone camera, or translate text components.  

These options already enable Sabrina to implement an initial proof of concept of her use case in order to identify potential obstacles to the final solution.  

Complexity level 2 – Out-of-the-box services

The platform also allows the integration of other services (outside the platform) during application development. They are, for instance, pre-trained for a specific area of application or their data processing may have superior scalability when requests increase. These services are available from major cloud providers such as Microsoft in the form of Azure Applied AI Services. However, organizations can also use Open AI interfaces with their large-scale GPT-3 language model. The important aspect here is that these interfaces are outside the actual application platform and therefore require professional developers (pro devs). Especially when used extensively, for instance in applications managing several thousand cues per day, this needs to be considered early in the application design. Another aspect to be clarified is whether the AI should perform task-specific activities.  

Sabrina's use case has an extremely high data flow due to the continuous stream of data. Additionally, the videos to be processed depend on the different cameras. Hence, models for the different cameras and a review of the processing quality of the data volume need to be evaluated in this case. 

Complexity level 3 – Low-code/no-code self-taught machine learning models

The focus in complexity levels 1 and 2 is on application development and the use of provided technologies and services. Should these AI solutions no longer be accurate enough, an organization will need to look at developing their own AI (or machine learning) solution. In this regard, cloud providers offer integrated platforms to create powerful applications that are tailored to a specific use case and require little prior coding experience or mathematical knowledge. Nevertheless, a data engineer needs to be involved in the development of the application in addition to pro devs. The data engineer develops a usable machine learning model on a platform such as Microsoft's Azure Machine Learning Services using either the drag-and-drop features of ML Studio or the auto-ML capabilities. Platforms such as Power BI also offer the option of integrating machine learning models or auto-ML solutions via data flows. These services are generally used when data is already available in an integrable repository such as a data lake or data warehouse. This makes it easy to quickly create business-specific forecasting models—for example with regard to sales forecasts or machine downtimes—and integrate them into applications. 

In Sabrina's use case, this approach provides the opportunity to develop tailored models that take into account the background of specific cameras. Proof of concept serves to test and evaluate the application under real-world conditions. This allows challenges for the productive deployment of this solution to be identified and addressed in a custom data pipeline. 

Complexity level 4 – Custom data pipelines with machine learning models

The implementation at complexity level 3 allows to understand the accuracy of a prediction under real-world conditions. This provide new insights, such as insufficient data processing, or the realization that the AI generates incorrect forecasts more often than should be the case. In this case, data scientists need to be involved in application development in addition to pro devs and data engineers. Organizations should choose a platform that allows for efficient collaboration as well as flexibility and experimentation. State-of-the-art platforms are designed to deliver exactly that. Azure Synapse Analytics and Azure Databricks are particularly suitable for this purpose, but Azure Machine Learning Services also offer many helpful components. These platforms are particularly well-suited in two cases:  

  1. When the same data source is to be used for multiple AI use cases, for example, when using production data for management reporting as well as predictive maintenance; 
  2. When data is very raw and preprocessing is required for an AI application, e.g. when using production data from different machines that need to be harmonized and linked using shared keys. 

Sabrina's use case will require implementation at a complexity level 4 in the medium term to optimize the efficiency of the AI solutions for the various cameras. This may be the case if the hazard potential of the identified objects is to play a role in the decision to trigger the light signal. 

Technology integration

However, anyone dealing with AI technology must not ignore the important aspect of interfaces. They are, in fact, the key to success, because tools and data per se do not achieve added value. They must be used for the right use cases and integrated into the relevant organizational processes. 

It is therefore necessary to clarify how the technologies can be linked with applications and workflows to achieve a smooth overall process. For example: The data generated by AI usually is also required by other systems such as CRM or marketing solutions. In many cases, Azure provides a suitable platform for this, as AI solutions can be integrated smoothly into Teams, Office or other Microsoft applications. 

In our above example, this means linking object recognition and the alert system for the shift supervisor. Here, the AI-calculated top alert level must result in the red light lighting up and an audible alert needs to be activated. To achieve this goal, the compatibility of the two systems needs to be verified. 

In addition, the use case needs to be embedded in the organizational processes. It needs to be specified what actions on the part of the shift supervisor are required to remove the object lying in the way. Will she do it herself or delegate the task to a safety officer? Is a log entry required or does the management need to be informed? 

These questions, in turn, may result in additional processes. If, for example, objects being in the way is a particularly frequent occurrence, production processes may have to be modified or safety precautions taken to mitigate such hazards. In the case of human error, training may be required to improve the handling of materials or tools. 

A data strategy is indispensable

AI solutions also need to be based on reliable and correct data to provide correct results. To this end, organizations should develop a comprehensive data strategy. This strategy serves to achieve the desired target state of the organization and of the IT platform, as well as provide a road map with appropriate guide rails. How this is done is explained in Part 2 of our article series. 

Becoming successful faster

IT creates the preconditions for the departments to become successful faster. This requires putting in place appropriate licensing agreements by using technology that optimizes the integrity between tools. Supported by governance rules and best practices, Sabrina can quickly start implementing the solution so that it will eventually be used in a productive manner. 

Sabrina could not have taken these preparatory steps on her own, as most of them require strategic decisions on the part of the management. Once these have been taken, Sabrina can implement her small use case without delay. She knows the technologies that are available, how to use them, and what to look for.  

At complexity level 2, she can evaluate initial models, at level 3, she can conduct a realistic proof of concept, and at level 4, she can implement the solution to make sure that significantly fewer accidents will happen in her factory aisles. 


Companies should now deal with the questions about IT technologies, i.e. create a database and develop a data strategy as well as define relevant use cases for the use of AI and integrate them into business processes. In addition, it is important to involve experts in all development steps and, if necessary, to carry out training courses. This will allow them to offer the right processes and licenses up front, before requests like Sabrina´s are submitted. Because one thing is clear: AI is becoming increasingly important and will be put to widespread use in the near future. The IT side needs to play an active role and enable digital, data-driven solutions, because the super power of AI is driven by data. 


Ingo Meironke

Innovation Manager & Co-Lead Future Operations

Trutz-Sebastian Stephani

Senior Manager | Head of Data & AI

Ramón Roales-Welsch

Business Lead Data & AI