Hardly any topic is given such prominence today as the use of cognitive solutions. There are many terms that are used interchangeably: Artificial Intelligence, Machine Learning, Robotic Process Automation (RPA), Deep Learning, Natural Language Processing, Predictive Analytics. The variety of these terms as well as the unclear definition and demarcation lead to confusion about what cognitive solutions are and how they can be used most meaningfully.
Cognitive solutions can be defined as applications trained by using self-learning systems and algorithms to perform activities and process steps previously performed by manual human work. In other words, cognitive solutions are applications that automatically perform what people still do manually today. They can make optimal decisions and make recommendations in critical situations, work much faster and are available around the clock.
To further analyze the potential uses of cognitive solutions, a classification and differentiation of the different types of cognitive solutions is helpful. Therefore, it is worth considering the process complexity and the form of the data input. The process complexity can be defined by the number of process steps, the set of influencing parameters or the number of process participants. Data input determines how the data is available, structured and classified data in databases are easier to handle than unstructured data in free form. Assembling these two parameters, the process complexity and the data input type, results in a classic 3×3 matrix for the classification of cognitive solutions. The following graphic describes the classification of cognitive solutions.
In cases in which simple process complexity is involved and the data is structured and defined, RPA systems are used. These solutions, like macros in MS Excel, can be programmed to automatically perform operations that were previously handled manually by people using computers. RPA can help organizations across multiple industries and areas to automate a variety of tasks, from patient health care registration to updating the customer profile in the financial services industry. The advantages are mainly the speed of execution, the avoidance of manual errors and the 24/7 availability of such systems. Thus, RPA-based solutions bring considerable time and cost savings. However, today they are still unable to fully respond to changes or inaccuracies. For example, if a new previously unknown data field is inserted, the preprogrammed system will face an error and can only be restarted by reprogramming or similar human intervention.
RPA differs from Machine Learning systems. On the one hand, Machine Learning systems are more suitable for carrying out more complex activities and can also work with unstructured data. They usually access other databases or programs to solve complex problems. Prominent applications that run based on Machine Learning are assistance systems such as Alexa, Siri or Cortana. If Alexa is asked about the weather, the application knows where to look to find the information needed. Such systems are also able to learn steadily from user behavior. WhatsApp uses self-learning in the suggestion words that appear when typing in messages. These suggestion words are based on a standard dictionary. Nevertheless, WhatsApp also stores the words that users regularly use in their individual messages.
Systems that use Artificial Intelligence go one step further. With AlphaGo, Google has developed a program that was able to beat the world champion in the complex Chinese board game Go. In doing so, the application has independently mastered the board game by playing the game millions of times. Artificially intelligent systems are not yet able to accommodate human capabilities such as anticipation, creativity and contextualization in all areas. Nevertheless, the development of computing power as well as the availability of data is increasing so rapidly that it is only a matter of time before artificially intelligent systems can independently learn and solve even more complex problems.
After classifying the different types of cognitive solutions, let´s highlight those areas that are particularly suited to the use of such systems in Procurement and in the Supply Chain. The following diagram gives an overview:
The areas of Analytics & Reporting in Procurement are particularly suitable for the use of Cognitive Solutions. In many cases today the reports are elaborately and manually created. The topics of Predictive Analytics and Big Data are already topics that are tackled heavily with artificially intelligent systems. Thus, relevant data is compiled very compact in real time, forecasts for future events can be made and prepared reports are presented on all relevant KPIs. In the Help Desk and Support area, Chatbot solutions are very much in trend. The biggest hope lies in the vision that the largest proportion of user inquiries can be reduced by using intelligent bots. For example, Alibaba has introduced an artificial intelligence Chatbot that companies can customize and train to improve customer service. General inquiries by e-mail and by telephone can be heavily reduced. IBM Supply Chain Insights leverages Supply Chain-driven Watson technologies to provide a complete visibility into the entire Supply Chain. This can be used to predict risks and disruptions in the Supply Chain and to improve processes. If one puts these application areas and the previously defined classification on top of each other, then the following picture emerges:
Cognitive solutions such as Machine Learning, Chatbots and Artificial Intelligence have reached a peak of expectations. However, according to Gartner’s latest Hype Cycle for Emerging Technologies 2017 report, most of the solutions are still far way, at least 5-10 years before they reach mass usage.
Cognitive solutions today already influence the areas of Procurement and the Supply Chain and open up new possibilities for automation. However, Predictive Analysis methods and logistics optimizations are not yet fully illustrated. The higher the complexity of the processes and the more unstructured the data entry types, the more complex is the use of cognitive solutions.
The following concrete recommendations are proposed by way of conclusion:
- Start with the automation of simple, repetitive tasks and processes using RPA solutions. These are already well developed today and can be used in the areas of Reporting & Analytics, Order Fulfillment and Master Data Management.
- Chatbot and virtual assistant systems have also matured; their use can improve end user and vendor support.
- More caution is needed on more complex topics such as Machine Learning and Artificial Intelligence solutions. Here, the hype has caused a flood of marketing. Even if you remain skeptical and cautious, initial experience in the form of pilot projects and development partnerships can prove very effective.
If you would like to learn more about this topic, please contact me.