Analyzing business processes is just the start of the transformation. To improve the efficiency and effectiveness of their operations, organizations need to explore ways to introduce intelligent automation.
But they first need to get their houses in order, and when faced with the complicated reality of their own processes, the best way to achieve this is to apply a series of measures in a defined sequence, starting with the elimination of wasteful tasks before redesigning and automating those that remain.
At Capgemini, we use the Digital Global Enterprise Model (D-GEM) – our proprietary business transformation platform – which encompasses the tools and techniques for reshaping and streamlining processes, helping our clients remain competitive in a rapidly changing, business context (see Figure
5). This, in turn, enables an organization to seamlessly connects processes and people, intelligently, as and when needed to create the Frictionless Enterprise.
However, it’s not only about replacing people with machines (real or virtual) – it’s also about identifying tasks that can be performed better and/or faster with AI, and then applying a proven methodology for automating their work within a given process.
Process optimization also plays a part in deciding between humans and machines. It must be assisted with advanced analytics and modeling in order to assess as precisely as possible the efficiency and the value added by AI.
People or machines?
Our proven D-GEM platform helps to decide between machines and humans while optimizing a business process. It comprises three steps:
- Identifying tasks that can be performed better and/or faster with AI
- Measuring the value that AI can add in the whole process
- Designing human-in-the-loop solutions when the expected efficiency is not reached by machines alone.
So, how should organizations decide which process should be handled by which means?
In order to rationalize the decision, it is important to use the same metrics to compare the performance of people and machines. Quantitative measures that are also often probabilistic are used to characterize machine learning algorithms – for instance, metrics such as confusion matrices – can also characterize human performances, making it more straightforward to identify tasks that can be performed
better with AI.
Then it is also important to measure the value that AI might add in the whole process, and not only for one task. We just looked at process mining as a means of revealing the truth about process flows. When data scientists and business experts use this real-world information to redesign processes before automating them, their first step in our three-stage methodology is to make decisions between machines and humans for each element of a task by characterizing human operators in the same way as by machine learning, using the same metrics for both.
The second step is to check than the whole process is efficient. For this, we need to carry out advanced quantitative modelling of the business process in order to simulate numerically all the possible scenarios, and to assess as accurately as possible the potential value that might be
added by AI. Sometimes, a machine that is less efficient but faster than a human can bring value; this quick-and-dirty effect can only be revealed by modeling the whole business process. It depends on the position of the node within the graph representing the business process.
Finally, in some cases, these two steps cannot reach the client’s expectation in terms of efficiency. The third and final stage of our methodology is to keep humans in the loop for performing such a task. For this, an AI-based operator should be added in order to orchestrate operations between machines and humans. Such an AI orchestrator is also characterized by a confusion matrix that takes into account all the possible outcomes.
In order to illustrate the confusion matrix that can be used for characterizing machines and human performances, we focus here on the case of the AI orchestrator.
This matrix provides a probabilistic measure of all the possible outcomes associated with the AI orchestrator that is based on machine learning algorithms. Such algorithms have learned to predict operations that can be totally managed by machines and those that can be handled only by human operators.
In this case, the confusion matrix is a 2 x 2 matrix. Each element of this matrix has its meaning (see Figure 6). The elements in green correspond to the true classification rates, the efficiency with which the orchestrator is doing the job. The other elements in red are the error rates with which the AI orchestrator misclassifies machines (human) operations as human (machines) operations.
The quantitative modelling of this scheme is based on this kind of matrix and enables the expected cost per operation and the efficiency of the process to be calculated. We can demonstrate that this simple human-in-the-loop design is better than machine only, at a cost that is lower than person-only.1
Figure 6. The confusion matrix for an AI orchestrator between machines and humans
“Our proven D-GEM platform helps to decide between machines and humans while optimizing a business process.”
Principal AI Scientist, Capgemini’s Business Services