Digital transformation is about doing things differently, not just doing them better than everyone else. And because I come from the world of business process management technology development, I have a few thoughts about how bringing artificial intelligence into business process management can help you achieve both sides of this coin.
Many business process management (BPM) projects are aimed at cost reduction and efficiency improvement. This operational improvement approach is important, because doing things better is, well, better. But simply improving operational effectiveness does not necessarily provide strategic competitive advantage.
Artificial intelligence (AI) is taking its place in both operational and strategic BPM, and here is where we are going to see some interesting advances that lead to both “better” and “different” ways of doing things.
Using Artificial Intelligence in BPM
Here are three use cases illustrating how AI could be applied to BPM.
Case 1: “Intelligent” robotic process automation (RPA) could be used to observe, and learn, what people are actually doing, and then automate those patterns so people don’t have to do error-prone redundant work.
Case 2: Machine learning could be applied during process execution to, for example, trigger a new process or reroute running processes according to predictions.
Case 3: Machine learning can also make recommendations — for a next best action, for example.
These three examples depend on unique data generated by each process to make predictions.
According to Wil van der Aalst, a professor in the department of mathematics and computer science at the Eindhoven University of Technology in the Netherlands, classic data mining techniques do not work for processes, because processes express behavior that is far too complex, with concurrency, loops, decisions and so on.
There is another area where AI can be applied using a more generic approach, independent of data. As process-based applications get more sophisticated, the effect of delays — or blocked business processes — remains a problem. Delays can be the result of human error, resource unavailability, peak workloads and external dependencies, among other factors. So it is possible to look at the process itself for prediction.
How can we use data about a process itself to facilitate intelligent continuous improvement of process-based apps?
Related Article: Is Robotic Process Automation Finally Here?
Choosing an Appropriate Model: Failures and Successes
The model that is needed has a dual objective: to allow predictions (duration, path, data) and recommendations of improvements on a process-based application. Further, for processes, we need a model that does not depend on existing data.
Although linear regression has the advantages of being easy to model and use, it requires good data! Linear regression won’t work on processes.
Neural networks use cool algorithms to learn from observation but are rather “black-box-like” and hard to retrain when data changes, and it’s difficult to use the results to make recommendations.
Consider techniques for process mining instead.
Conformance process mining is used when there is an a priori model. The existing model is compared with the process event logs; discrepancies between the logs and the model are analyzed. This model can be enhanced to make predictions. There is an a priori model, but it is not used to check conformance; instead, it is used to improve the existing model — to enable it to predict process duration, for example.
Enhanced process mining offers key advantages for process AI. There’s no need to know the data in advance — it’s relatively easy to understand and interpret. The same algorithm can be used for prediction and recommendation, and it can be extended for further prediction use cases.
So in short, an appropriate process mining algorithm for prediction can apply to any process model, and use the unique data stored in its particular event logs to enhance the process model (using time stamps on events, calculating time remaining based on deadlines, computing average completion times, etc.).
Related Article: Robotic Process Automation's Reach Expands in the Enterprise
A Straightforward Example: An SLA Monitoring App
One goal for AI applied to the process behind an app might be to predict when an organization may not be able to meet the terms of a time-sensitive service level agreement (SLA). The person responsible could be alerted when a deliverable might be late, because a step in the process is flagged as taking too long.
AI permits applications to analyze their own historical patterns and flag future constraints that are not easily perceived by a human observer.
And when some manual steps in the process might be found to be systematically taking longer than anticipated, user interface analysis data could be a handy way to improve predictions and recommendations.
Ideally, the applied AI can recommend both immediate corrective actions for current applications and suggest redesigns for future applications.
Intelligent Continuous Improvement of Running Applications: Current Developments
AI applied to process applications helps to anticipate future constraints and align them with available resources — that is, become proactive rather than reactive — and thus avoid bottlenecks and delays.
We can use process-flow pattern detection and process and business metrics predictions to guide corrective actions and updates of running applications. Applications can analyze their own historical patterns and flag future constraints that are not easily perceived by a human observer.
Some advanced decision automation capabilities can be based on both process-mining algorithms and machine learning techniques. Business processes execution can also be analyzed, so certain behaviors can be predicted. AI can be used in this way to make recommendations for automated or human decision-making.
AI-Enhanced BPM Has a Central Role in Digital Transformation
There’s a bright future ahead for BPM and AI. AI is a powerful technology that empowers human intelligence to continually improve user interactions and process execution.
BPM process-based applications already make back-office processes smooth and satisfying to end users. Digital user experiences for customers and employees are more and more personalized with customized user interfaces. And human decisions will be enhanced with artificial intelligence for intelligent ongoing improvements.
Looking forward, AI will be instrumental in improving the user experience with BPM applications, by applying other AI techniques on user interactions and browser uses, and by analyzing page flow. As applications provide better and better overall user experience, we’ll be seeing both better operations and better strategic differentiation for the organizations that employ them.