How can Artificial Intelligence layer on top of Process Mining techniques to provide substantial profits to businesses.
$300 MM losses in operative risk control events and $54,157MM in unusual fluctuations were identified in the financial industry according to a study based on the 2017 ALGO FIRST report by IBM .
The author of this article believes many of these cases could have been previously identified and foreseen applying Artificial Intelligence models on Process Mining techniques, saving millions to the industry and creating new earning opportunities.
So what is “Process Mining” ?
Process mining consists in a set of techniques to extract insights from event logs that are generated by any Enterprise controlling system (customer care, CRM, ERP, HelpDesk, Digital Analytics) in order to discover, model, monitor and optimize those processes  .
Process analytics is an emerging field that combines several disciplines. Starting from the Business oriented cases, such as Business Process Management (BPM) in its process modeling capacities that describes, documents and explains them .
There are several process modeling techniques and many of them are well known from years past such as Petri Nets of Gantt Diagrams. However the recent advances in modeling languages such as UML or BPMN provide bigger functional capacities that allow the modeling of complex structures. 
This is how process mining goes far beyond mere process modeling. Combining aspects of Business Intelligence and Analytics (BIA) it surpasses the limits of abstract process modeling from the information extracted in the event logs.
In this context, data integration and ETL (Extraction, transformation and load) can be used by process analytics in order to integrate several data sources along the end-to-end process.
In a similar manner process mining can use data mining techniques and the analytics visualization capacities in order to create useful insights and predictive and optimal models with direct impact on Business models.
Process mining is also closely related with some other techniques such as Business Activity Modeling, Business Operation Management, Business Process Intelligence and Data Mining.
How can Process Mining benefit Corporations?
Among the main benefits of process mining we can outline:
- Process map & Discovery: From event logs the process mining techniques allow describing and modeling the actual processes that are being run in the company.
- Conformance checking: Having precise documented trails of the process executions allow identifying their conformance with the established compliant procedures.
- Throughput time and bottleneck detection: It’s not only possible to model the sequence of tasks but also to determine the intensity of execution and the throughput and overhead times, simulating, predicting and identifying bottlenecks
Log analysis can be executed either once processes are completed (thus one would be analyzing post-mortem data) or in real time (thus pre-mortem data). for Operative support functions .
With process mining techniques one can thus:
- Improve time related KPIs: minimize throughput times, overhead time, response times, maximize the % of workload done in a time interval, etc.
- Improve cost or quality related KPIs
- Redesign (improve) processes by promoting structural changes in processes based on insights (ex, adding more control units, working tasks in parallel, …)
- Adjust processes with non-structural smaller changes, for example, modifying the number of resources given small fluctuations in the number of cases or workload.
- Provide operational support exploiting online pre-mortem data systematically in order to, for example, recommend the task that would minimize throughput time.
How does Artificial Intelligence add to Process mining?
The opportunities that rise form applying AI techniques, both as machine learning or as in its deeper counter part neural Networks are many and provide multiple benefits in a variety of business cases. Enriching log events to customer and Business intelligence data provides bigger datasets to model, increasing the sets of explanatory variables and the models’ sturdiness.
Clustering and segmentation:
Using machine-learning clustering algorithms such as “k-means nearest neighbors “ it is possible to identify customer segments, task, cases that can be grouped together resulting in specific domains that can predict specific characteristics to these segments.
- Friction points and bottlenecks prediction. It is possible to establish predictive models that can foresee if, given a specific situation, new friction points and / or bottlenecks could arise in order to prevent them.
- Next task prediction: Which task will be executed next? Models can predict with high accuracy, using Random Forest algorithms  or LSTM deep learning networks which will the task with the biggest probability to be executed. For instance the possibility of the customer reaching the final checkout page in an eCommerce, determining beforehand the real interest of the customer in shopping based on his/her behavior and previous visits (or whether he/she is just window shopping)
- Processes final state determination: ¿How will the process end? Using classification algorithms it is possible determining in which of the possible final tasks /states will the process end based on the process own variables, such as task times, throughput time, loops, bifurcations, … and /or business’ own variables (resources, customers, products, segments, price, channels)
- Resource allocation: How many resources will the process need? Thanks to regression analysis it is possible to determine the number of resources needed to execute certain tasks and so reduce the task and overhead times below bottleneck thresholds.
- Process variables dependency explanation. The use of these procedures serves to identify which variables of the process are the ones who have a bigger impact in the final outcome , allowing the businesses to act on them in order for better business results.
- The model recommends a process bifurcation: Given a certain scenario and a forecast, the system can recommend a fork in the process, the next best task to execute for a most efficient process
- The model recommends adding / removing resources to a task:knowing where and when bottlenecks can be produced, the system can recommend assigning new resource to a task (or removing non productive resources reducing the overall cost)
Direct measurable benefits
As a recent poll published by Accenture  states, the application of machine learning is already providing its fruits: 88% of companies that use it improved in over 200% in their Business process main KPIs; however only up to 9% of them use AI’s full potential
As this poll describes, 34% of the Business who participated in this study totally agree in that their new processes are showing hidden value in its data, which provides them with better decision-making and the offering of new products and services. Moreover, 82% of them assure that processes with machine learning helps the find Solutions to unseen problems, thanks to data that was not available until then.
We’ve seen there are huge benefits for corporations, and equivalently big market opportunities, in the synergic mix of these two disciplines: Process Mining and Artificial intelligence. Little has be done so far even if both are in pure booming state regarding investment and growth expectations and so we expect to see, in the near future, a new bunch of start up companies who will take on this high profitable opportunity
Credits and References
 This Article is a summary of an extract of the Master’s final study by the AlphAI team (Raul Alcubierre, Ignacio Chaparro , Manuel Gutierrez de Diego, and the author Ignacio Martí Carrera) for the “Artificial Intelligence and Deep Learning Executive Program” at the Escuela de Organización Industrial, Madrid, Spain.
 IBM Algo FIRST https://www.ibm.com/us-en/marketplace/ibm-algo-first
 Van der Aalst, Wil. et al. (2011): Process Mining Manifesto, http://link.springer.com/chapter/10.1007/978-3-642-28108-2_19
 To learn more about process mining please refer to the “Process Mining: Data science in Action” MOOC by the Eindhoven University of Technology from where most of the insights about process mining techniques described in this paper were extracted.
 “Process mining on the loan application process of a Dutch Financial Institute. BPI Challenge 2017” Liese Blevi, Lucie Delporte, Julie Robbrecht KPMG Technology Advisory, Bourgetlaan 40, 1130 Brussels, Belgium 
 “Process, reimagined” Accenture