Walking the Way from Process Mining to AI-Driven Process Optimization
Download as PPTX, PDF1 like1,087 views
The document discusses various organizational and technical pitfalls in using prescriptive and predictive models for process optimization, highlighting issues such as lack of buy-in, insufficient data, and poor validation practices. It emphasizes the importance of pilot testing and the incremental adoption of process optimization capabilities to derive sustained value. Tools like Kairos are presented as solutions for prescriptive monitoring of business processes, enabling better decision-making and resource allocation.
Walking the Way from Process Mining to AI-Driven Process Optimization
- 1. ICPM 2023 – Rome – October 25, 2023
- 9. Organizational Pitfalls Lack of buy-in from operations Lack of trust in predictions Prescriptions not linked with actions Lack of processes to validate, monitor, and maintain prescriptive models Technical Pitfalls Insufficient data availability & quality Lack of uncertainty modeling Drifts and out-of-distribution predictions Correlation ≠ Causation
- 10. No amount of organizational and technical readiness will save you from the sin of using simulation models without pilot testing or predictive models without A/B testing 0
- 11. Results achieved: Even though the prediction of the most likely customers to report incorrectly was reasonably accurate, the intervention did not have a preventive effect. No root cause was identified to explain why the intervention did not have the desired effect.
- 13. +/- - - + Kairos: A Tool for Prescriptive Monitoring of Business Processes (demo) https://kairos.cloud.ut.ee/
- 14. - Automate tasks X and Y - Add +5 resources to task A - Remove -2 resources from task B …
- 16. Everywhere! GenAI brings context recognition: What types of processes, activities, KPIs are we talking about? LLMs enable conversational process optimization across all layers of the pyramid Descriptive Process Mining •Where are the sources of waiting time? •Where are the rework loops? •Where are we over-processing? •What are the sources of variance? •Which cases require the most touches? •Where are we violating our KPIs? •Are we abiding to our business rules and policies? •… Predictive Process Optimization •By how much would we reduce order-to-delivery times is we: •Shorten inter-batch cycles? •Move resources to packaging? •Automate verification steps? •By how much would we reduce costs if we: •Consolidate touchpoints? •Reorder verification steps to reduce over-processing? •Reduce rework rates? •… Assisted Process Optimization •How can we slash KPI violation rates? •What are the practices of the teams with highest performance? •What should we change to reduce the number of touches? •Which checks should we add to reduce compliance violations? •How can we cut the cycle time at constant cost? •How to allocate resources to optimize time at constant capacity? •…
- 17. Faster or Cheaper Horses • Structured data prep (ETL scripting) • Data querying (if your tool requires SQL coding) • Data synthesis / enhancement (simulation?) Better Than (Hallucinating) Horses • Preprocess unstructured data • Explain (or mis-explain) findings, patterns • Figure out which options/questions to consider
- 19. Organizational Pitfalls Lack of buy-in from operations Lack of trust in suggestions / prescriptions Prescription flooding / over-prescription Prescriptions not linked with actions Lack of processes to validate, monitor, and maintain prescriptive models Technical Pitfalls Insufficient data availability & quality Neglecting inter-process dependencies Lack of uncertainty modeling Drifts and out-of-distribution predictions Unreliable prescriptions Lack of feedback loop
- 20. No amount of organizational and technical readiness will save you from the sin of using AI-driven improvement recommendations without validation and pilot testing or prescriptive models without A/B testing 0
- 24. No amount of organizational and technical readiness will save you from the sin of deploying proactive or adaptive process optimization without building a capability to derive sustained value from process mining, predictive, and prescriptive optimization 4
- 26. • Lay the foundations, start climbing, keep climbing, don't hold off Getting data for process mining is often a challenge. But there are both short- term benefits (bottom of the pyramid) and long-term ones (top) • Don't skip the layers The lower layers of the pyramid provide a foundation to draw business value from the upper layers. • Align strategically and build governance incrementally Apply these capabilities first and foremost to processes that matter. Adopt these capabilities incrementally, one process at a time. Build success stories internally, ensure each layer of the pyramid yields value.
- 27. Want to know more?
Editor's Notes
- #7: https://github.com/AutomatedProcessImprovement/Simod
- #8: Lost Make example of risk objectives
- #15: https://www.if4it.com/core-domain-knowledge-critical-foundation-successful-design-thinking/ https://towardsdatascience.com/minimum-viable-domain-knowledge-in-data-science-5be7bc99eca9
- #19: https://github.com/AutomatedProcessImprovement/Simod
- #23: https://github.com/AutomatedProcessImprovement/Simod
- #26: Building an analytics engine to support different automation and enterprise platforms Why are we having this SLA violation? (example of conversational bot used for diagnostics) What if demand increase by 20%? (example of conversational bot used for prediction) What intervention policies (same) For each question, the bot will automatically do an AB test of the improvement intervention/compare cases with and without the issue to show the impact and identify root cases
- #27: Lay the foundations, start climbing, keep climbing, don't hold off. Many managers postpone the adoption of process mining by stating “we don’t have the data”, or “our data is not good enough”. Yes, getting the data for process mining is often a challenge. But the benefits have been demonstrated repeatedly, in thousands of successful deployments. And getting the data to do process mining opens many doors. The data that is used today for process mining can be used tomorrow for predictive process monitoring or to build digital process twins. Once the obstacle of data collection and curation has been overcome, the possibilities are endless. Note that task mining provides an additional channel for collecting data— when the enterprise system does not allow us to do so. Don't skip the layers. The lower layers of the Augmented BPM pyramid provide a foundation to derive business value from the upper layers. Organizations that wish to maximize the benefits of adopting the upper-layer capabilities need to master the lower layers. Align strategically and build governance incrementally. Any process mining, predictive monitoring, or prescriptive process improvement initiative needs to be grounded on the strategic priorities of the organization. The capabilities in the augmented BPM pyramid should first and foremost be applied to business processes that matter to the organization. It is also important to adopt these technologies incrementally, one process at a time. Over time, a governance structure is needed to ensure that the technologies in the pyramid create value predictably and repeatably. But before getting there, it is important to have a few success stories internally, to gain executive support, and to keep this support by showing that every capability in the augmented BPM pyramid produces tangible value.