The document proposes a framework that combines process mining and causal machine learning to discover causal rules from event logs. It identifies candidate treatments using action rule mining on event log data. It then uses uplift trees to discover subgroups where a treatment has a high causal effect on outcomes, addressing confounding. The approach was tested on a loan application dataset, discovering 8 causal rules that recommend process changes to increase loan selections. Future work includes incorporating other recommendation types and addressing unobserved confounding.