Process Mining and Process Monitoring
Scope
Deriving process insight from event logs and monitoring running processes.
Core rules
- Event logs need at least case ID, activity and timestamp, plus optionally resource and attributes. From these come process discovery (a model from data), conformance checking (log against model) and enhancement (repair or extend a model using data).
- Algorithms differ: Alpha miner handles concurrency poorly, Heuristics miner tolerates noise, Inductive miner produces block-structured sound models, Fuzzy miner produces simplified maps.
- Discovered-model quality dimensions: fitness (how much of the log the model explains), precision (how over-permissive it is), generalisation and simplicity. These trade off against each other.
- Variants are distinct execution paths in the log; variant frequency sets improvement priorities.
- Monitoring: process dashboards and cockpits, KPIs near real time, thresholds and alerts, business activity monitoring, and predictive monitoring using remaining-time models.
Worked example
A log of 10,000 purchase orders shows 42 variants, the top one covering 61%. Conformance checking shows 8% of cases skip the 'check budget' activity. The analyst does not declare non-compliance but investigates: orders below a budget threshold may legitimately skip the step, in which case the model needs updating rather than the organisation needing discipline. Reasoning: deviations can be model error, log error or genuine exception.
Traps that cost marks
Reading a mined model as causal truth, when it shows observed order only. Confusing fitness with precision. Ignoring data quality problems such as incomplete logs, batch timestamps and missing case IDs.