Event Trace Process Mining¶
Process-mining tool — instantiates Funnel Attrition Localization
Reconstructs the actual paths people took from raw event logs, exposing the loops, skips, back-steps, and side-routes that a clean linear funnel silently assumes away.
Event Trace Process Mining is software that reads timestamped event logs and rebuilds the real process graph — which sequences actually occurred, and how often — instead of assuming the idealized funnel. Its defining move is to treat the funnel itself as a hypothesis to be checked against behavior. It is the only sibling that questions whether the process is even the straight chain everyone drew, surfacing reentry, loops, and alternate routes that make a single stage-to-stage drop-off number meaningless. Where the other mechanisms presume the ordered stages are correct and measure within them, this one asks whether those stages describe what people do.
Example¶
An insurer's assumed claims funnel is submit → review → approve → pay. Mining the event log tells a different story: roughly a third of claims loop review → request-more-info → resubmit two or more times before approval, some ≈8% skip straight from submit to a fast-track payout, and a slice bounces between review and appeals. The review → approve conversion everyone had been quoting was fiction — many claims pass through review three separate times, so the "population at review" was being counted repeatedly. The tool renders the real, tangled graph and quantifies each path's volume and where the reentry concentrates. The team learns the loss they were chasing is not a single stage at all but a rework loop, and re-models the funnel to account for it before measuring drop-off again.
How it works¶
Ingest logs keyed by case identifier and timestamp; discover the directed graph of observed transitions with their frequencies; detect loops (a stage revisited), skips (a stage bypassed), and branches (divergent routes). Then compare the discovered graph against the assumed funnel and quantify how much volume deviates. The output is twofold: a map of what actually happens, and a forced policy question — do loops and reentry count as attrition, as progress, or as churn? What sets it apart is that it derives the process from behavior rather than accepting a stage model as given.
Tuning parameters¶
- Path-frequency threshold — how rare a route must be before it is pruned as noise; low keeps every quirk (unreadable spaghetti), high shows only the trunk (and may hide a costly minority loop).
- Case / identity key — what defines a single case traversing the process; the wrong grain fabricates or erases loops outright.
- Loop treatment — whether a revisit counts as reentry (the same journey continuing), a fresh attempt, or a failure; this changes what "conversion" even means.
- Time-gap cutoff — how long a pause before the same actor's later events are treated as a new case rather than a continuation; it governs where journeys are cut.
- Conformance strictness — how tightly discovered paths must match the assumed model to be called conformant; tighter surfaces more deviations to explain.
When it helps, and when it misleads¶
Its strength is exposing the single most damaging hidden assumption in attrition analysis — that the funnel is linear — and localizing loss to loops and rework rather than a phantom stage. It rests on the established discipline of process mining and its conformance-checking[n1] technique, which measures exactly where reality departs from the intended model. Its failure modes come from the same richness: real logs mine into unreadable spaghetti, and it is easy to over-fit a narrative to noise, or to mistake a gap in instrumentation (a missing event) for a genuine skip. It needs clean case identifiers; without them, loops are artifacts. Its classic misuse is mining until some path pattern confirms a prior belief and then presenting that route as "what users do." The discipline is to set the frequency threshold and case key before interpreting, validate surprising paths against the raw log, and separate missing-instrumentation from real behavior.
How it implements the components¶
branch_path_model— it discovers and quantifies the divergent routes and alternate paths the population actually takes, replacing the single-line funnel with the real branching graph.leakage_reentry_and_loop_policy— it detects loops, back-steps, and reentry and forces the explicit call on whether each counts as attrition, rework, or progress.
It does not define the intended stages it measures against (ordered_stage_model, stage_entry_and_exit_criteria — Event Instrumentation Specification), nor tally clean per-stage conversion for a funnel already known to be linear (per_stage_conversion_and_loss_metrics — Conversion Funnel Dashboard).
Related¶
- Instantiates: Funnel Attrition Localization — it tests the archetype's core assumption that the process is an ordered, denominator-preserving chain.
- Consumes: Event Instrumentation Specification supplies the logged events and the assumed model the discovered paths are checked against.
- Sibling mechanisms: Event Instrumentation Specification · Stage Drop-Off Waterfall · Conversion Funnel Dashboard · Cohort Transition Table · Denominator Reconciliation Checklist · Segment Funnel Comparison · Loss Pareto Review · Stage Conversion Anomaly Alert · Survivorship Bias Audit · Funnel Experiment Backlog
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Event Trace Process Mining operates as a computation, comparison, model, or analytic representation used to infer, estimate, or choose because it reconstructs the actual paths people took from raw event logs, exposing the loops, skips, back-steps, and side-routes that a clean linear funnel silently assumes away.
Independent corroboration: The frozen evidence defines Event Trace Process Mining as 'Reconstructs the actual paths people took from raw event logs, exposing the loops, skips, back-steps, and side-routes that a clean linear funnel silently assumes away', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Computer Science & Software Engineering
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Process-mining research established algorithms that reconstruct actual workflow paths, loops, and deviations from case-linked, timestamped event logs.
Related originating lineages:
- Data Science & Analytics — Sequence analytics supplies discovery and frequency analysis across traces.
- Organizational & Management Science — Business-process management supplies the workflows and conformance questions.
Review resolution: The Process Mining Group's event-data and process-discovery references define both the required event-log fields and the automated discovery of models, directly locating the mechanism in computer science.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
Notes¶
[n1] Conformance checking — the process-mining technique that measures how far the actually-observed process, discovered from logs, diverges from the intended model. It is what turns "here is the real spaghetti" into "here is exactly where reality departs from the assumed funnel." ↩