Privacy-Preserving Conformance Checking: Auditing Business Processes Without Exposing Sensitive Event Logs

Forum topic · 小凯 · 2026-05-04

Summary

A Chinese forum post introduces an arXiv paper (2605.00283, 2026-04-29) by Luis Rodríguez-Flores, Luciano García-Bañuelos, Abel Armas-Cervantes, and Astrid Rivera-Partida titled "A Privacy-Preserving Approach to Conformance Checking." Conformance checking compares a reference process model against real event logs to detect deviations, but logs often contain sensitive customer and transaction data, blocking outsourced audits, cross-organization collaboration, and compliance work. The proposed approach uses secure multi-party computation so the model owner and log owner can jointly compute conformance results without revealing their inputs, combining cryptographic techniques such as homomorphic encryption and secure data sharing to output aggregated deviation statistics rather than raw events. The post explains why this matters: raw log exposure risks data leaks, GDPR-style compliance barriers, and lack of trust between organizations, while privacy-preserving analysis keeps data in place and lets only insights flow. It closes with the analogy that analysts can learn "average height" without knowing any individual's height, arguing that cryptography makes privacy and analytics complementary rather than a zero-sum trade-off.

Paper Overview

Paper: A Privacy-Preserving Approach to Conformance Checking
Authors: Luis Rodríguez-Flores, Luciano García-Bañuelos, Abel Armas-Cervantes, Astrid Rivera-Partida
arXiv: 2605.00283 | 2026-04-29

The Compliance Dilemma: You Need the Process, But Can't See the Data

Conformance checking compares how a process *should* run (the process model) against how it *actually* ran (the event log) to find deviations and improve operations. The problem: event logs contain sensitive information — customer data, transaction details — that cannot be shared externally.

This blocks common scenarios such as:

Outsourced process audits
Cross-organization collaboration
Regulatory compliance checks

The goal is to perform conformance checking while keeping the underlying data private, using secure computation.

The Proposed Approach

The core idea: complete conformance checking via secure computation when either the process model or the event log must remain confidential.

Key components:

1. Secure multi-party computation (MPC) — the model holder and log holder jointly run cryptographic protocols that compute results without revealing their inputs. 2. Privacy-preserving techniques — combining approaches such as homomorphic encryption, secure data sharing, and (where relevant) differential privacy. 3. Privacy-aware conformance metrics — deviations are computed and reported in aggregated/statistical form, without exposing individual events. 4. Application scenarios — outsourced auditing, supply-chain collaboration, healthcare processes, financial compliance, and any domain with confidential data.

An intuitive analogy:

*Traditional audit* = the auditor reads all raw records and learns every detail (privacy risk).
*New method* = the auditor sees only aggregate results (e.g., "3% deviation") without access to specific transactions — the audit still gets done, privacy is preserved.

Why Privacy Matters in Process Mining

Without privacy protection:

Data leakage risk — raw logs expose customer information and trade secrets, creating legal exposure.
Trust deficit — organizations refuse to share data, blocking collaboration and process improvement.
Compliance barriers — regulations such as GDPR restrict data movement, making audits hard.

With privacy protection:

Security — raw data never leaves its domain; guarantees are cryptographic.
Collaboration — parties can cooperate without mutual trust.
Compliance — analysis satisfies regulatory requirements.

Key Takeaway

Conformance checking needs not the *raw data* but the *patterns within it*. Just as you can know an average height without knowing anyone's individual height, cryptography lets us extract the pattern without exposing the data — "data stays put, knowledge flows."

Questions worth asking if you work with process mining or sensitive data:

1. Does my process audit truly require exposing raw data? 2. Can privacy-preserving techniques help with compliance? 3. Is secure multi-party computation applicable to my scenario? 4. Can analysis be completed while data never leaves its domain?

Privacy and analytics are not a zero-sum game — cryptography can deliver both. When process mining learns to "check without seeing," it transforms from a data snoop into a privacy guardian.

*Note: The original post discusses the paper at a conceptual level; consult the arXiv paper (2605.00283) for protocol details and evaluations.*