S SARAH LOCKETT

01 / AI case study / Business operations

Designed · not yet built

The Workflow X-Ray

Feed it the event logs a business already has, and an agent team reconstructs how the work actually flows: where it loops, stalls, and leaks money.

TYPEProcess intelligence
PLANNED BUILDPlanned · ~3 weeks
STATUSDesigned · not yet built

PROVEN EXPERIENCE → NEW BUILD

Rooted in finding what the topline hides

At EdgeWater and Nimbus USA I surfaced a dealer-concentration risk, one dealer holding 75%+ of sales, by reading the data everyone had but no one had interrogated. This case study is the same instinct, aimed at process data and accelerated by agents.

THE BRIEF

A process-discovery system: agents read raw event logs from operational systems, rebuild the real process map, quantify rework and delay, and rank fixes by effort against impact.

The problem

Every business believes it knows its process. The logs say otherwise: undocumented loops, approval detours, rework cycles, and hand-offs that quietly add days. Traditional process-mining tools exist, but they are expensive, and someone still has to interpret them.

The system

An agent team ingests event logs, reconstructs the as-is process map, narrates the gap between the documented process and the observed one in plain English, and proposes fixes ranked by effort. A human validates every finding against ground truth before it becomes a recommendation.

  1. Event logs
  2. Sequence reconstruction
  3. Variant analysis
  4. Bottleneck detection
  5. Narrative + ranking
  6. Human validation
Claude agentsPower QueryPower BISynthetic event logsProcess mapping

DESIGNED OUTCOMES

What “working” will mean.

  • The observed process map, not the imagined one
  • Loops, stalls, and rework quantified in time and cost
  • A plain-English narrative of where work drops
  • Fix list ranked by effort against impact

MEASUREMENT PLAN

How it gets judged.

  • Variant coverage
  • Bottleneck detection accuracy vs. seeded truth
  • False-finding rate
  • Time from log to map
  • Human validation pass rate

THE BUILD PLAN

How this gets built
when it gets built.

Days 1–4

Synthesize

Design a synthetic order-to-cash event log with realistic pathologies: loops, skipped steps, aging queues, rework.

Days 5–10

Reconstruct

Build the agent pipeline that maps variants, timings, and hand-offs from raw events.

Days 11–15

Diagnose

Add bottleneck detection, cost-of-delay estimates, and the plain-English findings narrative.

Days 16–20

Present

Power BI companion views, validation notes, and the written case study.

Transparency note. This is a self-directed AI case study built on synthetic data. Results will be reported as evaluation findings against seeded ground truth, not client outcomes. When the build is complete, this page gains: interactive process map, findings narrative, synthetic event-log dataset, power bi bottleneck views, validation log, written case study.

NEXT CASE STUDY / 02The Reconciliation Sentinel
→