Course outline for Workflow Performance Engineering

Module 1: Establish an end-to-end baseline

• Cycle time: Measure elapsed time across the whole case, separating generation, waiting, review, and rework rather than timing one task.
• Throughput: Count accepted completed outputs over a defined interval and reconcile case records before comparing workflow variants.
• Cost: Combine setup, operation, review, and rework costs to estimate break-even volume under explicit workload and pricing assumptions.
• Quality: Define acceptance guardrails for matched outputs so faster completion cannot conceal unsupported recommendations or increased correction effort.

Practical outcome: Calculate baseline measures including setup, review, operating cost, and accepted outputs for matched synthetic cases.

Module 2: Find the actual capacity constraint

• Rework: Measure correction effort and repeated processing separately from initial generation to expose their contribution to total workflow cost.
• Bottlenecks: Locate stages whose limited service capacity creates waiting, including review queues that grow after generation becomes faster.
• Constraint analysis: Compare arrival and service rates under stated capacity assumptions, explaining when queue growth undermines apparent local speed improvements.

Practical outcome: Compare arrival and service rates, locate review-queue growth, and state assumptions behind a capacity estimate.

Module 3: Decide what to improve and retest

• DORA-style metrics: Select delivery measures appropriate to the workflow rather than transferring software delivery metrics without checking their meaning.
• Workflow observability: Build a case-level measurement view that flags missing timestamps and invalid records instead of silently counting them as zero.
• Continuous improvement: Define a quality-guarded experiment and scale, simplify, or stop threshold while acknowledging uncertainty in before-and-after comparisons.

Practical outcome: Choose relevant delivery measures and a quality-guarded experiment, then estimate break-even volume and a scale-or-stop threshold.

Practical exercise

• Build a comparison dashboard from matched manual, model-driven, and hybrid case records.
• Reconcile case counts and elapsed times; flag missing timestamps and exclude invalid records with a reason rather than silently treating them as zero.
• Diagnose the fixture where generation accelerates but reviewer queues grow, and calculate total effort including rework.
• Estimate break-even volume under supplied cost assumptions and recommend scale, simplify, or stop with uncertainty and quality guardrails.