AI curriculum

Course outline for Workflow Performance Engineering

Measure useful completed outcomes across the whole work system before adding AI. Trace waiting, handoffs, quality, review, rework, cost, and human capacity. Test whether a local speedup improves the total result or merely moves the bottleneck. Compare constraint-focused redesign and simpler non-AI options. Design a bounded improvement test with feedback and stop criteria.

Workflow Performance Engineering Course Objectives

  • Define system boundaries, useful outcomes, measurement intervals, and usable case records.
  • Distinguish local task efficiency from whole-system throughput, cost, quality, and completion time.
  • Calculate cycle time, throughput, effort, cost, quality, and rework separately.
  • Test bottleneck shifts, feedback, and capacity changes under a whole-system objective.
  • Select appropriate delivery measures without turning them into universal targets.
  • Compare local AI speedups, constraint-focused redesign, and simpler non-AI alternatives.
  • Design an improvement test and explain its causal and statistical limits.

Pre-requisites

  • Describe a familiar workflow and its completion condition.
  • Use spreadsheet formulas, timestamps, counts, and simple charts.
  • Workflow Foundations for AI Automation or equivalent process, case, handoff, and queue knowledge.
  • No programming, model account, or observability-platform experience is required.

Before this course

Complete the required courses or bring equivalent experience. Recommended courses add useful depth; they are not enrolment gates.

Required preparation

Recommended preparation

Readiness checks

  • Describe a familiar workflow and its completion condition.
  • Use spreadsheet formulas, timestamps, counts, and simple charts.
  • Workflow Foundations for AI Automation or equivalent process, case, handoff, and queue knowledge.
  • No programming, model account, or observability-platform experience is required.

Describe a familiar workflow and its completion condition.

Find your place in the learning map

Lab Setup

Delivery requirements: the environment, access, and exercise materials must be prepared and verified before delivery. This outline does not certify that a lab package is ready.

Spreadsheet, synthetic case events, resource calendars, cost inputs, software deployment records, and separately labeled hypothesis values. Materials include formula scaffolds, one worked case, and independent cases. OpenTelemetry is a conceptual comparison, not an SDK/server implementation. Delivery requires checked formulas, time zones, reference answers, and current DORA definitions. No model account is required.

Detailed Course Outline

Module 1: Measure complete work rather than isolated activity (1.5 hours)

  • System boundary: System boundary, useful accepted outcomes, case identity, and measurement windows.
  • Event quality: Event records, missing timestamps, inconsistent clocks, and invalid durations.
  • Flow measures: Cycle time, handling time, waiting, throughput, and work in progress.
  • Outcome denominators: Accepted, failed, withdrawn, and unfinished cases with separate denominators.
  • Quality and rework: Quality criteria, first-pass acceptance, correction effort, and repeated work.
  • System throughput: Local activity/utilization versus accepted system throughput and misleading averages.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.5 hours for practice, 0.25 hours for feedback, and 0.25 hours for questions and contingency.

Practical outcome: Build a reconciled measurement sheet without hiding incomplete or unsuccessful work.

Module 2: Find the constraint and test its movement (1.5 hours)

  • Demand and capacity: Arrival rates, service rates, resource calendars, and available capacity.
  • Bottlenecks and queues: Bottlenecks, utilization, queue growth, and uneven case requirements.
  • Shared resources: Batches, handoffs, parallel steps, and shared-review resources.
  • Local optimization and feedback: Local optimization, shifting bottlenecks, and feedback from errors to review load.
  • Rate-model limits: Bounded rate arithmetic versus variable real waiting times.
  • Stable-interval assumptions: Stable-interval assumptions and limits on work-in-progress relationships.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.5 hours for practice, 0.25 hours for feedback, and 0.25 hours for questions and contingency.

Practical outcome: Predict the next constraint and show a faster AI stage that does not improve accepted system throughput.

Module 3: Connect observations to cost and quality (1.5 hours)

  • Case-level observations: Case-level event views and links between logs, traces, and metrics.
  • Instrumentation gaps: Instrumentation gaps, sampling, and missing or duplicated observations.
  • Full lifecycle cost: Total setup, operation, review, rework, integration, practice, and recovery costs.
  • Break-even sensitivity: Break-even volume and sensitivity to price, quality, and demand.
  • DORA event definitions: DORA measures for software changes and their explicit event definitions.
  • Software delivery measures: Change lead time, deployment frequency, recovery, failure, and deployment rework.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.5 hours for practice, 0.25 hours for feedback, and 0.25 hours for questions and contingency.

Practical outcome: Explain a cost comparison and select measures that fit one software and one nonsoftware process.

Module 4: Choose and evaluate one improvement (1.5 hours)

  • Process simplification: Simplification, capacity changes, smaller batches, and better handoff information.
  • Four improvement options: Unchanged workflow, local AI speedup, constraint redesign, and non-AI alternatives.
  • Comparable baselines: Baselines, matched cases, observation windows, and workload changes.
  • Delayed effects and guardrails: Delayed quality/capability effects, review capacity, and scale, revise, or stop thresholds.
  • Confounding and small samples: Confounding, small samples, and limits of before-and-after evidence.
  • Improvement ownership: Improvement ownership, follow-up measurements, and the next experiment.
  • Guided-time allocation: 0.5 hours for explanation and a worked example, 0.5 hours for practice, 0.25 hours for feedback, and 0.25 hours for questions and contingency.

Practical outcome: Defend a bounded improvement proposal and reject a locally faster but globally worse option.

Practical exercise

  • Analyze a repair desk using synthetic case events and resource schedules. Reconcile received cases with accepted completions, failures, withdrawals, and unfinished work. Separate cases entering the window from cases completing in it. Calculate elapsed time, waiting, handling effort, quality, correction effort, throughput, and cost. Flag missing or contradictory records instead of silently turning them into zero. Explain which summaries exclude unfinished durations.
  • Build a spreadsheet capacity comparison with inspection and review as competing constraints. Vary each capacity and arrivals independently. Compare faster intake, extra review capacity, and reduced rework. Include continuing arrivals and shared resources. Scenario calculations are not observed waiting-time predictions.
  • Compare manual processing with a supplied AI-assisted hypothesis, including setup, inference, review, and error correction. Keep hypothesis values visibly separate from recorded case data. Compute a break-even range and show a case where quality or review capacity invalidates the saving. Map current DORA definitions to a small software-change dataset, then explain why they do not directly measure repairs.
  • Create a system-level comparison of four options: the unchanged workflow, a local AI speedup, a constraint-focused redesign, and a simpler non-AI improvement. Use the same demand, outcome boundary, acceptance policy, observation window, and resource assumptions. Report useful completions, lead time, queue/WIP location, quality, total human effort, and full cost. Reject an apparent gain that merely moves waiting, increases rework, excludes failed cases, or assumes free review. Distinguish temporary backlog reduction from sustainable completion capacity. Account for integration, maintenance, protected skill practice, and recovery. Describe feedback and delayed effects that the bounded worksheet cannot predict. Recommend no AI when another option meets the need better.
  • Create an improvement plan with an observation window, independent quality check, owner, workload comparison, and stop threshold. Save a prediction for a new capacity case before revealing the worked feedback. Do not claim the proposal proves a production benefit, model accuracy, causation, or statistical significance.

Delivery and timing

  • Guided hours include teaching, demonstrations, learner practice, feedback, questions, and contingency as allocated above. Breaks, bulk downloads, and prerequisite setup are additional. Allocations need a timed learner pilot before delivery; they are not measured completion times.

What you will produce

Measure whole-workflow performance and defend an improvement using capacity, quality, effort, and full cost.

How the work is checked

  • Counts reconcile across all case outcomes and unfinished work.
  • Calculations use explicit windows, units, denominators, and exclusions.
  • Capacity tests expose a local speedup with no system gain and locate the shifted queue.
  • Four-option comparisons use consistent boundaries and include full human/technical cost.
  • The recommendation can reject AI and explains feedback, delays, and quality/capacity limits.
  • Delivery measures fit their workflow and do not become individual quotas.
  • The improvement plan includes quality, ownership, uncertainty, and stop criteria.

6 guided hours. Whole-system performance comes before AI transfer. Compare unchanged work, local AI speedup, constraint redesign, and simpler non-AI options. The current DORA guide includes five measures, including deployment rework rate. Rate arithmetic and break-even estimates depend on explicit assumptions; they are not measured waiting-time predictions, causal certification, or statistical significance. No telemetry deployment or live model is required.

When another approach is better

  • Do not use software-delivery metrics as universal targets for unrelated business workflows.
  • Do not scale a faster stage when the next stage lacks capacity or quality deteriorates.

How we train

Contact us for full course details, including duration, delivery options and lab requirements.

Ask about exercises, instructor feedback and the prior knowledge you need. Tool-specific courses marked provisional may become modules in a broader course.

Discuss your learning goals