Demonstrate a broken workflow before proposing AI

If the problem is duplicate events or missing ownership, an LLM may add cost without fixing it. Show the failure and baseline first.

Quantify a fictional workflow

The delivery lab uses twenty authored events.

from delivery_cases import broken_workflow_case

result = broken_workflow_case()
assert result["completion_rate"] == 0.8
assert result["failure_count"] == 4
assert result["baseline"]["median_minutes"] == 12
assert result["ai_required_to_show_failure"] is False

Map the current trigger, handoffs, decisions, systems, waits, exceptions and rework. Use sample evidence to measure volume, completion, error and cycle time. Reproduce one failure safely.

Then isolate cause. Deterministic validation, idempotency or ownership may solve it. Use AI where unstructured interpretation is genuinely required, and evaluate it against a simpler baseline. State what would make the proposal unnecessary.

The FDE for Freshers course ties solution choices to measured client workflows.

Exercise

Model a fictional manual process with twenty events and inject four failures. Recommend the smallest control first, then justify any AI step separately.

Continue learning

This article is part of the FDE career entry and client-delivery practice sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Explain a technical risk to a nontechnical client.
• Continue with Prepare a live demo with a recovery plan.

Reference: GOV.UK QA guidance.