Layer 3

L3Scenario Workbench

Compare candidate ways to deliver an Objective. Pick a focus, see the cost and timeline implications, decide.

Objective

Scenario

Fast: triage agent only

Bolt an autonomous triage agent on top of ClaimDesk; leave adjudication unchanged.

Scenario

Chosen

Balanced: triage + L1 copilot

Autonomous triage + supervised-autonomous L1 adjudication. L2 stays human.

Scenario

Deep: terraform claims end-to-end

Reshape the function so 'adjudicate claim' becomes 'review-and-sign' for most cases.

Fast: triage agent only£207k · 1.9 mo

Cost

Calendar

Balanced: triage + L1 copilot£400k · 3.7 mo

Cost

Calendar

Deep: terraform claims end-to-end£738k · 7.5 mo

Cost

Calendar

Side-by-side comparison

The decision moment lives here. Numbers are committed estimates — not contracts.

Dimension	Fast: triage agent only	★ Balanced: triage + L1 copilot	Deep: terraform claims end-to-end
Focus	Bolt an autonomous triage agent on top of ClaimDesk; leave adjudication unchanged.	Autonomous triage + supervised-autonomous L1 adjudication. L2 stays human.	Reshape the function so 'adjudicate claim' becomes 'review-and-sign' for most cases.
Total cost	£207khigh	£400kmed	£738klow
Platform	£38k	£64k	£110k
Dev sessions	£92k	£178k	£340k
Training	£18k	£56k	£92k
Calendar	1.9 mohigh	3.7 momed	7.5 molow
Critical path	1.4 mo	2.6 mo	5.2 mo
Effort	2.6 mo	5.3 mo	10.7 mo
Sensitivity	Stable — depends on ClaimDesk API access.	Hinges on rate-card data freshness and policy-drift signal quality.	Legal sign-off gate is the dominant unknown.
Risk	Cycle-time impact is meaningfully smaller than target. Adjudication queue still pages L2 weekly.	Workforce impact: 4 L1 adjusters need pathways into Exception Triager + Policy-Drift Analyst.	Two new emergent roles required immediately: Agent Steward and Policy-Drift Analyst.

See the full plan and decision record: Cut claims cycle time by 30% →