Approval one
Change the work
Is the evidence strong enough to change how the work gets done?
Test capability, adoption, quality, human review and whether the redesigned process performs well enough to use.
Executive decision resource
Test whether AI productivity evidence supports a proposed permanent reduction in roles, contractor or supplier capacity, or related operating cost.
AI can change how much work people need to do. That does not automatically establish how much paid capacity can safely and economically leave.
Why this matters
Organisations are moving from AI pilots and productivity claims to permanent choices about roles, supplier capacity and operating cost.
Those choices can create real savings, but they can also weaken service, controls, skills or resilience, and lost capability may be costly to rebuild. Waiting has a cost too. The point is to test whether the proposed amount, timing and mechanism are supported, and what would make a different answer preferable.
Separate the decisions
One business case can contain both. Evidence for the first does not automatically settle the second.
Approval one
Is the evidence strong enough to change how the work gets done?
Test capability, adoption, quality, human review and whether the redesigned process performs well enough to use.
Approval two
Is the evidence strong enough to remove this amount of paid capacity, and to remove it now?
Test workload, concentration, what must remain, economics, alternatives and uncertainty.
This Stress Test focuses here.
The Stress Test
Questions 1–3 establish the capacity bridge. Q4 then tests the credible structural choices, leading to the supported structural outcome, formal status and recommendation.
AI productivity → Net human work change → Released capacity → Structurally addressable capacity → Capacity that must remain → Capacity available for choice
No scoring. No automatic headcount answer.
Time or operating capacity freed by the changed workflow. It can still be dispersed across roles, needed elsewhere or already being used for other work.
The part no longer required by the current operating model once its location, concentration, remaining operating requirements and risks have been tested.
Available does not mean it should be removed. It may be removed, redeployed, used to absorb growth, strengthen service or resilience, or retained while a material uncertainty resolves.
The net change in human workload across the complete AI-enabled workflow.
Baseline and current workload, volumes, human review, exceptions, escalations, controls, work newly created by the AI-enabled process and material work transferred to another team or supplier.
After everything still required or created elsewhere is included, how much human work is genuinely no longer required?
Whether the released time forms an identifiable structural unit that can be tested against the capacity the operating model still requires. For suppliers or contractors, the unit may be contractual rather than role-based.
Where the gain sits by role, team, shift, location or supplier. Test whether small fragments of released time can genuinely be consolidated into whole roles, roster positions, supplier units or another addressable block. Where the structure is discrete, identify the feasible packages or tranches rather than assuming the capacity can be adjusted continuously.
Can the released capacity be mapped to a specific structural unit that can actually be changed, or is it still scattered through existing work?
How much of the structurally addressable capacity is still required by the operating model.
Peak demand, shift coverage, service and quality requirements, exception loads, controls and compliance, scarce skills, resilience, AI or vendor failure, fallback capability, expected demand growth, and any binding legal, regulatory, licence or contractual minimum that determines how much capacity must remain.
What capacity still has to remain even after average workload has fallen?
Whether permanent removal creates more value than the credible alternatives, not merely whether the capacity can be removed.
Compare credible options and combinations: remove, stage, use attrition, redeploy, absorb growth, strengthen resilience or retain capacity. Test them on comparable evidence about value, feasibility, timing and reversibility.
If this capacity is genuinely available, what else could the organisation do with it, and why is permanent removal of this amount now preferable?
If the structural option changes the evidence: if a proposed structural option could materially change workload, AI performance, service demand, downstream work or another assumption used earlier in the Stress Test, rerun the affected earlier questions for that option before treating the capacity result as decision-ready.
If several credible uses remain after comparison: close material evidence gaps that could determine the ranking. If adequately evidenced options still depend on broader priorities, constraints or risk appetite, move to a broader resource-allocation decision. Different parts of the capacity may resolve differently.
Then try to change the answer
What evidence would change the amount, timing or structural choice?
Finish the test
The Stress Test does not prescribe the outcome. It makes clear what the evidence currently supports and what remains unresolved.
The amount and timing remain supported after workload, structural addressability, protected requirements, economics and sensitivity have been tested.
The proposed reduction is not the strongest available choice. A different amount, timing, mechanism or use of the available capacity may create more value.
Material evidence gaps still prevent a defensible structural decision, whether about available capacity, timing, economics, reversibility or a credible alternative.
Adequately evidenced uses remain competitive and the answer depends on broader priorities, constraints or risk appetite. Different tranches may have different statuses.
Worked examples
The framework is outcome-neutral. These fictional cases show how different evidence changes what a defensible structural decision looks like.
The evidence supports a 32-position customer-service reduction after workload, capacity, service, resilience and the credible alternatives are tested.
View worked exampleHuman effort per transaction falls materially, but higher demand and new AI-related work absorb the gain. The proposed structural reduction is not supported.
View worked exampleSupplier work has fallen, but the evidence does not yet support the proposed 50% capacity reduction. A bounded evidence bridge is required before structural approval.
View worked exampleTwenty FTE are available for choice. Eight are clear to remove, while the remaining 12 have credible competing uses and require a broader resource-allocation decision.
View worked exampleThe output
Another executive should be able to see what is proposed, which evidence supports it, which assumptions remain and what would cause leadership to choose differently.
If the decision is live
In a 30-minute Decision Review, work through the proposed change, the evidence, the constraints and what would have to change for a different answer.