Fictional worked example

AI works, but the supplier reduction is not yet supported

A regulated financial-services organisation has materially reduced standard document-review work with AI. Management proposes cutting external supplier capacity by 50% at the next renewal. The Stress Test finds that the technology case is promising, but the permanent supplier decision is ahead of the evidence.

"AI is working" does not require leadership to accept an unevidenced supplier reduction.

Decision at a glance

The case supports more evidence before it supports a permanent supplier band.

Proposal Reduce supplier capacity by 50%

Make the reduction permanent at the next contract renewal, six weeks from now.

Capacity available for choice Not yet established

Normal-period supplier hours are lower, but complete workload, peak demand and specialist requirements are not yet quantified.

Supported decision now Use a three-month evidence bridge

Keep the current commitment temporarily, close the specified evidence gaps and return with a defensible permanent capacity range.

Decision status 03: Require specific evidence before structural approval

The proposed 50% band is ahead of the evidence.

Supplier capacity Document review Renewal in 6 weeks AI operating at scale for 3 months

Evidence behind the case

AI productivity is real. The permanent supplier-capacity number is not.

Observed evidence

  • Standard document-review hours are 46% lower.
  • Automated document classes represent approximately 58% of normal supplier workload, but a smaller share during remediation periods.
  • Reported supplier hours are approximately 35% lower on average since deployment.
  • Multilingual, high-risk and exception work remains largely human-reviewed.
  • No major regulatory-remediation event has occurred during the three months of AI operation at scale.
  • Complete exception, secondary-review, rework and internal-team workload has not yet been measured together.
  • Deeply reduced supplier capacity could take approximately 4-6 months to restore.

Economics and remaining assumptions

50% band annual saving ≈ £2.4m
Three-month bridge cost ≈ £0.6m foregone saving
AI operating at scale 3 months
Supplier restoration lead time ≈ 4-6 months

Assumptions still carrying the proposal

  • Current AI performance persists.
  • Multilingual automation expands successfully.
  • Average-period utilisation represents future demand.
  • Half the current commitment can absorb remediation peaks.
  • Internal teams will not absorb material additional work.
  • Lost supplier capacity can be restored quickly enough if required.

The capacity bridge

The bridge breaks before the proposed supplier band.

The AI result is credible. What has not yet been demonstrated is the permanent external capacity that can safely become unavailable.

01

AI productivity

AI has materially reduced human review of standard documents.

02

Net human work change

Not yet fully established across supplier work, internal review, exceptions and rework.

03

Released capacity

Some supplier capacity is clearly released in normal periods, but the complete amount is uncertain.

04

Structurally addressable capacity

Lower contractual bands are available, but workload has not yet been mapped to a defensible permanent band.

05

Capacity that must remain

Peak remediation, multilingual, high-risk and exception requirements are not yet sufficiently quantified.

06

Capacity available for choice

No defensible permanent amount has yet been established.

50% proposed Permanent amount not yet established

The four questions

The technology case survives. The permanent supplier decision does not yet.

How much human work is actually no longer required?

Not yet established

Standard document-review hours are 46% lower, but that is not yet equivalent to a 46% reduction in total human work. Remaining work includes multilingual and high-risk cases, exceptions, secondary review, rework, controls and work potentially transferred to the internal team.

AI has clearly removed meaningful human effort. The complete net human work change across supplier and internal capacity is not yet known.

Is released capacity structurally addressable?

Partly demonstrated

Supplier capacity is structurally addressable through contractual bands, and normal-period supplier hours are approximately 35% lower. But the organisation has not shown how that workload changes under remediation demand, different document mixes or additional internal review.

A supplier-capacity reduction is worth examining. The current evidence does not establish a defensible 50% permanent reduction.

What capacity must remain to protect performance, coverage, skills and resilience?

Not established

Remaining supplier capacity protects multilingual capability, high-risk review, exception handling, specialist review and remediation peaks. No material remediation event has occurred since the AI workflow reached scale, and restoring deeply reduced capacity could take four to six months.

The minimum external capacity that must remain is not yet known. Therefore capacity available for choice cannot yet be quantified.

Is the proposed reduction, at this amount and timing, economically better than the credible alternatives?

Bridge is stronger

The proposed 50% band has clear savings, but the available-capacity number is not yet evidenced. The alternatives therefore have to be compared with the cost and value of closing that gap.

Option A

Reduce supplier capacity by 50% now

≈ £2.4m annual saving

The immediate saving is clear, but the permanent reduction relies on untested assumptions about peaks, document mix, internal workload, specialist capacity and restoration friction.

Saving evidenced. Capacity number not evidenced.
Option B

Make a smaller permanent reduction now

No defensible permanent percentage yet

A smaller band may ultimately prove preferable. Choosing 20%, 25% or 30% simply because 50% looks too aggressive would create another unsupported number.

Do not replace one unsupported number with another.
Option C

Use a three-month evidence bridge

≈ £0.6m of potential savings foregone

Extend the current commitment for three months at the existing rate. Use that bounded period to establish complete workload, peak demand, specialist capacity, internal transfer and restoration economics.

Strongest current choice.
Option D

Redeploy released spend elsewhere

Released budget not yet established

Redeployment may become relevant after permanent supplier capacity is quantified. There is no defensible pool of released spend to allocate before that number exists.

Premature until capacity available for choice is established.
Do not approve the proposed 50% permanent reduction or invent a substitute percentage. Use the three-month bridge to close the specific evidence gaps, then return with a defensible capacity range before selecting the next supplier band.

Stress Test conclusion

Require specific evidence before structural approval.

The Stress Test does not reject the AI workflow or conclude that supplier capacity cannot be reduced. It finds that the current evidence does not establish how much supplier capacity is genuinely available for permanent reduction.

Evidence bridge ≠ indefinite delay. Use the three-month extension only for the specified evidence work, assign clear owners and set the return decision before the bridge expires.

What could change the answer

Close five evidence gaps, then return to the structural decision.

Evidence required

  • Net human workload across the complete AI-enabled process, including internal review, exceptions, controls and rework.
  • Supplier utilisation under normal and peak/remediation conditions.
  • Minimum external capacity required for multilingual, high-risk and exception work.
  • Contractual and economic consequences of realistic reduction bands.
  • Restoration lead time, cost and procurement requirements.

Bridge controls and return point

  • Assign an owner to each evidence item.
  • Agree a peak-capacity simulation if a real remediation event does not occur during the bridge.
  • Obtain supplier pricing and restoration terms for realistic bands.
  • Do not turn the three-month bridge into an open-ended extension.
  • Return before the bridge expires with a defensible amount of supplier capacity available for choice.

What this example shows

A real productivity result can still be insufficient for a permanent supplier decision.

AI productivity is real supplier workload appears lower complete net-work change is not known peak and specialist requirements are unresolved capacity available for choice cannot yet be quantified 50% is unsupported do not invent a compromise percentage name the evidence required and return to the decision

The failure is not that AI has not worked. The missing link is the evidence from AI performance to permanent supplier capacity.

AI Restructuring Business Case Stress Test

Apply the same test to your own case.

Use the parent resource to test the bridge from AI productivity to a proposed permanent capacity decision.

Open the Stress Test

More worked examples

Browse all worked examples