AI does the analysis. People approve at the gates.
The platform produces the finding and the recommendation. A consultant chooses between bounded options at each checkpoint, and an expert steps in only on exception.
AI spend is growing faster than anyone can explain it.
The FinOps Foundation reports that 98% of practitioners now manage AI spend directly, up from 31% in 2024. Few organisations have the specialists to attribute it by hand.
From reporting spend to proving what it bought.
A consultant can run an engagement.
A consultant with no FinOps-for-AI background can run a checkpoint, because the platform does the analysis and shows its confidence and the evidence behind it.
- Each checkpoint is designed to complete in 20 minutes or less, with a visible time budget.
- Two to four bounded options, each with its consequence spelled out. No free-text decisions.
- Escalation to an expert is always available, with the reason and full context attached.
- Every decision is recorded: who decided, which option, when, and with what confidence.
Six principles shape how the work is organised.
AI-primary analysis
The platform produces the analysis and the recommendation. No step defaults to expert-led without a recorded reason.
Bounded decisions
People choose between two and four defined options, never open-ended judgement.
Every action has a gate
Optimisation changes run only after a recorded human approval.
One-to-many
One platform, many clients. No per-client forks and no calculation engine installed on the client side.
Honest instrumentation
If the platform cannot see something, it says so. Silent gaps would undo the CFO conversation the product exists to win.
Escalate by exception
Escalations carry full context and start an SLA clock. A checkpoint that escalates often is treated as a design defect.
Four groups, one platform.
The Capgemini engagement lead agrees scope and tier with the client and countersigns each outcome definition with the client owner.
FinOps-for-AI experts (L2) resolve escalations against an SLA, approve threshold and policy changes, and own scope and assumption reviews.
Generalist consultants (L1) run the checkpoints: G1 to G3 for Govern, then G4 to G7 as an engagement moves through Attribute and Optimise. They escalate on exception.
Capability owners (L3) calibrate detection and thresholds from observed outcomes, publish parameter versions and redesign checkpoints that escalate too often.
The roles behind the platform.
Agrees scope and tier with the client and countersigns outcome definitions with the client owner.
Sees tier, scope and engagement economics. No access to client data detail beyond what scoping needs.
Provisions tenants, manages connections and publishes parameter versions.
Runs checkpoints, picks from bounded options, confirms platform findings and escalates on exception.
Resolves escalations against an SLA, approves threshold and policy changes, and decides paused agent sessions that a consultant declines.
Calibrates detection rates and thresholds from observed outcomes and analyses escalation patterns.
Sets up connections, manages users, signs outcome definitions and proposes agent policy.
Read-only and operational access to the client workspace.
From a Diagnostic to continuous control.
-
Diagnostic and Tier 1
Govern
Connect providers, accept a cost baseline and agree a governance framework through gates G1 to G3.
-
Tier 2
Attribute
Sign outcome definitions and accept Cost Per Verified Outcome through G4 and G5. Verified Yield follows once a baseline is captured.
-
Tier 3
Optimise
Authorise recommendations at G6 and review sustained control at G7. The agentic add-on sits on top for clients running agents.
One platform, many engagements.
- One platform serves many concurrent engagements, with no per-client forks.
- Findings are consistent because they come from one platform, not from whichever team was staffed.
- Checkpoint durations, escalation rates and detection performance are measured, so the delivery model is tested rather than assumed.
Where does your organisation stand?
A short self-assessment of how you track, attribute and govern AI spend today.