Managed on Azure
- Run by
- Capgemini Invent
- Hosted on
- Azure Container Apps
- Your side
- Read-only provider credentials
The platform
One platform that ties AI spend to the outcomes it produces.
{{product.name}} connects to your AI providers, builds a cost baseline you can defend and computes cost per verified outcome. Checkpoints keep a person in control of every decision.
[Replace: product film for each moment of the day]
A day with {{product.name}}
You open the cost baseline and see AI spend for the period, with the coverage score beside the figure. Any provider not yet connected appears as a named gap, so you know what the number leaves out.
How it is built
Four layers, with billing truth at the base.
Experience on top, an intelligence engine beneath it, and a token data layer on the FOCUS schema fed by read-only connectors to your providers.
- A workspace for each client, with an internal view and a client view.
- The server decides what each viewer may see. Protected methodology is never sent to client mode.
- The menu shows the modules a tenant's tier and the user's role allow, and names the tier that unlocks the rest.
- Computes Cost Per Verified Outcome and the AYCI confidence score, with provenance on every figure.
- Raises optimisation recommendations that run only after a recorded human approval.
- Authorises agent sessions and watches them in flight. The calculation engine stays with Capgemini.
- Billing and usage are normalised to the FinOps FOCUS schema.
- Authoritative billing and real-time session data are kept in separate stores, so money figures never rest on estimates.
- Every row carries a tenant id, and queries are scoped by tenant. Prompt and completion content is not stored.
- Read-only connectors for OpenAI, Anthropic, Azure OpenAI, Amazon Bedrock and Google Vertex AI.
- Azure is verified end to end on real billing data. Bedrock invocation logging and a Google Cloud billing export are client-side prerequisites.
- Credentials are encrypted at rest and connections are read-only.
Ways to deploy
Capgemini runs it. You connect read-only.
{{product.name}} is a managed service. One platform serves many clients, with no per-client forks and no calculation engine installed on your side.
A second instance on AWS
- Run by
- Capgemini Invent
- Hosted on
- AWS
- Status
- Not live yet
Self-service Control Plane
- Run by
- Your team, with Capgemini onboarding
- Covers
- SaaS-embedded AI and a prompt coach
- Status
- Not live yet
Built so a CFO can defend every number.
Honest instrumentation, protected methodology and a human decision at every gate.
Coverage is stated, not implied
Every figure sits beside the share of AI spend the platform can see. Gaps are named, and unattributed spend stays in the ranking.
No figure without a signed definition
Until an outcome definition is countersigned, the platform shows what is missing instead of an estimate.
Every number has provenance
Open any figure to see the billing rows, outcome counts, verification mix and parameter version behind it.
Methodology stays with Capgemini
Clients get full detail on their own data and outputs. Calibration, thresholds and scoring weights are not sent to client mode, and automated tests check this.
Tenant isolation
Every row carries a tenant id and queries are scoped by tenant. Staff sign in through Microsoft Entra ID.
Narrow data footprint
Version 1 stores billing and usage metadata, not prompts or completions. A privacy review is cleared before any real client tenant is onboarded.