Pillar 04 · Cockpit
Track, compare and steer AI usage across the organisation — per process, per team, per model, per provider. Including the algorithms that aren’t ours.
Four steps from an AI estate you endure to an organisation that steers its algorithms.
Step 1
Instrumentation SDK, MCP connectors or natively NEXA processes. Everything that executes AI in the organisation reports to the same place, whatever its origin.
Step 2
Each project is attached to its department, criticality, owner and value-measurement strategy. Configuration happens in the platform, AI-assisted, or via Excel import for estates already inventoried.
Step 3
Cost in euros, tokens consumed, CO₂ footprint, number of uses and error rate — measured at every execution, consolidated per process, team, model and provider.
Step 4
Arbitrate an engine, restart a stalled process, cut a usage without value, multiply the one that has it. That is the shift from an organisation that endures its AIs to one that steers them.
How many AIs run in your organisation today — and which ones?
Per process
“Does this process cost me more than it brings back?”
What the Cockpit answers
Cost, usage, quality and value of a given process, from framing to production. ROI is configured by its owner, with their unit and valuation rate.
Per team
“Who actually uses AI here, and for what?”
What the Cockpit answers
Which department consumes, which produces, which validates. Active users and execution volume per entity, over rolling twelve months.
Per model and provider
“Am I building a dependency on one provider?”
What the Cockpit answers
Which share of the estate depends on which engine, at what comparative cost. Enough to arbitrate a provider change without rebuilding the processes.
Per criticality
“Which of my usages require mandatory human validation?”
What the Cockpit answers
The level of control applied to each process: active checks, substantiation thresholds, mandatory validation roles. And the ones that don’t have them yet.
Per family
“Where should I focus my next industrialisation effort?”
What the Cockpit answers
Generation, document knowledge, qualification, steering, acceleration. Families with an already-high acceptance rate are the cheapest to multiply.
Per maturity
“How many projects are stuck between POC and production?”
What the Cockpit answers
Framing, development, POC, industrialisation, run. The estate reads like a pipeline: you see where usages stall, and for how long.
Three mechanisms without which a cockpit is just another dashboard.
Configurable ROI
Saved analyst time, an avoided risk, an error not made and additional revenue don’t count in the same unit. The process owner picks the measurement strategy, unit and valuation rate: seven measurable value types. ROI becomes comparable across processes without being uniformised.
Extended scope
A cockpit that only sees its own executions governs nothing. The SDK instruments in-house algorithms; MCP connectors plug in third-party platforms and IT tools.
Roles and identities
Governance is not a dashboard, it is a distribution of responsibilities. Each role has its view and its power of arbitration.
A governance framing starts from what exists: what runs today, for whom, at what cost, with what level of control.
