AI governance map of a mobile carrier

Views in this map

1. 1
   AI landscapeSeven care agents, the tools they call, the data behind them and the models they run on.
2. 2
   Customer dataSix of seven agents can reach personal data, and every way in runs through a tool.
3. 3
   ApprovalsFive agents are approved, one is in review, and one went live with no review at all.
4. 4
   Billing-dispute agentReads bills and proposes credits; over $200, a supervisor decides.
5. 5
   Refund agentPays approved credits to the bill or the card; over $1,000 needs two approvals.

Landscape ‹ › Customer data ‹ › Approvals ▸ Billing · Refund

Flows to play

• ▷
  A customer disputes a billA $240 roaming charge billed twice, from the chat to the refund.
• ▷
  An agent nobody approvedHow the retention offers agent reached customer data without a review. In the Approvals view.
• ▷
  A credit over the limitWhat the guardrails do with the $240 credit the agent proposes. In the Billing-dispute agent view.
• ▷
  A refund over $1,000A returned phone: $1,240 back to the card it was bought with. In the Refund agent view.

The model behind it

56 records of 10 types. Every box on the map is one of them, so the filters and the record panel read the same data.

• Tools 10
• Systems 10
• AI agents 7
• Guardrails 7
• Teams 6
• Review decisions 6
• Models 4
• 3 more types 6

Filters

Model provider and Approval status: pick a value and everything else fades.