Agents your risk team can sign off
We build for organizations where every automated decision has to be explained, owned and audited: financial services, government and healthcare.
Financial services
Applies to federally regulated financial institutions from May 1, 2027, and covers AI and machine learning models alongside traditional ones.
Read the source ↗What's expected of you
- Every model, including AI agents, identified, inventoried and risk-rated
- A named owner and clear accountability for each model
- Validation before use and ongoing monitoring after
- Evidence you can show internal audit and the regulator
How we build for it
- Each agent is registered with an owner, a Microsoft Entra Agent ID and a risk rating from day one
- Evaluation test sets act as repeatable validation, rerun on every change
- Monitoring, transcripts and change history kept for audit
- An Agent Fact Sheet per agent covering purpose, data, model, limits and owner
Agents we build
- Client onboarding: documents checked, record created, file routed to the right advisor
- Exception handling: items triaged and queued, with a person approving every decision
- Internal knowledge: policy answers grounded on approved SharePoint content only
Government and public sector
Applies when federal institutions use automated systems, including generative AI, to make or support administrative decisions. An Algorithmic Impact Assessment must be completed and published before the system goes into production.
Read the source ↗What's expected of you
- An impact assessment before launch, scaled to the decision's impact
- Transparency, so people know when a system is involved in a decision about them
- A person in the loop where decisions affect people
- Bilingual, accessible services
How we build for it
- Design documentation shaped to feed the Algorithmic Impact Assessment and peer review
- Agents that assist and recommend; people make the decision
- Clear notices and audit trails built into the process
- Our bilingual, accessible Power Pages accelerator for public-facing intake
Agents we build
- Citizen intake: requests classified, routed and acknowledged in both official languages
- Case preparation: files summarised for an officer, who makes the decision
- Staff self-service: HR and policy answers from approved sources
Healthcare
The Information and Privacy Commissioner of Ontario advises health information custodians to complete a privacy impact assessment before introducing an AI system that involves personal health information.
Read the source ↗What's expected of you
- A privacy impact assessment before an AI system touches personal health information
- Only approved tools, with no "shadow AI"
- The minimum personal health information needed for the purpose
- Clear accountability for anyone acting on the custodian's behalf
How we build for it
- Agents run only in your approved Microsoft tenant, with the data flows mapped for your PIA
- Least-privilege access and sensitivity labels on what agents can reach
- Human review for anything clinical or urgent
- Usage monitoring so unapproved AI use is visible
Agents we build
- Referral intake: forms read, urgent cases flagged for a person, appointments proposed
- Scheduling and reminders: patient communications handled within approved channels
- Staff knowledge: policies and procedures answered from approved content
The evidence pack
Your compliance, privacy and audit teams make the call. We hand them what they need to make it.
The Agent Governance Checklist
30 checks for Copilot Studio agents, updated for Ignite 2026: identity, data access, ALM, models, quality and cost. Score yourself in 15 minutes.
- Microsoft Entra Agent ID and agent accounts
- Model choice, harness and knowledge sources
- Agent 365 cost and value tracking
Working under a regulator?
Tell us which framework you answer to. In 30 minutes we'll show you how an agent can meet it.