A risk team's approver authority matrix names who in the second line may approve the decisions that change how much risk agents carry: a new agent going live, a capability moving up a depth, a risk accepted, an appetite breach tolerated, or a control exception granted. Risk does not run agents; it consents to the risk they bring.
What does the risk team approve?
Not individual transactions. Risk's approvals sit above the queue, at the points where the organisation's exposure changes. The clearest is a depth increase: when an owner asks for an agent to move from depth 3 to depth 4 on a payments capability, risk reviews the certification evidence and the appetite, and either consents or declines.
Two principles shape the matrix. Depth is earned, not configured: risk's consent cannot raise an agent's depth beyond what certification evidence supports or the plan allows. Consent is a necessary condition, not a sufficient one. And appetite is written per capability and depth, so approvals can be checked against it. A risk manager approving a depth 4 capability should be able to point to the line in the appetite statement that permits it. The concept is set out in what an authority matrix is.
How do you fill it in?
- List the decisions on which risk's consent is required.
- For each, name the approver by seniority and the second approver.
- State the evidence risk must review before consenting.
- Add an expiry: risk consents should lapse and be renewed, as agent certificates do.
- Record the certificate the approver must hold, if approval rights depend on certification.
The template
| Risk decision | Approver | Second approver when | Evidence reviewed | Consent expires | Certificate |
|---|---|---|---|---|---|
| New agent go-live | |||||
| Depth increase to 3 | |||||
| Depth increase to 4 | |||||
| Accept a residual risk | |||||
| Tolerate an appetite breach | |||||
| Grant a control exception | |||||
| Rate incident severity | |||||
| Approve a new source class mapping |
A worked example
A hypothetical building society's risk function: two risk managers, a head of operational risk and a chief risk officer.
| Decision | Approver | Second approver when | Evidence reviewed | Consent expires | Certificate |
|---|---|---|---|---|---|
| Go-live | Risk manager | Customer-facing, head of operational risk | Certification result, owner named, reviewers certified | With first certificate lapse | Builder |
| Depth to 3 | Risk manager | None | Gate results on the society's own cases | 12 months | Builder |
| Depth to 4 | Head of operational risk | Always, chief risk officer | Gate results including hard cases, appetite line | 12 months | Architect |
| Residual risk | Head of operational risk | High rating, chief risk officer | Risk assessment, owner's acceptance | 12 months | Operator |
| Appetite breach | Chief risk officer | Board risk committee informed | Breach report and plan | 3 months | Architect |
| Control exception | Risk manager | Over 30 days, head of operational risk | Compensating control | 30 days | Operator |
| Incident severity | Head of operational risk | Severe, chief risk officer | Reconstruction from the record | n/a | Operator |
| Source class mapping | Risk manager | External to signed, head of operational risk | Origin of source | 12 months | Builder |
Depth 4 needs the chief risk officer as well, and the evidence must include hard cases: large values, new payees, unusual timing. A certification that cleared every gate on routine cases says little about the cases that matter.
Source class mapping appears because reclassifying a source changes what it may authorise. Moving a supplier portal from external to signed would let its data carry more weight in payment decisions, so a second approver is needed.
Consents expire. A depth increase approved for 12 months is reviewed at expiry against what the agent did meanwhile. That mirrors how agent certificates lapse after 90 days and people's certificates after 12 or 24 months.
Risk's consents interact with first-line matrices such as the approver authority matrix for the lending team and the approver authority matrix for the compliance team. Detailed thresholds are in the approver rule for the risk team.
Where the Academy fits
Pryme Intelligence Academy's Builder track teaches how agents are certified, which risk managers need to read gate results. The Architect track is for people leading regulated, multi-agent deployments and ends with a board review by two Pryme assessors; its certificate is valid for 24 months. The approver column above maps to those levels.
Questions people ask
Can risk approve a depth the evidence does not support?
No. Consent is necessary but not sufficient; depth comes from certification evidence within the role's ceiling. The skills each role needs are in AI agent skills by role.
Should risk consent to every depth 2 capability?
Usually not. Drafts that a person sends carry limited risk. Many firms require consent from depth 3 upwards, where agents act without a person on each item.
What does the board see?
Appetite breaches, depth 4 consents and severe incidents, with the record of who approved what and when.