An approver authority matrix for support leadership sets out which leaders (team leads, the support manager and the head of support) may approve each action an AI agent holds, at which depth and up to which limit. It also records the certificate each approver must hold, so authority lapses when competence does.
What is the matrix for, and when do you need it?
Frontline staff approve routine held items within small limits, and that layer has its own approver authority matrix for the support team. The leadership matrix covers everything above it: goodwill beyond the frontline limit, refunds that move money, final responses to formal complaints, account closures, exceptions to policy, and changes to the agent's own bounds.
That last group matters most. When the support manager raises a customer limit or agrees that an agent's refund capability should move from depth 2 to depth 3 after Attest evidence, they are widening what the agent may do without a person in the loop. That decision belongs in the matrix with a named owner, not in an email thread.
Build the matrix before any support agent works above depth 2. Review it whenever leadership changes, whenever an agent gains a capability, and at least once per certificate cycle. If you are new to the idea, start with what an authority matrix is.
How do you fill it in?
Work row by row from the agent's capability list, not from your organisation chart. Each row is one action the agent can hold.
- Action held by the agent: use the capability name exactly as the workspace shows it, so routing and the record match.
- Agent depth: the depth the agent runs at for that capability. Depth 2 means the agent drafts and a person sends; depth 4 means money moves or the action is hard to undo.
- Approver role and named approver: a person, never a shared inbox.
- Limit: the highest value or risk level that approver may release.
- Second approver above: the threshold at which a depth 4 action needs two people.
- Certificate required: the current certificate the approver must hold. A business can adopt the practice of making approval rights depend on that certificate, depth by depth, so an expired certificate removes the right automatically.
- Conflicts: who may not approve, such as the person who wrote the agent's brief or set its limits.
- Withheld actions get a row too, marked "no approver". They are refused at the platform floor for every agent at every depth, and no leader can release them.
The detailed routing logic sits in how to set an approver rule for the support leadership. The matrix is the decision; the rule is how the workspace enforces it.
The template
| Ref | Action held by the agent | Agent depth | Approver role | Named approver | Limit | Second approver above | Certificate required | Conflicts | Review date |
|---|---|---|---|---|---|---|---|---|---|
| L-01 | |||||||||
| L-02 | |||||||||
| L-03 | |||||||||
| L-04 | |||||||||
| L-05 | Withheld at platform floor | n/a | No approver | None | n/a | n/a | n/a | n/a | n/a |
Add a footer line naming who owns the matrix, who signed it off, and where the current version is kept.
A worked example
Take a hypothetical subscription business with three support team leads, one support manager and a head of support. Its refund and goodwill agent holds the following actions. All values are sample figures.
| Ref | Action held by the agent | Depth | Approver role | Named approver | Limit | Second approver above | Certificate required | Conflicts |
|---|---|---|---|---|---|---|---|---|
| L-01 | Goodwill credit above the frontline limit | 3 | Team lead | Lead A, Lead B, Lead C | 250 | n/a | Current operator certificate | None |
| L-02 | Refund to the original payment method | 4 | Support manager | Manager D | 1,000 | 500 | Current operator certificate | Not the person who raised the case |
| L-03 | Final response to a formal complaint | 2 | Team lead | Lead A, Lead B | n/a | n/a | Current operator certificate | Not the lead who handled the original contact |
| L-04 | Closing an account with a remaining balance | 4 | Support manager | Manager D | 1,000 | Any balance | Current operator certificate | None |
| L-05 | Raising the customer limit on goodwill | Bound change | Head of support with agent owner | Head E | n/a | Always two | Current architect or builder certificate | Not the person who wrote the brief |
| L-06 | Refund to a card not used for the purchase | Refused by customer limit | No approver | None | n/a | n/a | n/a | n/a |
| L-07 | Reversing a settled payment | Withheld | No approver | None | n/a | n/a | n/a | n/a |
Three rows teach the most. L-02 shows the decision chain at work: a refund of 700 is within Manager D's limit but above the second-approver threshold, so it waits for two people, and the verdict in the record names the customer limit as the bound that decided. L-06 is a customer limit the business chose: the agent refuses, and a leader handles the case by hand. L-07 is withheld by the platform, so it appears in the matrix only so nobody wastes time looking for an approver.
If Lead B's certificate lapses, L-01 and L-03 route to Leads A and C until Lead B renews. Pair the matrix with a reviewer rota for support leadership so every row has cover on every shift, and with the AI agent competency matrix for support leadership so you can see who is ready to take on a higher limit. The wider picture of what each role must know is in AI agent skills by role.
Where the Academy fits
Pryme Intelligence Academy's Operator track trains team leads and support managers to read a held action, check its sources and refuse with a reason, then certifies them through a written exam and a two-week practicum of real review decisions. The certificate expires after 12 months and can be checked by number at source, so the certificate column in this matrix points to something verifiable.
Questions people ask
Should the head of support approve individual refunds?
Usually not. If the head of support routinely releases held refunds, the limits below them are set too low or the team is short of certified approvers. Keep the head of support for bound changes and exceptions.
What if a leader's limit for agent work differs from their normal sign-off limit?
It can be lower, never higher. Many teams start leaders on lower limits for agent-held items and raise them once the record shows sound decisions.
How do we stop leaders approving their own changes?
Use the conflicts column. Anyone who wrote or edited the agent's brief, sources or limits should not approve the actions those changes make possible.