Support leadership needs a smaller rota with sharper edges. Name a duty manager each day for escalations that team leads cannot decide, keep an on-call line for incidents where an agent is giving customers wrong answers at scale, and hold a weekly slot to approve changes to agents' briefs and limits.
Each line has its own response target and its own authority.
What reaches support leadership?
Team leads handle most of what a support agent holds. What comes up to leadership is different in kind: refunds and credits above the team leads' authority, complaints at the final response stage, account closures with legal or reputational risk, threats of legal action, complaints that are spreading publicly, and requests to make an exception to policy. Leadership also owns changes to how the agent works: its brief, its sources, and the limits within which it acts.
The head of support owns the rota. The guide to AI agent skills by role covers what support leaders need to know, and what a reviewer rota is explains the term. The team-level rota beneath this one is described in how to build a reviewer rota for the support team.
What are the steps?
- Agent, depth 1: reports escalations by type and age. And the changes requested by team leads.
- Person (head of support): sets response targets. Escalations within a working day; final responses within the timescale policy or regulation sets; incidents acknowledged within minutes during service hours.
- Person (head of support): names a duty manager each day, with a deputy. They take escalations from team leads.
- Person (head of support): names an incident on-call manager. For each period the agent serves customers.
- Person (head of support): sets a weekly change review. Requested brief and limit changes, each with the evidence behind it.
- Agent, depth 2: drafts responses for escalated cases. With the history and the agent's earlier verdicts.
- Person (duty manager): decides each escalation. With reasons, within authority; above authority to the head of support.
- Person (incident manager): leads any incident. Contain, communicate, fix, and record.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Escalations | Drafts responses with history | 2 | Duty manager decides |
| Final responses | Drafts the response | 2 | Approval and sending |
| Account closures | Prepares and holds | 4 | Approval by named manager |
| Policy exceptions | Refuses; routes up | n/a | Exception decided by authority |
| Incidents | Files every verdict as it happens | 1 | Incident manager leads |
| Brief and limit changes | Works to the current version | n/a | Weekly review approves or rejects |
What does the incident line do?
An agent working at scale can repeat a mistake hundreds of times before anyone notices: quoting an old price after a change, giving outdated delivery times, refusing a refund it should allow because a source was updated wrongly. The incident manager's job is to stop it, find what changed, correct the source or brief, and decide what to tell affected customers.
Speed matters, which is why the line is on-call rather than "whoever notices". The record helps: every verdict is filed, so the incident manager can find every customer affected by the faulty answer rather than guessing. In the most serious cases, the fastest containment is to have the agent's depth lowered for that capability until the fix is certified.
Which controls matter, and what evidence do they leave?
- A duty manager is always named. Evidence: the rota.
- Escalations are decided within authority. Evidence: the approver and authority limit on each verdict.
- Incidents have an owner from the start. Evidence: incident record with manager, times and actions.
- Changes are reviewed with evidence. Qualification, not training: what changes is the brief, sources or bounds, never model weights. Evidence: the change, its evidence and the approver.
- Exceptions are recorded. Evidence: each exception, its reason and its approver.
What should be measured?
- Escalations decided within target.
- Final responses sent within the required timescale.
- Incident time to containment and time to fix.
- Changes approved at the weekly review, and whether they reduced the hold reason they targeted.
- Exceptions granted, by reason; a repeated reason means the policy needs changing.
The AI agent competency matrix for support leadership and the approver authority matrix for support leadership set who may sit on each line.
Where does it usually go wrong?
Changes are made between reviews. A leader edits the agent's refund wording during a busy afternoon. It may help; it is also an unrecorded change to how the agent behaves. Route every change through the review.
Incidents are found by customers. If nobody watches for repeated complaints about the same answer, the first sign of a problem is social media. The on-call manager should also review the hold and complaint trends daily.
The duty manager is in meetings all day. Escalations age. Keep the duty manager's diary light.
Fraud teams run incident and duty lines of their own; compare the rota for the fraud team.
Where the Academy fits
Support leaders deciding escalations and approving changes need the Academy's Operator skills, especially the modules on teaching the agent and when something goes wrong. Leaders who design or change agents' briefs can go further with the Builder track, which ends in Certified Pryme Builder, valid for 12 months.
Questions people ask
Should the duty manager also handle incidents?
In small teams, sometimes. If so, the rota should say so, and a deputy should take escalations while an incident runs.
Can team leads make small brief changes without the weekly review?
Only if leadership has set a category of small change that may be made between reviews, with a record. Otherwise every change waits for review.
Who decides whether to contact customers after an incident?
The incident manager, with the head of support for anything significant. The record shows exactly which customers received the faulty answer.