An operations rota is a grid: queues down one side, reviewers across the top, and a mark wherever a person is trained and authorised for that queue. Fill each day from the grid and name a floor lead to move people as holds shift.
Rotate reviewers so nobody approves the same queue all week, and keep a contingency plan for the day an agent drops to depth 1.
What makes an operations rota harder than a single-team one?
A back-office operations team often runs several agents at once: one handling account maintenance, one processing standing instructions, one reconciling exceptions. Each holds work for different reasons and needs different skills. The operations manager must staff all of them from one team, with people whose training covers some queues but not others.
The operations manager owns the rota. The guide to AI agent skills by role sets out what each reviewer needs, and what a reviewer rota is gives the basic idea. A ready-made layout is in the review queue rota template.
What are the steps?
- Agent, depth 1: each agent reports holds by reason, hour and age. Combined into one view across queues.
- Person (operations manager): builds the skills grid. For each reviewer, the queues they are trained and authorised for, and the depth they may approve.
- Person (operations manager): sets response targets per queue and hold reason.
- Person (operations manager): estimates capacity. How many holds of each type a trained reviewer decides well in an hour, from the record; not how many they could rush through.
- Person (operations manager): fills each day from the grid. Every queue has a primary and a backup; no reviewer is on more queues than they can watch.
- Person (floor lead): moves people intraday. Named each day, watching hold ages across queues.
- Person (operations manager): rotates reviewers weekly. Nobody approves the same queue for weeks on end; fresh eyes catch what familiarity misses.
- Person (operations manager): keeps the certificate-drop plan current. For each agent, who absorbs its work if it falls to depth 1.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Hold reporting | Each agent reports its holds | 1 | Reading the combined view |
| Routine work | Each agent works within bounds | 3 | Sampling quality |
| Exceptions | Each agent drafts and holds | 2 | Decisions by grid-qualified reviewers |
| Money or irreversible actions | Prepares and holds | 4 | Named approvers only |
| Certificate drop | Falls to depth 1 the same day | 1 | Contingency staffing |
| Intraday shifts | Keeps holding until someone acts | n/a | Floor lead moves people |
How should the rota handle a certificate drop?
Agents' certificates lapse after 90 days, and a failed Attest run drops an agent to depth 1 the same day. An agent that was completing hundreds of routine actions at depth 3 now only reads; everything it used to do becomes work for people or waits. For an operations team, that is a sudden surge with no warning.
The contingency plan names, for each agent, which reviewers take over its work, which lower-priority queues pause, and who requests recertification. It also names what must not happen: people approving in bulk what the agent can no longer do, without checking each item. The plan sits with the rota, and the floor lead activates it. Recertification dates are known in advance, so a lapse should never be a surprise; a failed run can be.
Which controls matter, and what evidence do they leave?
- Only grid-qualified reviewers decide each queue. Evidence: the reviewer on each verdict appears on the grid for that queue and depth.
- Every queue has a primary and backup each day. Evidence: the rota.
- Rotation happens. Evidence: the rota history shows reviewers moving between queues.
- The floor lead's moves are recorded. Evidence: a short intraday log of who moved where and why.
- Certificate drops trigger the plan. Evidence: the time of the drop, the time the plan was activated, and the actions taken.
What should be measured?
- Response targets met per queue.
- Reviewers' decisions per hour, alongside a quality sample, so capacity estimates stay honest.
- Queues with only one qualified reviewer.
- Time from certificate drop to contingency in place.
- Items approved in bulk during contingencies, checked for quality.
Where does it usually go wrong?
The grid is out of date. People are trained, leave or change role, and the grid lags. Review it monthly.
Capacity is set by the fastest reviewer. If one person decides sixty holds an hour, it does not mean sixty is right. Set capacity from decisions that pass the quality sample.
No rotation. Reviewers on one queue for months stop seeing it. Errors become habits.
Payments operations and legal operations teams apply the same grid with narrower queues; compare the rotas for the payments operations team and the legal operations team. For the reviewer's own role, see what an agent reviewer is.
Where the Academy fits
Operations managers are among the people the Academy's Operator track is written for. Its modules on reading the workspace, when something goes wrong and running the rhythm cover the skills grid, intraday moves and certificate-drop contingencies. Certified Pryme Operator is valid for 12 months, and the team admin view shows each reviewer's certificate and expiry, which helps keep the grid current.
Questions people ask
How many queues can one reviewer cover at once?
As many as they can watch without hold ages slipping, which the record will show. For most people that is one or two busy queues, or a few quiet ones.
Who should be the floor lead?
An experienced reviewer qualified on most queues, with the authority to move people. Rotating the role among senior reviewers builds the skill across the team.
Should the rota change when an agent's depth is raised?
Yes. A higher depth usually means fewer holds of one kind and more consequential holds of another. Revisit capacity and the grid whenever depth changes.