Over-caution shows in the record as declines that peers would have approved, vague reasons, and decisions overturned on escalation. Regrade a sample of the reviewer's declines, show them what each cost the customer or the business, find out what they fear, then clarify their authority and coach them to decide on evidence. Re-measure after a set period.
Why is declining too much a problem?
Declining feels safe. Nobody is blamed for a payment that was not released or a refund that was not given. But each unnecessary decline has a cost: a customer who waited and then got the wrong answer, a supplier paid late, a colleague who has to reopen the case, a complaint. And a reviewer who declines whatever looks unusual teaches the team that holds are dead ends, which pushes people to find routes around the queue.
Over-caution also hides a skills gap. A reviewer who cannot tell what the agent checked, or why it held, will decline to be safe. The team lead or reviewer's manager owns the coaching. The guide to AI agent skills by role sets out what reviewers should be able to do, and what an AI agent reviewer is describes the role's purpose: to decide, not to stop.
What are the steps?
- Agent, depth 1: reports each reviewer's decisions by hold reason. Approvals, edits, declines and escalations; time per decision; the reasons given.
- Person (team lead): compares with peers on the same hold reasons. Over the same period, so differences are about the reviewer, not the work.
- Person (team lead or quality reviewer): regrades a sample of declines. Marked as correct, should have been edited and approved, or should have been approved.
- Person (team lead): reads the reasons. "Not comfortable" or "insufficient information" without saying which information are signs the reviewer did not know what to check.
- Person (team lead): traces what happened next. Declines overturned on escalation, cases reopened, complaints received. The record links them.
- Person (team lead): has the conversation. Show three or four regraded declines and what followed. Ask what the reviewer was worried about.
- Person (team lead with reviewer): agrees a coaching plan. Clear authority, the evidence to check per hold reason, and a paired-review period.
- Agent, depth 1: reports the same measures after the period. The team lead regrades a fresh sample.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Decision report | Counts, times and reasons by reviewer | 1 | Reading the pattern |
| Peer comparison | Reports by hold reason | 1 | Choosing fair comparisons |
| Sample regrade | Draws the declines | 1 | The regrading judgement |
| Downstream trace | Links declines to reopenings and complaints | 1 | Weighing the cost |
| Coaching plan | Nothing | n/a | Agreeing behaviours and dates |
| Re-measure | Reports the same measures | 1 | Fresh regrade and decision |
What usually causes over-caution?
Fear of blame. The reviewer believes they will answer for anything that goes wrong after they approve, but not for what goes wrong after they decline. Address this directly: they answer for the quality of their decision, in either direction, and the record shows both.
Unclear authority. Reviewers unsure whether they may approve something decline it instead of asking. Show them their authority on the approver list and the matrix behind it. If the authority really is unclear, fix the matrix.
Not knowing what the hold means. A held action is not a suspected problem; it is an action that reached a bound. A refund held because it exceeds the agent's limit needs a check that the refund is right, not suspicion that it is wrong. Teach reviewers to read the deciding bound first and check what it points to.
Misreading provenance. Some reviewers decline anything resting on declared facts. Declared is not false; it means the fact needs the check policy sets for it. Coach on what that check is.
Which controls matter, and what evidence do they leave?
- Comparisons are fair. Evidence: the peer comparison by hold reason, kept with the coaching record.
- The regrade is independent. Evidence: the regrader's name and marks.
- The plan is written. Evidence: the reviewer coaching plan template completed and agreed.
- Escalation stays available. The coaching should never make a reviewer feel they must approve. Evidence: escalations with specific reasons are counted as good decisions in the re-measure.
What should be measured?
- Regraded declines that should have been approved, before and after.
- Declines overturned on escalation.
- Reasons that name the specific missing evidence, as a share of all declines.
- Time per decision; over-cautious reviewers are often slow as well.
- Complaints and reopenings traced to the reviewer's declines.
Where does coaching usually go wrong?
Pushing the rate down. The goal is right decisions, not fewer declines. A reviewer coached to approve more without learning what to check will swing to the opposite fault; coaching a reviewer who approves too much covers that pattern.
Ignoring the work mix. Some queues should produce more declines. A reviewer on an advice team's suitability holds may rightly decline more than one on routine refunds; compare within the same queue, as a rota like the one for an advice team makes easy to do.
Leaving the fear unaddressed. If the reviewer still believes approvals carry all the risk, they will drift back.
Where the Academy fits
The Academy's Operator track teaches how a governed agent decides and how to review its holds, so reviewers learn what each hold means and what to check. Module tests are set in the learner's own industry and role, and the practicum marks every decision, approvals and declines alike, against the record.
Questions people ask
Is a high decline rate always a problem?
No. Some queues hold mostly actions that should be declined. Over-caution is a decline rate higher than peers on the same hold reasons, confirmed by a regraded sample.
Should an over-cautious reviewer lose approval rights?
Rarely. Over-caution causes delay and cost but seldom a red-line miss. Coaching and paired review are the usual response; narrowing rights is for reviewers who approve badly.
What is the difference between declining and escalating?
Declining is a decision that the action should not happen. Escalating is passing the decision to someone with more authority or knowledge, with a reason. What an agent reviewer is covers both.