Idle Agents, Waiting Cases

Ranking by how much of the 60 cases each cause explains:

1. Eligibility, not capacity. 40% have no assignment group for the first 15 minutes, and closed cases are in the queue. AWA routes work items that match a queue; these do not match, or match and should not. This explains roughly half the backlog on its own.
2. Skill supply. Four of nine Billing agents cannot take Billing work, so effectively five agents cover the queue - which is why some agents are idle while others are saturated.
3. Presence/role. Two agents may lack the roles needed to receive and open work items; presence alone does not make them routable.

First three checks:

• Open the queue configuration: work item conditions, skill requirements, and which assignment groups map to which service channel. Confirm closed cases are excluded.
• Open the AWA agent records for the idle agents: presence, capacity used vs capacity, assigned skills, and whether they are members of the queue's group.
• Check the idle agents' roles and whether their inbox in the workspace renders at all - a role gap looks identical to 'no work available' from the agent's screen.

Only after those three would I look at case volume or the routing algorithm.

Fix the source, not the filter.

The tempting fix - loosening queue conditions so unassigned cases match any queue - spreads work unpredictably and defeats skills-based routing.

1. Set the assignment group at insert via the case type default, or a before-insert rule driven by category/product, so the 15-minute flow is no longer needed for routing. Keep the flow only for cases whose group genuinely cannot be known at creation.
2. If a deferred decision is unavoidable, use a holding queue with its own service channel and a short re-route timer, rather than letting the case sit outside all queues. Agents on the triage queue then own it visibly instead of it appearing to be nobody's work.
3. Add a queue condition that excludes terminal states (Closed Complete and equivalent), and review any other queue that could pick up closed records.

Verify with a day of traffic that the 'unassigned with no queue' count goes to zero, and that no case is offered to an agent lacking the required skills.

Measure the pipeline, not just the outcome.

• Coverage per queue: count of available agents holding the full required skill set, versus open work items in that queue. A ratio below 3:1 is an early warning that a queue will starve - this is the metric that would have flagged four of nine Billing agents being ineligible before Monday.
• Work items with no assignment group older than the routing window; trend it and alert on any rise.
• Queue wait time percentiles (not just average), per queue, because the average hides a long tail.
• Agent utilisation and idle-with-queue-non-empty, which is the direct symptom of a configuration mismatch.
• A weekly check that each queue's skill requirements are held by at least the number of agents the forecast needs.

Put these on a workspace dashboard the team lead actually looks at, and set a threshold alert rather than a monthly review.