Adventure Works: Chat Agents Drowning in Sessions
Adventure Works runs chat and SMS through Omnichannel for Customer Service with 80 agents and a target of five concurrent sessions per agent. During peak hours agents report six to eight simultaneous chats, while other agents sit idle with no conversations at all.
Average handling time has risen from 12 to 19 minutes. Some sessions stay open long after the customer has left, and those agents then receive no new work. The knowledge panel configured in the session template also appears blank for many agents, so they switch to a browser tab to search. The operations lead wants the session model fixed before the next peak season.
1. Agents receive more concurrent sessions than intended, and others receive none. Diagnose what controls concurrency and what you would verify.
Read the worked answer
Concurrency is controlled by capacity profiles, not by a single global setting.
Check for over-allocation:
- Confirm which capacity profile each agent is assigned. Agents on a profile with a higher session unit, or on the default profile, will accept more work than intended.
- Check whether different workstreams assign different units. If chat and SMS both draw on the same capacity pool, an agent can look compliant per channel while being over capacity overall.
- Verify the work distribution setting on the queues feeding the workstream.
Check for agents receiving nothing:
- Presence must be Available and not Do not disturb; an agent who closed the browser may be left in a non-routing state.
- The agent must be a member of the queue that the workstream routes to, and must hold the skills any routing rule requires.
- Their capacity must actually be free - stale open sessions hold units, which explains idle agents after customers leave.
Recommendation: set a deliberate capacity profile (for example three chat sessions) and test with a small pilot group before applying it to all 80 agents.
2. Many sessions remain open after the customer leaves. Explain the operational and configuration causes and how you would confirm the fix.
Read the worked answer
An open session holds capacity even when nobody is talking. That is why idle agents stop receiving work; the routing engine sees them as full.
Causes to investigate:
- Agents leaving the tab open or closing the browser without ending the conversation, so the session is never closed.
- No wrap-up or auto-close behaviour configured for the workstream, leaving the session to be closed manually.
- Session timeout settings that are longer than the real conversation lifecycle, or that were never tuned from defaults.
Fix:
- Enable and configure wrap-up time so agents get a short, bounded period after the conversation ends and the session then closes.
- Confirm the close-conversation action actually ends the session, and that agents know it is part of the workflow.
- Consider closing sessions automatically when the customer disconnects after a defined wait.
Confirm with data: track concurrent sessions per agent and average session duration before and after, and check that available-agent count during peak rises. If sessions still linger, sample individual sessions with their timestamps rather than relying on agent recollection.
3. The knowledge panel in the session template is blank for many agents. What would you check, and how would you redesign the template?
Read the worked answer
The blank panel is usually a context problem, not a knowledge problem. The panel needs something to search against - typically the linked case, the customer, or the conversation text. If no case is created or linked when the session starts, the panel has no anchor.
Check:
- Whether a case is created and linked automatically at session start for each workstream, and whether agents are expected to link one manually.
- Whether the article search control is filtered by product or language in a way that excludes results for these customers.
- Whether the agents who report blanks are on a different template version than the ones who see results.
Redesign the template around the conversation: include customer summary, the conversation panel, the linked case, and knowledge search positioned beside the conversation rather than behind another tab. Fewer, well-placed panels beat a crowded template - the goal is to remove the browser tab, which is where the real handling-time loss is coming from.
Measure article-link rate per session after the change, not just panel usage, so you can show the tab habit actually declined.