Plan requirement
| Subscription | Suite Professional or higher, Explore Professional or higher |
| Access | Agent |
What the AI agent takes off the queue, measured in a way that survives a rise in traffic, and the third outcome most reports leave out.
Build it
- Use the AI agents dataset, where the work is already done: Conversations, Automated resolutions, Escalations and Drop-offs, with % Resolution rate, % Escalation rate and % Drop-off rate as each of those divided by conversations.
- Group by month, and by channel.
- Add tickets reaching agents on the same subjects, from the ticket dataset.
Share, not count
Resolutions rise when traffic rises. Only the share tells you whether the agent got better, which is why % Resolution rate belongs on the dashboard and the raw count does not.
Read all three outcomes
Every conversation ends as resolved, escalated or dropped, and it is the third one people forget. A drop-off is a conversation that was neither resolved nor passed to a human: the customer left. A high resolution rate next to a high drop-off rate is not a good result, it is a queue that emptied because people gave up.
Escalation rate is the least worrying of the three. A conversation that reaches a person is a conversation that was handled.
The comparison that matters
Human ticket volume on the subjects the AI agent handles, before and after it went live. That is what the business is asking: did work leave the queue.
Resolution counts alone do not answer it, because an AI agent can handle conversations that would never have become tickets.
Add repeat contact
Customers coming back within a day on the same subject. Alongside the drop-off rate this is the check on whether resolved means solved, because a conversation can end tidily and still leave the customer needing you tomorrow.
One distinction to hold on to while you read the rest of the dataset: a generative reply is any response the AI agent produced, greetings included, while a generative answer is one that used your content. The second is the one that says something about your knowledge base.
Split by channel
Messaging and email behave differently enough that a combined figure describes neither. Email is where a poor answer costs most, so it deserves its own line.
Tag the tickets
If AI-handled conversations carry a tag, every future report is straightforward. Adding it costs one trigger and saves a lot of inference later.
Report it with what it cost
Resolutions are billed. The honest report puts the volume handled next to that, so the conversation about value is based on both halves.
Comments
0 comments
Article is closed for comments.