Governance and control
Human oversight in practice
How your people oversee your agents before, during and after every conversation, and how you can pause an agent or route a channel back to your team at any time.
On this page
- The agent says what it is
- Oversight before, during and after each conversation
- Watching live conversations
- Approvals and drafts
- Handover
- Quality review of every conversation
- Reviewing flagged conversations
- Regular reviews against agreed measures
- Pausing an agent or routing a channel back to your team
- In practice
- What your team controls
- Related
Your people stay in charge of your customer conversations. They set what the agents can do before a conversation starts, they can watch, approve and receive handovers while it happens, and they review what happened afterwards. And at any moment, your organisation can pause an agent or send a channel back to your own team. This page describes each form of oversight and who typically holds it.
The agent says what it is
Every customer knows they are talking to an AI agent. The agent identifies itself as an AI agent acting for your organisation, and offers a person whenever the customer asks for one. A customer who feels misled is a far bigger problem than one who knows what they are talking to and got their answer.
Oversight before, during and after each conversation
| When | Form of oversight | Who typically holds it |
|---|---|---|
| Before | Scope, limits, escalation rules and approval gates, set and changed through the controls your team holds | Operations and compliance leads |
| Before | Testing and approval of every change before release | Named approvers |
| During | Watching live conversations as they happen | Team supervisors |
| During | Approving actions that wait at an approval gate | Team supervisors |
| During | Reviewing drafts before they are sent | Your team, on the inboxes you choose |
| During | Receiving handovers, with the full conversation and a brief | Your team |
| After | Quality review of every conversation, scored against your standard | Quality team |
| After | Review of conversations flagged against your compliance rules | Compliance |
| After | Regular reviews against the measures you agreed | Operations lead, with Telonic |
| At any time | Pausing an agent or routing a channel back to your team | Named people in operations |
Watching live conversations
Supervisor monitoring lets a supervisor watch conversations as they happen, on any channel the agent handles. It is most used during the first weeks of a deployment and when a new workflow goes live, and it reassures teams that want to see the agent at work before they rely on it. Supervisor monitoring is configured during implementation for the teams that want it.
Approvals and drafts
Some actions should wait for a person. Actions you mark as sensitive, such as cancelling a reservation that carries a fee, are prepared by the agent and held until a person on your team approves them. The customer is told their request is with your team.
On email, the agent can prepare a reply and hold it for a person to approve rather than send it. This is a common first step for sensitive inboxes, and a way to build confidence before replies are sent directly. Drafts for review are configured during implementation for the mailboxes you choose. See Email: drafts for human review.
Handover
When a conversation reaches a trigger you have set, it goes to a person on your team with the full conversation and a written brief. Your colleague starts informed, and the customer does not repeat themselves. See Escalation rules and Handover to your team.
Quality review of every conversation
Every conversation is scored against your own quality standard, not a sample. Your quality team sees the scores in the Quality view of the console (the Telonic web application your team uses), and reviews the conversations that matter: the lowest scoring, a sample of the rest, and any a supervisor marks for review. Scoring every conversation means your people spend their review time where it is most useful. See Quality review on every conversation.
Reviewing flagged conversations
Compliance flagging checks conversations against rules your compliance team sets, and marks any that need a person to look at them: a customer who mentions a regulator, a complaint that was not recorded as one, a disclosure that should have been given. Flagged conversations are listed for review, and the reviewer records what they found and what was done. Compliance flagging is configured during implementation. See Compliance flagging.
Regular reviews against agreed measures
Before a deployment starts, you and Telonic agree what will be measured and what good looks like. Typical measures include how many conversations are resolved without needing a person, satisfaction on every conversation, handling time, first contact resolution, and the reasons customers get in touch.
You review those measures with Telonic at an agreed rhythm, more often in the first months. The review covers what is working, what is going to your team and why, and what should change. Changes agreed in a review follow the normal path to release. See How changes go live.
Pausing an agent or routing a channel back to your team
Your organisation can take any channel back at any time. On voice, the routing sits in your own phone system, so your telephony team can send calls back to your team by changing a routing rule on your side. In the console, a person with the right role can pause an agent, a workflow or a channel. While it is paused, new conversations go to the destination you have set for that channel, such as your team's queue or shared inbox.
Conversations already in progress when you pause are handed to your team with their brief, so no customer is left without an answer. Pausing and resuming are logged, with who did it and when. See Audit trail and decision records.
Decide during implementation who can pause an agent, and where paused conversations go on each channel. That way, the people who need to act in an incident already have the access to do so.
In practice
Rimal Air, a Gulf airline, goes live with a Telonic agent on booking changes for one market.
- For the first two weeks, two supervisors watch live conversations for part of each shift and mark any they want the quality team to look at.
- Refunds are held at an approval gate. The refunds team approves or declines each one in the console.
- The quality team reviews the lowest scoring conversations each morning. One pattern stands out: passengers asking about baggage on codeshare flights are getting a generic answer. The team proposes a change to hand those to a person.
- Compliance flagging marks three conversations in which a passenger mentioned a complaint to the aviation regulator. The compliance team confirms each was handed to the complaints team.
- One evening, the airline's reservation system has an outage. The duty manager pauses the booking change workflow in the console. New conversations go to the reservations team's queue, and the agent keeps answering flight status questions.
- At the end of the first month, the operations lead reviews the agreed measures with Telonic and approves extending to a second market.
What your team controls
- Who can watch live conversations, approve actions and review drafts.
- Your quality standard, and the compliance rules conversations are checked against.
- The measures you review, and how often.
- Who can pause an agent, and where conversations go when it is paused.