Governance and control
Decision boundaries
The three zones every request falls into, what the agent never decides, and how those boundaries are enforced.
On this page
Some decisions belong to your people and should stay there: a claims decision, a refund outside policy, a goodwill gesture, anything heading towards a legal dispute. Telonic draws that line with you before a deployment goes live, and builds it into what the agent is able to do. The agent handles the conversation and the process around a decision. The decision itself stays with your team, and the agent has no way to make it.
Three zones
Every request a customer makes falls into one of three zones. Because the agent is built for your industry, it recognises which zone a request belongs to, and your team decides where the lines between them sit.
| Zone | What the agent does | Who decides |
|---|---|---|
| 1. Handles | Completes the request within your limits, using information from your systems | The agent, within limits your team set |
| 2. Prepares and hands over | Gathers the facts, pulls the history and explains what your policy says. When the customer wants a decision, it passes the conversation to a person with a written brief | A person on your team |
| 3. Stays with your team | Recognises the request, does not attempt it, and routes it straight to the right person | A person on your team. The agent has no access to the decision |
What the agent never decides
In each industry, some decisions stay with your team in every deployment. For some, the agent gathers the facts first (zone 2). For others, it routes the customer straight to a person (zone 3). These are the defaults we start from. Your team can add to them, and none of them moves to the agent without your agreement.
| Industry | Decisions that stay with your team |
|---|---|
| Property | Price negotiation, cancellations and refunds, changes to a payment plan, disputed snagging items (defects reported at handover), anything heading towards a legal dispute |
| Travel and hospitality | Refunds and compensation outside policy, goodwill gestures, anything involving safety, health or a guest in distress, VIP guests who expect their own contact |
| Insurance | Every claims decision: assessing liability, setting a settlement amount, approving or declining a claim. Medical judgements. Anything that looks like fraud. Claims involving serious injury or bereavement. Any complaint already with the regulator |
In insurance, the boundary is absolute. The agent takes the first notice of loss (the first report of a claim), answers the policyholder's questions from the approved policy wording, requests documents and keeps the policyholder updated. It does not assess liability, set settlement amounts, or approve or decline a claim.
How the boundaries are enforced
A boundary written as an instruction depends on the system following it. Telonic enforces boundaries in four ways. The first three remove the agent's ability to act. The fourth checks every turn of the conversation for topics outside scope.
Decision actions are not available to the agent. The functions that approve a claim, issue a refund outside policy or change a payment plan are not connected to the agent at all. There is nothing for it to call, however a customer phrases the request.
Permitted values are fixed. Where the agent can act, the values it can use come from your rules. It can offer the rebooking options a fare permits, or a voucher within your policy limit, and nothing beyond them.
Sensitive actions wait for approval. Actions you mark as sensitive are prepared by the agent and held until a person on your team approves them.
Topics outside scope are not discussed. Subjects you place out of scope, such as legal advice or investment advice, are declined and routed to a person, including when a question is rephrased. Scope is tested before every release.
The technical detail
Each connection to your systems exposes a defined set of actions, and each action has its own permission, set with your IT team. Read access and write access are granted separately, system by system. An action that is not granted cannot be invoked by the agent. See Permissions.
Escalation triggers are defined per workflow. They include the customer's request, what the agent detects in the conversation, such as distress or frustration, and cases where the agent is not confident enough in a consequential answer. Escalation combines rules set per type of question with a confidence threshold measured for the model before deployment. When a trigger fires, the conversation goes to the person or queue you have named, with the full history. See Escalation rules.
Every request, lookup, decision and handover is logged, with the sources used, the rule or trigger that applied, and the agent's stated reason. When a customer or a regulator asks what the agent did and why, you can show them. See Audit trail and decision records.
When a request crosses a boundary mid-conversation
Conversations do not stay in one zone. A customer checking a claim's status may go on to ask why part of it is not covered, then say they intend to complain.
The agent moves with the conversation. It answers the status question itself. For the coverage question, it explains what the policy wording says and records the question.
When the customer says they intend to complain, it hands the conversation to your team with a brief that covers all three. The customer does not start again, and the person picking up can see exactly where the conversation crossed each line.
In practice
A motor insurer in the UAE sets its boundaries with Telonic before going live on its claims line.
- The claims and compliance teams agree the defaults: every claims decision stays with the claims team, and complaints go to the complaints team.
- They add one boundary of their own. Any claim where the policyholder mentions an injury goes straight to a senior handler, whatever else is said.
- The insurer's IT team grants the agent read access to policies and claims, and permission to create a new claim and attach documents. No permission to change a claim's status or reserve (the amount set aside to pay a claim) is granted.
- At 10pm, a policyholder calls to report a collision and mentions a sore neck. The agent checks he is safe, takes the details the claim needs, and routes him to the on-call senior handler with a brief, as the insurer specified.
- The next morning, the compliance team opens the conversation and sees the injury mention, the trigger it matched, and the time of the handover.
What your team controls
- Which requests sit in each zone, and which stay with your team in every case.
- The escalation triggers for each workflow, and who receives each handover.
- Which actions in your systems the agent can take, and which need a person's approval.
- The topics the agent never discusses.
Boundaries are set with you during implementation and changed through the controls your team holds, with every change tested before it goes live. See The controls your team holds and How changes go live.