Adoption · Operating decision guide
Which service cases should AI resolve, draft or escalate?
Otomat editorial · Revised
Classify cases by the commitment the response could make and the evidence needed to make it. Give uncertain, sensitive or exceptional cases a named human path.
Start with the promise to the customer
A reply may describe a policy, approve a refund, change an account or promise a delivery date. Those actions have different consequences. List the permitted sources and approval authority for each. A fluent answer that invents an entitlement is not successful service, even if the customer initially rates it highly.
Make uncertainty an operational state
Give the workflow an explicit way to say that a source is missing, contradictory or out of date. Show the human reviewer the relevant evidence and the unresolved decision. Sending every uncertain case to an unstaffed queue is not a fallback. Define queue ownership, working coverage and how urgent work proceeds when the reviewer is unavailable.
| Check | Evidence to bring | Decision or stop condition |
|---|---|---|
| Policy explanation | Current approved policy and matching customer context | Draft or answer only within the supported policy. |
| Refund or account change | Identity confirmation and action-specific authority | Require authorization above the accepted limit. |
| Unknown or conflicting source | Evidence gap, urgency and staffed review owner | Escalate with context; do not invent an answer. |
| Repeat contact | Reopen reason and correction required | Count the full resolution rather than the first response. |
Test more than the happy path
Use synthetic cases for a policy change, an angry repeat contact, an ambiguous customer identity and a request to exceed refund authority. Include an attachment that attempts to change the assistant’s instructions. Check whether the system exposes information from another customer while constructing a helpful response. Compare drafts and final customer-visible actions, not just the first model answer.
Count accepted resolution
Measure reopened cases, downstream corrections, reviewer effort and elapsed time to accepted resolution. A fast automated response can create another contact later. Keep case mix and service coverage visible when comparing periods. Expand the resolution scope only after the service owner accepts the evidence; a drafting assistant can remain useful even when autonomous resolution is not appropriate.
Questions before the decision
Is deflection the right success measure?
It can be one measure, but should be checked against reopened cases, customer effort, errors and accepted resolution. A missing follow-up is not proof the issue was solved.
Can the same agent answer and approve a refund?
Only if that exact action has an authorized, tested boundary. Separating the answer from consequential approval is often a useful initial scope.
Take the evidence sheet to the meeting
Download the blank CSV, assign an owner to each check and record the evidence. Keep completed sheets in your own approved system.
Download the evidence sheet (CSV)Use the working tools
Blank templates are ungated. Adapt them to the workflow; inputs are not submitted.
Sources and limits
These are Otomat’s proposed operating methods and original decision artifacts. External references inform risk management; they do not certify this method or establish client results. Linked sources were selected for this review; their live revision dates may change.
For anonymous documentation and proposal patterns, see delivery evidence and its boundaries.
Related decisions
Bring a workflow to the discussion
Share the business goal and constraints through the mandate process. Do not send sensitive production data in the initial brief.
Book a Return Review