Start with one repeated customer request that has a clear outcome, a known policy and a named owner.
Observe real work. Review a few recent cases with a support agent. Capture the information they need, the decisions they make and the exceptions they escalate.
Write the procedure together. Describe the trigger, the required context, the allowed actions and the handoff. Ask a team lead to challenge the assumptions.
Test before expanding. Include ordinary cases, missing information and requests outside the policy. Compare the proposed action with the reviewed expected outcome.
Begin with human review. Keep a person responsible for consequential actions. Expand the scope only when the evidence supports it.
Quality framework / 02
How to evaluate human and AI service quality
Use shared definitions when comparing human and AI work. Compare similar request types and separate policy failures from missing information.
Resolution quality: was the request actually completed, and was the action correct?
Customer satisfaction: collect feedback in a consistent way and distinguish feedback availability from the score itself.
Escalations: was the handoff appropriate, timely and accompanied by enough context?
Cost per resolved request: include the human work required for review, corrections and follow-up.
Review a sample of conversations together. Keep the evidence behind each assessment and turn disagreements into clearer evaluation rules.
Team playbook / 03
A practical guide to permissions and human handoffs
Define permissions around the action and its consequences. Reading an order, preparing a change and executing a refund need different decisions.
Name the owner. Every workflow needs someone who can approve policy changes and handle exceptions.
Separate read, prepare and execute. Give the agent only the access required for the agreed workflow.
Specify the handoff conditions. Examples include missing context, policy exceptions, a value threshold or a request the agent cannot reliably handle.
Make review useful. Show the proposed action, supporting information and reason for escalation, so the reviewer can decide.
Learn from reviewed outcomes. Update the procedure and retest affected scenarios before rollout.
Blog / Field notes
The best starting point is a real case.
A useful discovery conversation starts with a recent customer request. Ask the team to walk through what happened, where they switched tools and how they decided what to do. That walkthrough exposes the context an agent will need and the judgment that should stay with a person.
Bring the support agent, team lead and business owner into the same review. Agreement on the procedure is part of the work of building the agent.
Webinars & walkthroughs
Explore a workflow at your own pace.
Our current walkthrough is an interactive browser simulation. Explore a return request, customer permissions and the decisions behind an approved action.
We’ll share customer stories when there are real deployments and reviewed outcomes to report. For now, explore the illustrative workflows and bring your own customer care challenge.