Skip to scenario
Back to the business caseAutomated ticket resolution · Implementation

Separate policy evidence from permission to act

The assistant may handle the conversation, but the policy and action integration must establish what it may do. Uncertainty needs a supported handoff.

Follow the process
Business goal
Extend scope (expense)
Business metric
Self-service rate

Assess, authorize, act, and confirm

Use an approved policy revision, reliable billing records, enforced tool permissions, and a staffed escalation route. Test without issuing transactions first.

The task in this example

We were charged a late fee on an invoice we paid on time. Can you refund it?

  1. Step 1

    Recognize the request

    Map the fee dispute to the late-payment policy.

    Taxonomy

    Maintain customer language and policy identifiers together.

  2. Step 2

    Read billing facts

    Retrieve the invoice, payment date, and plan.

    Product or customer data

    Billing records determine whether the payment falls within the approved grace period.

  3. Step 3

    Retrieve current policy

    Select the policy for the account and effective period.

    Content team

    Retain scope and effective dates with eligibility rules and exception handling.

  4. Step 4

    Check action limits

    Evaluate whether this case may be refunded automatically.

    Content team

    State authority conditions explicitly.

    System

    Enforce those limits in the action layer; the model’s confidence is not authorization.

  5. Step 5

    Act or hand off

    Issue the permitted refund or transfer the case.

    System

    Record the transaction result and prevent duplicate actions on retries.

  6. Step 6

    Confirm and monitor

    Explain the recorded result and observe reopens.

    Model

    Do not claim a completed refund before the billing tool confirms it.

    People

    Support owners review exceptions and policy errors.

This framework provides a practical baseline shaped by experience. Every organization operates under distinct technical and operational realities that require tailored design and governance. While you can adapt these measures independently, we frequently partner with teams to customize governance, integrations, and rollout. Performance outcomes will naturally vary.

Why these content layers matter

Intelligent content

Maintain policy, effective date, eligibility conditions, explicit limits, and handoff instructions as one controlled unit.

Taxonomy

Map customer phrases such as “late fee” and “overdue charge” to the policy and classify its plan and region scope.

Verify the result and handle exceptions

Test an eligible fee, a payment outside the grace period, missing data, an action beyond authority, and a repeated request. Confirm the recorded outcome rather than relying on the final chat message.

Missing or conflicting evidence, exceeded authority, or a failed tool call requires a handoff. Preserve context and keep the case unresolved until the appropriate outcome is confirmed.

How we work
with your team

We can guide your team, work alongside it, or deliver the work for you. Combine these approaches as your needs and capacity change.

Lead

Build capability

We guide your people through the work, building capability that stays with your team.

Support

Add specialist backup

Your team owns the work. We provide experienced backup and specialist help where you need it.

Do

Delegate delivery

We handle work that is one-time, specialized, or outside your core business, without adding an ongoing burden to your team.

Which routine cases have rules clear enough to automate?

Intuitive Stack can help structure policy evidence and action boundaries with your support and system owners, then define a controlled self-service pilot.