Audit intermediate Works with: Any AI

Refund Decision Consistency Audit Across Your Team

Table output Has example

Check that similar complaints get similar outcomes.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Audit the consistency of refund decisions. A sample of past cases with the complaint, evidence, decision and amount: [CASE_SAMPLE]. My written guidelines: [GUIDELINES]. The staff who decide: [DECIDERS].

Group the cases by type and compare decisions. Make a table with Case, Type, Evidence, Decision, Amount, Matches guidelines, and Concern. Identify cases where similar facts led to different outcomes, decisions with no recorded reason, and amounts that deviate from the guidelines. Calculate the share of consistent decisions. Propose a decision matrix (situation, evidence needed, standard remedy, who can approve exceptions), a short reason code list, and a monthly sample review routine. Add a note on documenting exceptions.

Use only the cases provided. Do not recommend remedies that break my stated policy without flagging it.

This is not legal advice. Cite no time limit you cannot source. Use [JURISDICTION] and [RELEVANT_LAW_OR_POLICY_IF_KNOWN] as given. Flag where I need expert help. Promise no result.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

Cases: 20. Matches guideline: 14 (70 percent). Similar cases, different outcomes: 3 (damaged on arrival: full refund by Alex, 50 percent by Sam, replacement by Dana). No reason recorded: 4. Decision matrix: situation, evidence, standard remedy, who can approve exceptions.
Reason codes: damaged, wrong item, late, changed mind, other. Monthly sample: review 10 cases and discuss.
Outcome of the review: the three damaged-on-arrival cases will be set to full refund or replacement at the customer's choice, and the four cases with no reason recorded will be completed from the ticket history. Re-test next month on a fresh sample of 10.

• worksheet