Home · Business · Operations & Logistics · SOPs & process docs

YES

As of 13 August 2026, AI can create a customer refunds procedure.

This still needs a person who signs their name to it.

Can you do it?

5 minutesto a draft.

30 minutesto something you’d act on.

Cost, all in£0

Skill neededchat-fluent

Who has to check ita colleague

What the alternative costsA process-documentation tool such as Scribe can turn a workflow you click through into a step-by-step guide, while a human can document the process from your supplied business information.

If this goes wrong: staff apply the wrong refund rule, customers are treated inconsistently, and your business may face complaints, lost money or an avoidable dispute.

What to actually do

  1. Hand it to a person

    The route this page recommends

    A person who owns the outcome does this end to end, worth it when the failure is dear.

  2. Use a tool built for this

    Second choice
  3. Do it yourself

    The distant third

    A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

    How to actually do it

    1. Open your current refund policy, terms and conditions, payment-system instructions, approval matrix and escalation list, then gather the latest versions in one folder.
    2. Paste the gathered material into the prompt, replacing each bracketed slot and removing customer names, payment details and other unnecessary personal data.
    3. Ask the chatbot to produce the procedure, decision tree, staff message templates and verification table in one document.
    4. Compare every drafted eligibility rule, deadline, approval limit and exception against the source policy and mark any mismatch for correction.
    5. Test the decision tree with at least one ordinary refund, one refusal, one payment-method exception and one ambiguous case, then record where a staff member would need to escalate.
    6. Ask a manager or process owner to check the procedure against the live payment systems and current terms, obtain specialist review for unresolved legal points, and publish the approved version with an owner and review date.

    Prompt

    Create a customer refunds procedure for a UK business using only the information below. Do not invent legal rights, deadlines, approval limits, system steps, fees or policy exceptions. Separate statutory consumer rights from our voluntary goodwill policy, and mark any point that needs checking against current UK law or professional advice.
    
    Business type and products or services:
    [INSERT]
    
    Our current refund, cancellation and returns policy:
    [INSERT]
    
    Relevant terms and conditions:
    [INSERT]
    
    Payment methods and systems used:
    [INSERT]
    
    Who can approve refunds, and approval limits:
    [INSERT]
    
    Common refund reasons and examples:
    [INSERT]
    
    Escalation contacts and response targets:
    [INSERT]
    
    Records we must keep and where they are stored:
    [INSERT]
    
    Write the procedure for front-line staff. Include: purpose and scope; definitions; a decision tree for common cases; step-by-step instructions for checking eligibility, approving or refusing a refund, issuing it in each payment system, and recording it; customer message templates for approval, refusal and escalation; fraud or repeat-claim warning signs; exceptions; quality checks; and version-control fields. Use plain UK English. Put unknown information in [CHECK] brackets rather than guessing. End with a table of every factual, legal or operational assumption that a manager must verify before publishing. This is not professional advice.

    Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

What it gets wrong

  • AI cannot know which refund rules your business actually intends to apply unless you provide the current policy and examples.
  • AI cannot reliably resolve unusual cases involving disputed rights, misleading information, damaged goods, digital content or vulnerable customers without appropriate specialist judgement.
  • AI cannot confirm that its system instructions match the live payment platform, order system or approval workflow.
  • AI cannot take responsibility for an unlawful refusal, inconsistent treatment or money lost through an incorrect refund decision.

Even on a YES, the friction has a name: legal accountability, judgement under ambiguity and context depth.

How we scored this

Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.

AxisScore (0–2)
Output2
Inputs2
Verification1
Liability1
Effort delta2
Total8 / 10

FAQ

Can ChatGPT write a customer refunds procedure?
Yes. It can turn your refund policy, systems and approval rules into a structured procedure, decision tree and set of customer messages. You must check the result against your live process and obtain specialist review for unresolved legal points.
Can AI decide whether I should give a customer a refund?
It can apply decision rules that you have supplied, but it should not make unreviewed decisions in ambiguous or disputed cases. Your business remains responsible for the decision and its consequences.
What information does AI need to write a refunds procedure?
Give it your current refund and cancellation policy, relevant terms, payment systems, approval limits, common refund reasons, escalation contacts and record-keeping requirements. Remove unnecessary customer personal data and tell it to flag missing information instead of guessing.
Do I need to check an AI-written refunds procedure?
Yes. Compare every rule and system step with your current documents and live workflow, test it against representative cases, and have a manager approve it before staff use it. A serious or legally uncertain case needs appropriate professional advice, not an unchecked chatbot answer.

Nearby answers

Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.

The newsletter

AI news, new answers and product picks, straight to your inbox.