SupportSaves about 5 minutes per escalated caseTested on Claude Sonnet 5.5 and GPT-6, Nov 2026

Write a reply that hands a case to a human agent

What this does

This playbook helps a support team write a customer reply and an internal handoff note in one step. You paste the customer message, the policy and your escalation rules. The AI suggests whether a person is needed, writes a reply that promises nothing the policy does not allow, and records the reason, urgency and what was already tried.

Fill in your detailsDownload skill
On this page

Before and after

Before

  • An agent reads the message, looks up the policy, decides whether to pass the case on and writes a reply.
  • The handoff note is often a single line, so the next agent asks the customer the same questions again.
  • Replies sometimes promise a refund or date that the policy does not allow.

After

  • The agent pastes the message, policy and rules, and gets a decision, a draft reply and a structured note in one pass.
  • The agent checks the decision, settles who leads the case and checks the promises, then sends.
  • The next person can pick up the case from the note without going back to the customer.

Saves about 5 minutes per escalated case

From my practice

I see handoffs go wrong when the customer is promised something the next agent has to undo. A good reply says what happens next and who does it, and leaves out refunds and dates the policy does not allow. The note should save the agent from asking the customer the same questions again. A person reads both before sending.
· AI workflow and adoption consultant

What to have ready

  • The customer message, pasted as received, with details you do not need (full address, card or bank details) removed.
  • The policy text that applies to this case, such as returns, refunds, delivery or warranty, copied from your own document.
  • Your escalation rules: when an agent must take over, who receives the case and how urgent each type is.
  • Case facts from your helpdesk or order system: dates, order value, earlier contacts and anything already tried.
  • The channel (email, WhatsApp or chat), the tone you want and how you sign off.

Build your prompt

  1. 1. Gather your inputsHave the items above to hand.
  2. 2. Fill in the formYour prompt updates as you type.
  3. 3. Copy or open in your AIClaude, ChatGPT or Gemini.
  4. 4. Review before you sendCheck it against your notes.
Fill in your details

Fill these in. Your prompt updates as you type, and stays in this browser tab.

Your prompt · 0 of 6 filled

You are helping the customer support team at Your business name and city. Read the customer message, decide whether a human agent must take over, then write a reply to the customer and an internal handoff note. Customer message: The customer's message, pasted as received Case details and what has already been tried: Facts from your order or helpdesk system, including what has already been tried Policy that applies: The policy that applies, copied from your document Escalation rules for human agents: Your rules for when a human agent must take over Tone and channel for the reply: Channel, tone and sign-off for the reply Rules - Use only the policy, rules and case details above. Do not invent order facts, dates, names or policy. - Do not promise a refund, replacement, exception, compensation or timeline unless the policy text allows it for this exact situation. If the policy is silent, or the customer asks for more than it allows, say that a colleague will review the case. Do not give a date unless the policy gives one. - Do not admit legal fault or argue with threats. Stay calm and respectful. - If no escalation rule applies, still write the reply and state that no handoff is needed. - Mark anything you had to assume with [ASSUMED]. Output in this order 1. Decision: Handoff needed or No handoff needed, and the escalation rule that applies, in a few words. 2. Reply to customer: under 150 words, plain language, using the customer's name. Acknowledge the specific problem, say what happens next and who will do it, and ask only for information the agent truly needs. 3. Handoff note for the agent (skip if no handoff is needed): Reason: Urgency: High, Medium or Low, with one line on why Customer wants: Already tried: Policy points that apply: What the customer has been told: Suggested next step for the agent: Open questions: Before finishing, check the reply against the policy line by line and remove any promise the policy does not support.
Open in
Open in Claude (opens in a new tab)Open in ChatGPT (opens in a new tab)

Gemini cannot be filled from a link, so the prompt is copied first. Paste it with Ctrl+V or Cmd+V.

Use it as a skill

Add this playbook to your assistant once, then ask for it by name.

Download skill for ClaudeView SKILL.md (opens in a new tab)
How to install
  1. Claude: download the zip and upload it as is under Settings, Capabilities, Skills.
  2. Other assistants: open SKILL.md, copy it, and paste it into the assistant's custom instructions or project.
  3. Ask for the playbook by its task, and give it your details when it asks.

The skill is plain text with no scripts, links or commands. Read SKILL.md before you install it, and use a company-approved AI workspace. Need help setting it up? Book a call.

See a real result

Sample input: Kaveri Home Appliances, Pune · Hi, I ordered a mixer grinder (order KHA-20418) and the… · Customer: Rohan Mehta. Order KHA-20418, Rs…

Sample output from Claude Sonnet 5.5, run 15 Nov 2026

1. Decision: Handoff needed. Three escalation rules apply:

  • The customer mentions a consumer forum (legal mention), so this goes to the support manager.
  • The report is outside the 7-day window, so this goes to the returns team lead.
  • The order value is above Rs 5,000 and a refund is requested.

2. Reply to customer:

Subject: Your cracked mixer grinder jar – order KHA-20418

Dear Rohan,

I'm sorry your mixer grinder jar arrived cracked. Thank you for letting us know.

Our damage-on-delivery policy asks for reports within 7 days of delivery, and your order was delivered on 6 November. Because of this, I have passed your case to our support manager and returns team lead, who will review it and decide the next steps. I can't confirm a refund, replacement or dispatch date myself, as those decisions sit with them.

To help them, please reply with a clear photo of the cracked jar.

Thank you for your patience.

Warm regards, Kaveri Support Team

3. Handoff note for the agent:

Reason: Three escalation rules apply:

  • Consumer forum mentioned (legal mention, goes to support manager).
  • Report made 9 days after delivery, outside the 7-day window (goes to returns team lead).
  • Order value Rs 6,499 is above Rs 5,000 and a refund is requested.

Urgency: High. The customer has mentioned a consumer forum and wants action by "tomorrow".

Customer wants: Full refund and a new jar sent by tomorrow. He says he will go to the consumer forum if it isn't sorted.

Already tried: Order and delivery record checked (KHA-20418, Rs 6,499, UPI, delivered 6 Nov 2026). This is the first contact on this issue and no earlier reply was sent. No photo has been received.

Policy points that apply:

  • Damage must be reported within 7 days of delivery with a photo. The window ended on 13 Nov, and today is 15 Nov.
  • Reports after 7 days need manager approval.
  • After inspection, the returns team replaces the item or refunds to the original payment method.
  • Refunds take 5 to 7 working days after the returned item reaches us.
  • Only the returns team confirms replacement dispatch dates.

What the customer has been told:

  • He has been acknowledged and told the report is past the 7-day window.
  • The support manager and returns team lead will review the case.
  • He has been asked for a photo of the cracked jar.
  • No refund, replacement, exception, date or timeline has been promised, and the forum threat was not addressed or argued with.

Suggested next step for the agent:

  1. Support manager to own the case because of the forum mention. Returns team lead to decide on the out-of-window request.
  2. Get the manager approval that policy requires for late reports before anything is committed.
  3. Review the photo once it arrives.
  4. Reply to the customer with the decision. Only the returns team should confirm any replacement dispatch date.

Open questions:

  • Will the manager approve an exception, given the customer says he was travelling? There is no proof of this and the policy has no travel exception.
  • The policy says "manager approval" and the escalation rules say "returns team lead". [ASSUMED] Both the support manager and the returns team lead should be involved. Please confirm who gives the final approval.
  • The customer asked for both a full refund and a new jar. The policy offers replacement or refund, not both, so which should be offered if approved?
  • Will the customer send a photo? Without one, the policy requirement is not met.
  • Does the cracked jar need to be returned for inspection? The policy implies this but does not say how.

Was this playbook helpful?

Want this set up for your team?

I can fit this playbook into your CRM, helpdesk or inbox and train your team to use it.

Book a 30-minute review

Review before you send

Check the result

Watch out for

  • Over-promising. the AI may try to calm the customer by offering a refund or a date. Check every commitment in the reply against the policy and delete any the policy does not allow.

  • Several rules at once. a legal mention, a late report and a high order value can all apply to one case, each with a different owner. The AI tends to list them together. A person should decide who leads so the case does not sit between two teams.

  • Quiet assumptions. the AI may turn 'tomorrow' into a date or read a limit into the policy that is not written there. These look like facts in the note, so check them against the source.

  • Personal data. customer messages can hold phone numbers, addresses and payment details. Remove what the AI does not need, use an approved workspace, and follow your obligations under India's DPDP Act or GDPR. This is general guidance and not legal advice.

  • Wrong urgency. the AI judges from the text you give it. If a safety issue or legal threat is missing from your escalation rules, the AI will not flag it. Keep the rules complete and current.

  • Rubber-stamping. if agents send replies without reading them, errors reach customers. Ask reviewers to check four things each time: the decision, the lead owner, the promises and the facts in the note.

  • Outdated policy. the AI follows the policy you paste. Paste the current version each time, not an old copy from a chat history.

Tips for your industry

This playbook works for any business. These notes help you adapt it.

Damaged items, late deliveries and refund requests are the usual handoff triggers. Order value, days since delivery and payment method are the facts agents most often need in the note.

Questions people ask

It suggests, and your escalation rules guide it. A person should confirm the decision before the reply goes out, especially when the AI says no handoff is needed. Cases involving safety, legal threats or large amounts should always be read by a human.

Last checked · Updated