← The Service Station ~ dish 03 of 10 · the service station

Standard Response Generation

Your team has written hundreds of refund replies, reschedule confirmations, and status updates. You're not writing new language. You're showing the station what your language looks like. Then it drafts the next one.

The job

A customer requests a refund. Or a rescheduled call. Or an update on a ticket that’s been sitting. These are not judgment calls. They’re template moments. The problem is templates read like templates. Your team today writes the reply from scratch every time because starting with a template feels like you’re handing the customer a canned response. Takes five minutes per message. Times thirty a week, that’s two and a half hours of repetitive writing.

Standard response generation reads your archive of real replies and drafts the next one in your voice. Not a template. Your voice. The station has read what you actually write when you’re being efficient. It knows your refund voice (sympathetic but clear), your reschedule voice (matter-of-fact), your status-update voice (specific and grounded). Draft comes back in 60 seconds. Team reviews. Sends.

Plated well: the customer reads the refund reply and thinks “they understand why I’m frustrated and they’re fixing it clearly” not “this is a form letter.”

The recipe

All seven ingredients still apply. The leverage on this dish is Examples (Ingredient #4). The station’s output is only as good as the examples you show it. You can’t give it a framework and expect refund voice. You give it three real refunds you’ve written. Now it understands.

Training sets the house style at the granular level. This is not high-level tone. This is verb choice, sentence length, how you open an apology. Context matters because the customer’s account history changes the message (refund for a long-time customer reads different than refund for someone in their first month). Examples are load-bearing because they’re the only way the station learns what you actually do versus what you think you do. Many teams say “we’re sympathetic” and then their refund template opens with “We appreciate your feedback.” That’s not sympathetic. That’s corporate. Examples expose the gap. Output Over Process means the destination is clear: send a refund that lands as human, not robotic. Measurement means you’re watching whether customers feel heard.

How to build it

  1. Pull your archive. Extract twenty of your best standard responses. Refunds, reschedules, status updates, billing explanations. Get a mix. The station learns from the pattern, not a single flavor.

  2. Name your response categories. What types of standard messages do you actually send? Refunds (full, partial, store credit). Reschedules (moved by customer, moved by you). Status updates (update delivered, escalated, waiting on customer). Billing corrections. Billing explanations. Feature requests acknowledged. Pick the top five you send most.

  3. Tag your examples by category. For each of your twenty saved responses, mark which category it belongs to. The station needs to know “this is how we do refunds” versus “this is how we do reschedules.”

  4. Audit for voice patterns. Read your twenty examples. Do they sound like one person or five? If you’re hearing different tone, different opener, different vocab, the station will default to the most common pattern. Sharpen toward one house voice first.

  5. Set the review checkpoint. Station drafts. A human reads before send. This is not auto-reply. You’re measuring quality and catching misses. Flag which categories need tighter guardrails (refunds should never read apologetic on behalf of the product; status updates should always include next step).

  6. Track send velocity and reply rate. Did the standard response resolve the ask or did the customer reply asking for clarification. If eight of ten refund replies closed without follow-up, it’s working.

What breaks it

  • Generic examples. Your twenty saved responses are from a template library, not from your team’s actual writing. The station learns template voice, not your voice. Pull from real threads. Not the polished final reply. The actual thing that went out.

  • Mixed voices. The station learned from five different people’s writing styles and averaged them into something that sounds like nobody on your team. Assign one person as the “house voice anchor” for each response category. They write the three examples. The station learns from that one voice.

  • No category clarity. The station drafts a reschedule in refund voice or a status update that reads like an apology. You didn’t mark examples by type clearly enough. Add category labels. Make the grouping obvious.

  • Zero guardrails on what not to say. The station writes “we value your feedback” for a refund because that’s in the examples somewhere. Set one rule per category. Refunds never open with “we appreciate.” Status updates always include next step. Billing explanations never admit system error without escalation path. Guardrails close the gaps.

When it’s working

By week two, standard responses are drafted in sixty seconds. By week four, the team is sending drafted responses without major edits. The team saves three to four hours per week on refunds, reschedules, and status updates. Reply rate stays flat or drops (customers don’t need follow-up because the first reply was clear). Measure it: ask five customers if the response felt personal or templated. If four say personal, it’s working.

Monday Move

Pull five of your best refund replies from the last month. Read them alongside each other. What words do you actually use. What’s your opening move. What’s your closing move. Have the station read those five and draft a new refund. Compare. Does it sound like you or like a template that studied you.

Dish 3 of 10 on the Service Station. Build-note leverage: Examples (Ingredient #4).