The job
Your knowledge base is full. Onboarding guides. Feature walkthroughs. Billing explanations. Troubleshooting steps. A customer asks “How do I reset my password?” The answer is in your docs. Instead of a human finding the doc and rewriting it in your house voice, the station reads the doc and drafts the answer. Fifteen seconds instead of five minutes.
Plated well: the customer reads the answer and never notices it came from the station. It sounds like the team. It points to the specific doc if they need more. It connects the feature to what they’re trying to do, not just the steps.
The recipe
All seven ingredients still apply. The leverage on this dish is Context (Ingredient #2). The station is only as good as the knowledge base it’s pointed at and the account context it can read.
Context is load-bearing because the station needs your knowledge base plus the customer’s account history. Just retrieving the docs creates template-level answers. Adding account context means the station can say “based on your plan” or “for your setup” instead of generic steps. Training sets the house voice. Examples show the station how to wrap a knowledge-base answer so it sounds like the team. Output Over Process means the destination is “customer understands and can act” not “customer read all the docs.”
How to build it
-
Audit your knowledge base. Pull the 30 most-asked-about topics. Do those docs exist and are they current? If not, write them first. The station can’t retrieve what isn’t there.
-
Set house style for KB answers. How do you usually answer these questions. Do you open with empathy or jump to steps. Do you link back to the docs or embed the answer. Do you offer a next step or leave it hanging. Write one answer you’re proud of and show the station the standard.
-
Pipe in account context. The station should read the customer’s plan level, setup type, and usage history. That way it can say “your plan includes X” instead of generic language.
-
Create three answer examples. Take three real customer questions and three answers you’ve already written to those questions. Show the station the destination. Then have it draft answers to new questions and compare.
-
Set the review checkpoint. The station drafts. A human reads it before it goes out. You’re measuring quality and catching misses. This is not auto-send.
-
Track deflection rate. Did the KB answer actually resolve the issue or did the customer reply asking for clarification. If six of ten answers resolved without follow-up, it’s working.
What breaks it
-
Knowledge base garbage. The docs are outdated. The station points the customer at a guide that’s no longer accurate. Or the station can’t find the doc and defaults to generic language. Keep your KB current or the station’s output becomes liability.
-
No account context. The station drafts “here are the steps to upgrade your plan” without knowing the customer is on the highest plan already. Generic answers read like the station wasn’t paying attention.
-
Template voice. The station wraps the KB doc in canned language. “Thanks for reaching out. Please see the attached guide.” The customer reads it and feels like they hit a wall, not talked to someone.
-
No escalation path. The station drafts an answer to a question that actually needs a phone call. A technical setup question. A billing dispute. If the KB answer won’t work, the station needs to escalate, not send the wrong doc.
When it’s working
By week two, KB-answerable questions are being answered within 24 hours. By week four, customers are getting drafted answers that sound like the team. The support team saves 30-45 minutes per day on repetitive questions. Measure it: ask customers whether the KB answer they received actually solved their problem. If eight of ten say yes, it’s working.
Monday Move
Take three customer questions from this week that could have been answered by your knowledge base. Have the station draft answers to them. Read them against your KB docs. Do they match your house voice or read like a template.
Dish 2 of 10 on the Service Station. Build-note leverage: Context (Ingredient #2).