Every customer email triaged and drafted for approval
AI classifies every incoming message first, answers the routine ones from your real policies, drafts the rest for approval, and escalates anything sensitive immediately.
The problem
How it works by hand
A shared inbox at a small business is where response times go to die. Everyone can see it, so nobody owns it. The easy questions wait behind the hard ones, the hard ones wait for the one person who knows the answer, and the customer who emailed on Tuesday is chasing by Thursday.
A worked example
What a working version looks like
Classification comes before generation, which is what makes this safe to run on real customers. Every incoming message is first classified: routine question, order or booking query, complaint, legal or safeguarding issue, spam. Routine questions get answered automatically from your documented policies and prices, nothing improvised. The middle tier, messages that need a human decision but not human typing, get a drafted reply attached, grounded in the customer’s history, waiting for one-click approval or a quick edit. Complaints and anything sensitive skip drafting entirely and go straight to a named person with full context. Every category boundary is yours to set, and every automated answer cites the policy it came from.
The exact tools change per business. The shape does not.
What it needs
Honest inputs, nothing exotic
- 01Your shared inbox (Google Workspace or Microsoft 365)
- 02Your real policies, prices and FAQs in written form
- 03A named person for escalations
- 04Your call on where the line sits between auto-send and approve-first
The payoff
What you get back
Response times stop depending on who happens to look at the inbox. Routine questions get answered in minutes, the rest get drafted so a human only has to decide, and the sensitive messages reach the right person faster than they did when everything was manual.
Do it yourself
How you would build this yourself
No course, no upsell. This is the order we would build it in, with the tools named, and a prompt to start from.
- 1
Spend a week tagging the shared inbox by hand: routine question, order query, complaint, sensitive, spam. Those tags become your categories, and there are fewer than you think.
- 2
Write the real answers down: policies, prices, FAQs, in one document. The AI is only as good as what you give it to answer from.
- 3
Build classification first: a script that reads each new email and applies a label, nothing more. Run it silently alongside the humans for a week and check its calls.
- 4
Then add drafting: for routine and middle-tier messages, Claude drafts a reply grounded in your document and saves it on the thread. A human still clicks send.
- 5
Hard-wire the escalations: complaints and anything legal or safeguarding go straight to a named person, no draft, no delay.
Build me an email triage system for a Google Workspace shared inbox. Step one: classify every incoming email as [routine question, order query, complaint, sensitive, spam] and apply a Gmail label, nothing else, so I can check its accuracy for a week. Step two, once I approve: for routine and order emails, draft a reply grounded only in the attached policies and price list, citing which section it used, and save it as a draft on the thread. Complaints and sensitive emails forward to [name] immediately with no draft. Nothing ever auto-sends.
Copy it into Claude Code, fill the brackets, and it will plan the build with you before writing a line of code.
We would rather show you how than bill you. The whole ladder of free help, answers, guides and the weekly build-along, is on the do-it-yourself page.