Alle Artikel
AI & Automation5 min readEN

The Inquiry Inbox: Sort, Prepare, Miss Nothing

Inquiries, invoices, spam and supplier mail in one shared inbox. How automation sorts it and drafts replies without ever sending to customers on its own.

Armin Stadler
Armin Stadler

Monday morning, the shared info@ inbox. Somewhere in there: a customer asking about a new bathroom, two supplier invoices, a newsletter, a rescheduled appointment, a complaint, and three messages that look like spam but might not be.

Whoever looks after the inbox does the same work every day: read, classify, forward, ask back, reply. Reading is not the problem. The problem is that an inquiry gets stuck between two invoices and is only noticed when the customer chases it, or never.

This post is the counterpart to missed calls: that one was about the phone, this one is about the inbox. Same logic. Nothing should disappear quietly.

What can be automated

Sorting by type. Customer inquiry, invoice, appointment change, complaint, supplier, advertising. This is where language models genuinely help: they read unstructured text and assign it to a fixed list of categories. You define the list, not the model.

Flagging urgency. A burst pipe is more urgent than an inquiry for next spring. Clear signals can be written down as rules: words like "emergency", "leaking", "today", known customers with an active job. Anything ambiguous goes to a person.

Extracting the key data. Name, address, phone number, the service wanted, the time frame. That becomes a task where work already happens: in the CRM, the job system, a shared list. The email stays linked, so nobody has to ask where the original is.

Preparing a draft reply. For typical inquiries the system writes a draft: a confirmation, a request for missing details, an appointment suggestion from free slots. The draft is ready. A person sends it.

Passing invoices on. Incoming invoices are recognised and handed to invoice processing. What happens to them there is a separate topic and does not belong in the inquiry inbox.

Why nothing goes to customers automatically

The temptation is real: if the draft is already good, why not send it straight away? Because even a draft that usually fits will at some point contain a wrong commitment, the wrong tone or an answer to the wrong question. And nobody sees that case if nobody looks.

An automatic acknowledgement ("Your message has arrived, we will get back to you") is a different thing. It promises nothing the business cannot keep. Replies with substance stay drafts.

Escalation for complaints

Complaints and annoyed customers need their own rule. Not into the general list, but straight to a named person, with a flag. No friendly, generic draft for someone who is angry. What counts here is that a person responds quickly and personally.

The same goes for anything the system cannot classify with confidence. "Unclear" is a category of its own, and it goes to a person, not to spam.

What stays with people

Tone. How a business writes is part of its reputation. A draft can provide the frame; the final sentence comes from a person.

Decisions. Whether a job is accepted, whether an appointment works, whether a complaint is justified. Those are decisions, not sorting.

Anything unusual. The inquiry that fits no category is often the most interesting one.

Settle data protection first

Emails contain personal data, sometimes very personal data. When a service reads them, classifies them or extracts data from them, that content is being processed. Which providers are suitable, what needs to be agreed contractually and how long what is stored: settle that with your data protection officer before anything is built.

Frequently asked questions

Can't the spam filter do this?

The spam filter separates wanted from unwanted. It does not tell an inquiry from an invoice or a complaint, and it extracts no data. It stays the first stage anyway.

What happens when the system gets it wrong?

Then an email lands in the wrong list. That is why every category should be reviewed by a person regularly, especially at the start. Misclassified cases show where the rules need tightening.

Is it worth it with only a few emails a day?

That depends less on volume than on how much gets left behind. If inquiries are regularly overlooked, that is the pain that matters. Weigh effort against benefit honestly before you build.

Do I need a new email system for this?

Usually not. Common email services offer interfaces a workflow like this can connect to. Which one you have is the first technical question.

What you can do today

  1. For one week, tally which kinds of email land in the info@ inbox. A tick list is enough.
  2. Note every inquiry that was only handled after the customer chased it.
  3. Define the categories you want, including "unclear".
  4. Decide who receives complaints and how quickly they get a response.
  5. Run the case through the AI automation ROI calculator: hours per week, people involved, hourly rate, automation share, build and maintenance cost. The result is a capacity value, freed time priced at the hourly rate, not money saved in the bank.

How we build workflows like this without anything reaching customers unchecked is described under AI & automation. Or talk to us directly in a first call.

ai-automationsmetradesemaillead-handling

Ready to put AI to work in your business?

We help you find the right levers — pragmatic, measurable, and with real business impact.