What is email reply automation?#
Email reply automation means using an AI tool to identify the type of incoming email, prepare a suitable draft reply, and automatically sort messages by urgency. Instead of thinking about what to write for every email, the AI prepares a draft based on your rules - you simply review and send.
This article is not theory. It is exactly what I did for my company over two hours one Tuesday morning. Here is the full process.
Why I decided to automate#
I run AI Solucija - an agency that sets up Claude Code for Slovenian companies. And ironically, I fell into the same trap I warn my clients about.
Every day I received between 15 and 25 emails. Most of them were predictable:
- "How much does your service cost?"
- "Can you help with automating X?"
- "What exactly does Claude Code do?"
- "Do you have a free slot for a meeting?"
- Confirmations, thank-yous, administrative messages
For every email I thought about what to write. I opened previous replies, copied chunks of text, adjusted the details, checked the tone. Each email took 5-10 minutes.
15 emails x 8 minutes on average = 2 hours a day. Just to reply to messages I had already answered 50 times before.
Two days a week were lost to typing the same replies in different words.
Something had to change. And since I sell exactly this kind of service myself, I already had the tool at hand.
Hour 1: Analysis and rules (the part that actually takes the most time)#
Step 1: I reviewed the last 50 emails (20 minutes)
I opened my inbox and went through the last 50 received messages. I was looking for patterns - which types of emails keep coming back?
I identified 6 categories:
- Service enquiry - someone asks what we offer and how much it costs
- Technical question - someone asks what Claude Code actually does or how it works
- Booking a slot - someone wants a meeting or a call
- Follow-up - someone replies to a previous conversation
- Administrative - confirmations, thank-yous, invoices
- Other - anything that does not fit into the categories above
Step 2: For each category I wrote rules (30 minutes)
For each of the 6 categories I wrote down:
- What tone to use (friendly, professional, without excessive formality)
- What the reply must contain (key information, price, next step)
- What it must never contain (promises I cannot keep)
- Examples of good replies (from my previous emails)
An example rule for service enquiries:
Step 3: I gathered 10 examples of good replies (10 minutes)
From my sent folder I copied 10 replies that I knew had landed well - clients reacted positively, the conversation kept moving. These examples showed Claude Code my tone, length and style.
Hour 2: Setup and testing#
Step 4: I set up Claude Code (20 minutes)
I gave Claude Code:
- All 6 categories with rules
- 10 examples of good replies
- Instructions for sorting by urgency (urgent, today, can wait)
- A rule: whenever anything is not 100% clear, flag it for manual review
The setup did not require any code. I wrote everything in plain Slovenian, as if I were explaining to a new employee how to reply to emails.
Step 5: I tested on 20 real emails (25 minutes)
I took 20 real emails received in the last two weeks and ran them through the system. For each one I checked:
- Did it correctly identify the category?
- Is the draft reply good?
- Would I send it as is, or does it need changes?
Results of the first test:
- 16 out of 20 correctly categorised (80%)
- 12 out of 20 drafts ready to send without major edits (60%)
- 4 drafts needed minor edits (tone, a missing detail)
- 4 were correctly flagged for manual review
Step 6: I refined the rules (15 minutes)
Based on the test I refined the rules for two email types where Claude Code missed the tone. I added another example for technical questions, since those were the most varied.
Second test: 18 out of 20 correct, 15 out of 20 ready to send. Good enough to start.
Results after 2 weeks#
After two weeks of use the numbers look like this:
| Metric | Before | After | Change |
|---|---|---|---|
| Time on email per day | 2 hours | 25 minutes | -1.5 hours/day |
| Average time per reply | 8 minutes | 2 minutes (review + send) | -75% |
| Emails sent without edits | 0% | 70% | Most go straight through |
| Emails for manual review | 100% | 15% | Only the unusual ones |
| Response time to clients | 4-8 hours | Under 1 hour | Faster response |
The biggest surprise: clients started replying back faster. When you get a reply within 30 minutes instead of half a day, the conversation moves forward. According to Harvard Business Review, companies that respond within the first hour of an enquiry are 7 times more likely to close the deal.
Two weeks later I refined the rules once more - I added a new category (partnership enquiries) and improved the tone for follow-up emails. Since then the system has run without major intervention.
What I learned#
1. Rules matter more than the technology
I spent 90% of the time thinking about the rules, not setting up the tool. What tone? What to include? What to leave out? These are decisions you have to make - Claude Code then simply executes them consistently.
2. 80% is good enough to start
At first I wanted every draft to be perfect. That is a mistake. If 70-80% of drafts are ready to send, you already save a huge amount of time. Perfection comes with time and adjustments.
3. "Manual review" is not a weakness
When Claude Code flags an email for manual review, that is not a failure. It means the system knows where its limits are. I would rather have a system that says "I do not know this one" than one that quietly sends a poor reply.
4. Time saved is only part of the story
Yes, I save 1.5 hours a day. But the bigger effect is that email no longer drains me. Before, after two hours of replying to messages I was mentally tired. Now I review drafts, send them off, and still have energy for the work that actually matters.
The full setup by the numbers#
- Setup time: 2 hours
- Tool cost: ~5 EUR/day (Claude Code API)
- Daily saving: 1.5 hours
- Monthly saving: ~30 hours
- ROI: The setup pays for itself in less than a day
For comparison: if I hired someone to reply to emails for 20 hours a month, it would cost 400-600 EUR. Claude Code does comparable work for 150 EUR a month.
"Will this work for my business?"#
If you receive more than 10 emails a day and most of them repeat - then yes, very likely.
This approach works for:
- Service businesses - enquiries about prices, slots, scope of services
- Online shops - questions about products, delivery, returns
- Consultants and agencies - booking meetings, follow-ups, proposals
- Tradespeople and contractors - enquiries about slots, prices, availability
It does not work for:
- Emails that require creative thinking for each individual case
- Negotiations where every reply is a strategic decision
- Communication where the personal relationship is a key part of the service
Frequently asked questions#
How long does it take to set up email automation with Claude Code?
For a basic setup like the one I describe in this article, you need about 2 hours. The first hour goes on analysing your emails and preparing the rules. The second hour goes on setting up Claude Code and testing. If your emails are more varied, it can take 3-4 hours.
Does Claude Code send the email itself?
No. In my setup, Claude Code prepares the draft and you review and send it. That is a deliberate choice - I want to have the final say. Full automation is of course possible, but I recommend reviewing the drafts to start with.
Does this work for emails in Slovenian?
Yes. Claude Code understands and writes fluently in Slovenian. All of my email replies are in Slovenian and the quality is on par with English.
What if the types of emails change?
You update the rules. Same as with a new employee - when a new type of question appears, you explain how to reply to it. With Claude Code that means adding a new category and an example. It takes 10 minutes.
Do I need technical knowledge to set this up?
No. Everything I described in this article I did in plain Slovenian. There was no code, no technical configuration. If you can write an email, you can set up this system.
Does Claude Code read my emails?
Claude Code only accesses what you allow it to. In my case I give it the contents of an email and it prepares a reply. It does not read your entire inbox without your knowledge. Anthropic, the company behind Claude Code, does not use your data for further model training.
Next steps#
If you want to set up something similar for your business, you have three options:
- Do it yourself - follow the steps in this article. 2 hours of your time, no cost beyond your Claude Code subscription.
- Free pilot project - each month we take on 3 companies for which we set up one automation free of charge. Email replies included.
- Free consultation - tell us how many emails you receive per day and what type. We will tell you whether automation is worth it.
In previous articles we explained what Claude Code is, gave 5 concrete use cases, and looked at how companies lose 16+ hours a week. Today I showed how I solved one of those problems in practice - step by step.
Next time: a comparison of Claude Code with tools like Zapier and Make. When to use which, for what, and whether a combination is worth it.
Tit Dolinsek is the founder of AI Solucija. He helps small and medium-sized businesses automate repetitive tasks so they have time left for the work that actually matters.