Inbox Zero with
an Agent —
26h a Week Back.

When founders say they have an email problem, they actually mean three different things. Each one costs you hours. And depending on which one you mean, the solution is different.

Problem A: too many emails

Anyone getting 300 emails a day has a volume problem. Most of those emails shouldn't be landing in your inbox at all — newsletters, automated notifications, CC-only threads, internal system updates. That junk alone eats 8–10 hours a week. An agent doesn't reduce this category — it brings it to zero.

Problem B: too many decisions

The harder problem is the decision problem. Every email implicitly asks you a question: how important is this? Do I need to reply now? Who else needs to know? Can this wait? These micro-decisions cost energy — and after 50 of them, your cognitive capacity for the day has visibly dropped. Psychologists call this decision fatigue. In other words: you don't just lose the hours in the inbox — you lose the hours afterward too, because your head is empty.

An inbox agent makes exactly these decisions — faster, more consistently, without fatigue. It sorts, prioritizes, labels. You no longer open an inbox in the morning — you open a short, pre-built task list. And you stay in control: what goes out is your call.

Problem C: reacting too slowly

The most business-critical problem is response time. Anyone who replies to a customer inquiry after 4 hours has a 60% lower conversion rate than someone who replies within an hour. That's not theory — that's HubSpot data from 2024. An agent that responds to incoming emails immediately, at least with an acknowledgment, solves this completely — without costing you a second.

Before you picture this wrong: an inbox agent is not a fully automatic autopilot that answers all emails itself. That would be the wrong goal — and it would be dangerous. You stay in control. What a good inbox agent takes off your plate:

What the agent does

• Classifies every incoming email into predefined categories (details in the next chapter)
• Labels and sorts — every email gets a label and lands in the right folder
• Deletes or archives automatically whatever is clearly irrelevant (newsletters, ads, automated notifications)
• Creates reply drafts for standard requests — which you only review and send
• Sends acknowledgments for customer inquiries — instantly, automatically, personalized
• Summarizes long email threads — so you know what's going on in 30 seconds
• Escalates urgent emails to you via Slack or SMS immediately
• Maintains the CRM automatically — when a customer emails, the contact gets updated

What the agent doesn't do

• Independently answer emails on sensitive topics (complaints, legal inquiries, price negotiations)
• Make decisions that need your personal judgment
• Send anything under your name without your approval

The goal is not: "The agent answers everything." The goal is: "I open my inbox in the morning and only find what genuinely needs my time. Everything else is done or prepared — and I have 20+ hours a week back." If it doesn't work for you, we'll tell you.

The inbox agent isn't an assistant that thinks for you. It's a filter that makes sure you only think when it's worth it — and the rest of your time is yours.

The heart of every inbox agent is its classification system. The agent scores every incoming email against these seven categories — and then automatically carries out the defined action. Every email it handles on its own is one that no longer eats your hours:

In a typical inbox with 150 incoming emails a day, our experience shows this rough split: ~5% Critical/Urgent, ~20% Standard, ~15% Info, ~25% Newsletter, ~30% Automated, ~5% Spam. In other words: 75–80% of incoming email doesn't need your time at all. The agent handles it completely — and gives you back the hours you spend sorting today.

Here's the complete system prompt of a production inbox agent — adapted from a real setup for a consulting firm with 8 employees. Names and specific details anonymized. You can use it 1:1 as a template:

You are the inbox assistant for [Company Name].
You process incoming emails and classify,
prioritize, and prepare replies.

ABOUT THE COMPANY:
- Consulting firm in digital transformation
- Clients: mid-sized companies, 50-500 employees, in DACH
- Main contacts: CEOs, IT leads, project managers
- Key active projects: [list updated daily]

CLASSIFICATION:
Score every email against these categories:
- CRITICAL: customer escalation, legal letters, press
- URGENT: new inquiries, open quotes, today's deadlines
- STANDARD: existing customers, internal coordination, suppliers
- INFO: confirmations, updates, FYI emails
- NEWSLETTER: subscribed and unsolicited newsletters
- AUTOMATED: system emails, tracking, notifications
- SPAM: ads, cold emails, phishing

ACTIONS PER CATEGORY:
CRITICAL → instant_alert(channel=slack, urgency=high)
URGENT → create_draft() + label("🟠 Today")
STANDARD → create_draft() + label("🟡 This week")
INFO → label("ℹ Info") + archive_after(hours=24)
NEWSLETTER → folder("📰 Digest") + summarize_weekly()
AUTOMATED → delete() or archive()
SPAM → delete() + block_sender()

WHEN WRITING DRAFTS:
- Write in the company's style: professional, direct, warm
- Formal with unknown contacts
- Informal with internal contacts and contacts who've agreed to it
- Maximum length: 120 words (excluding signature)
- Never start with "I hope you're doing well" or similar filler
- Always use the sender's name when known
- For customer inquiries: confirm the content + state the next step
- For meeting requests: check the calendar + suggest 2-3 alternative times

SIGNATURE (always append):
[Handler's name]
[Company Name] | [Phone]
[Email] | [Website]

ESCALATION:
If category is unclear → choose STANDARD
For emotional or conflict-laden emails → CRITICAL
For emails about money/contracts → URGENT, never reply directly
For emails in English → reply in English

OUTPUT FORMAT:
{
  "category": "URGENT",
  "priority": 2,
  "sender": "Name",
  "subject_summary": "One sentence",
  "recommended_action": "Describe the action",
  "draft": "Email text if relevant",
  "alert_text": "For Slack if CRITICAL"
}

This prompt is 400 words long. That's the right size — enough context for good decisions, not so much that the agent gets confused. Important: the prompt contains explicit escalation rules (unclear → STANDARD), explicit prohibitions (never reply directly on money/contracts), and a clear output format (JSON) that makes downstream processing easy. You stay in control, the agent does the hourly work.

Test the prompt before it goes live

Before you let the agent loose on your real inbox, test it with at least 30 historical emails. Take real emails from the last two weeks — a mix across all categories — and have the agent classify them. Compare its decisions to yours. Wherever it's off, adjust the prompt. That takes 2–3 iterations until the error rate is under 5%. If it doesn't work out in the end, we'll tell you — before you flip the switch.

This is the question I hear most often: "What does my morning actually look like then?" A concrete answer — based on the setup at a Hamburg marketing consultancy that's been running an inbox agent for four months and has gotten roughly 22 hours a week back:

This isn't an idealized scenario — it's the actual routine the owner of this consultancy described to us after 60 days. Before: 90 minutes on the inbox in the morning, plus another scattered 40–50 minutes through the day. After: 20 minutes, done, checked off. Roughly 22 hours a week back — life hours, not an "efficiency gain."

What happens with the digest folder

Newsletters and info emails don't enter your awareness at all — they land in a digest folder. Once a week, Friday at 4pm, the agent sends a summary: "12 newsletters this week. Highlights: [3 sentences on the key content]. Relevant articles: [3 links]." You get the essentials without having read a single one of those emails yourself — and get another 2–3 hours a week back.

What you need

• Email account with API access: Gmail (Google Workspace) or Outlook (Microsoft 365). Both have stable APIs that are easy to connect. Personal Gmail accounts work too, but Google Workspace is more stable for continuous operation.
• LLM provider: Anthropic Claude for classification and draft generation. Cost: roughly €0.002–0.005 per processed email. EU-hosted on request.
• Workflow engine: n8n (open source, self-hosted) or Make.com. n8n is simpler for this use case because it has native Gmail and Outlook nodes. Can be set up GDPR-compliant.
• Notification channel: Slack or Telegram for real-time alerts.
• Optional CRM: HubSpot, Pipedrive, or a Notion database for automatic contact maintenance.

The technical workflow

In n8n, the workflow looks like this:

1. Trigger: Gmail webhook that fires on every incoming email
2. Preprocessing: extract email content, convert HTML to text, identify attachments
3. Classification: LLM call with the system prompt + email content → JSON output with category and recommendation
4. Routing: switch node that triggers the right action based on the category
5. Actions: set label, archive, create draft, send Slack alert, update CRM
6. Logging: log every processed email to a Notion database or Google Sheet for review and optimization

Important technical details

Rate limiting: set a maximum of 200 API calls per hour. If an unusually large volume of email comes in (e.g. a newsletter bounce), you don't want to pay for 1,000 unnecessary calls.

Error handling: if the agent can't classify an email (e.g. because the LLM service is briefly unreachable), the email isn't deleted — it's marked STANDARD and stays in your inbox. Fail safe, not fail open. You stay in control.

Attachments: in the base version, attachments aren't analyzed — only the email text. If you also need attachment analysis (e.g. for contract PDFs), that's a separate workflow with a document agent.

Costs

At 150 emails a day, 5 working days a week: roughly 750 emails/week × €0.004 = about €3 a week in LLM costs. Plus n8n (about €20/month) or Make.com (€9/month). Total cost: €25–35 a month. For that, you get 20+ hours a week back.

We've built this setup for a lot of clients by now. Here are the six mistakes that keep happening — and that cost you hours instead of getting them back:

1. Letting the agent loose on the main inbox right away.

   Always test first with a separate test account. Forward your emails in parallel to the test account for three days and watch what the agent does — without real consequences. Only once the error rate is under 5% does it move to the real inbox.

2. Too-aggressive deletion rules.

   An agent that's too confident about deleting will eventually delete something important. Start conservative: when in doubt, archive instead of delete. You can increase aggressiveness after two weeks once you know its behavior. You stay in control.

3. No audit logs.

   If you don't know what the agent did in the last 24 hours, you can't find mistakes. Log every processed email with category and action into a table. It costs nothing and is priceless when something goes wrong.

4. Sending drafts directly instead of reviewing them.

   The agent creates drafts — not emails it sends itself. At least for the first 60 days. After that, you can turn on automatic sending for standard categories (e.g. appointment confirmations, acknowledgments) — but step by step.

5. Never adjusting the prompt.

   Your inbox changes. New customers, new topics, new patterns. Look at the prompt once a month and adjust it to what you've observed. A prompt from month 1 usually isn't optimal by month 6 anymore — and costs you hours you could be getting back.

6. No feedback loop.

   If the agent misclassifies an email, record it somewhere. Not by clicking (nobody does that), but with a simple system: an _agent_errors folder you drag wrong emails into. Once a week you check it and adjust the prompt accordingly.

One of our clients — the owner of a mid-sized ad agency in Frankfurt (anonymized as "MediaWorks") — had the classic inbox problem in its worst form. 200–250 emails daily, spread across three different accounts mixing inquiries, project emails, suppliers, and personal communication. Roughly 25 hours a week in the inbox — over 1,300 hours a year of his life.

Before the setup

• 3 email accounts, all mixed in one client
• No folders, no labeling system
• Average response time to customer inquiries: 6.5 hours
• Estimated daily time in email: 3.5 hours — roughly 25h/week
• Personal stress level from email (1–10): 8

What we built

Setup in 14 days. We first split the three accounts by function: customer communication, internal operations, personal. Then we built an inbox agent for the customer account with seven categories and automatic CRM maintenance in HubSpot. For the operations account, a lighter agent that only sorts and archives, without creating drafts.

Results after 90 days

The response-time improvement was the biggest surprise. Not because the owner replied faster — but because the agent sent automatic acknowledgments instantly. Customers felt heard, waited more calmly, and the conversation stayed calmer once the real reply came. Bottom line: roughly 22 hours a week back — nearly 1,150 hours a year of life.

I thought email was my job. Now I realize: email was an obstacle to my job — and it stole my life. Those are two different things.

The inbox agent in its base version is a good start. But it's not the end. Here are three extensions we typically add after the first stable month — each one gets you more hours back:

Extension 1: Smart follow-up

If you sent an email and got no reply after 48 hours, the agent reminds you automatically. Not with a generic reminder — with a prepared follow-up draft that ties back into the context of the original email. "Following up on my email from [date] about [topic] — did you get a chance to look it over?"

Extension 2: Sentiment analysis

The agent evaluates not just what an email is about — but how it feels. An email from a customer that's factually harmless but sounds emotionally tense is treated differently than a friendly inquiry. Tense emails always get a manual review flag — even when the content would otherwise be standard. You stay in control of anything emotional.

Extension 3: Meeting intelligence

When an email discusses an upcoming meeting, the agent automatically extracts the relevant details — date, time, attendees, topic — and creates a calendar entry. At the same time, it prepares a short briefing: what was the last contact with this person? What open items are there? What should be discussed?

These three extensions turn the inbox agent from a filter into a genuine digital assistant. Each takes 2–4 days to set up, and gets you another 3–5 hours a week back. If it doesn't work out for you, we'll tell you.

Inbox Zero isn't a myth. It's a result — one that emerges once you stop managing your inbox yourself and start training an agent to do it. Setup in 14 days. 26 hours a week back. 56 days of life a year.

The hardest part isn't the technical side. The technical side is solvable. The hardest part is the willingness to hand over control. The fear of missing an important email. The distrust of a system making decisions for you.

That distrust is justified — and that's exactly why we always start with a parallel test. The agent classifies, you check. After two weeks you know the system. After four weeks you trust it. After eight weeks you'll wonder how you ever worked without it — and what to do with the 22+ hours a week you now have back. If it doesn't work out for you, we'll tell you. You stay in control.

If you want to think through more concretely what an inbox agent would look like for your situation — with your accounts, your industry, your requirements — a 30-minute strategy call is the most sensible next step. Setup in 14 days · GDPR-compliant · EU-hosted · Free · 30 minutes · No obligation.