Email is the tax you pay for being reachable. Most founders pay it between 7am and midnight, seven days a week, with no sign of the bill going down.
The math is brutal: at 80 emails per day, even 2 minutes per email is 2.7 hours of daily inbox work. That's 13 hours a week. 676 hours a year. Roughly 17 full work weeks, gone — not to customers, not to product, not to the decisions that actually move the company.
There's a better way. Here's exactly how to set up AI inbox automation without turning your communications into the email equivalent of hold music.
Why Inbox Automation Fails (And How to Avoid It)
Most founders who try inbox automation either abandon it within two weeks or end up with a system that actively damages their relationships. The failures follow a predictable pattern:
- Too much autonomy too fast. They flip to full-auto on day one and immediately send a weirdly formal response to their best customer.
- No voice training. They give the AI zero context about their writing style and wonder why everything reads like a LinkedIn post from 2019.
- Wrong task boundaries. They automate things that should stay human (investor updates, sensitive negotiations) and keep doing manually the things that are genuinely automatable (vendor scheduling, newsletter responses).
The fix isn't a different tool. It's a better onboarding process.
The Right Mental Model: AI Drafts, You Decide
Before setting anything up, internalise this hierarchy:
- AI monitors — scans your inbox in real time, categorises everything
- AI drafts — writes responses for routine email types
- AI queues — puts drafts in a review queue with context and priority flags
- You decide — approve, edit, or decline; add nuance where needed
- AI sends — after approval, handles the mechanics
You're not removing yourself from the inbox. You're compressing 2 hours of processing into 20 minutes of decision-making.
Step 1: Define Your Email Taxonomy
Before loading anything into Vega, you need a taxonomy of the emails you receive. Not every email is the same kind of problem. Map yours into 4–5 buckets:
Routine / automatable (target: fully delegated)
- Vendor scheduling requests
- Newsletter subscriptions and event invites
- Basic customer support queries (with a support team in place)
- Cold outreach and sales pitches (auto-archive or polite decline)
- Internal team status updates
Hybrid (target: AI draft + your edit)
- Customer questions requiring real answers
- Partner and collaborator communications
- Press and media enquiries
- Job applications
Human-only (target: you, always)
- Investor communications
- Board updates
- Key customer escalations
- Anything where the relationship is the asset
Most founder inboxes are 60–70% Routine, 20–25% Hybrid, and 10–15% Human-only. You're automating the first two categories.
Step 2: Build Your Voice Document
Your voice document is what determines whether your AI emails sound like you or sound like a press release. Keep it short — 300–500 words beats 3,000 words.
Include:
- Three to five tone adjectives with examples from emails you've actually sent
- Words and phrases you never use (e.g., "circle back," "reach out," "per my last email")
- Words and phrases you always use (your shorthand, your jokes, your sign-off style)
- Formality calibration by recipient type (investors: more formal; teammates: casual; customers: warm but direct)
- Two or three sample emails you consider "perfectly on-voice"
This document becomes the first thing you load into your AI agent. The difference in output quality between founders who do this and founders who skip it is not subtle.
Step 3: Configure Triage Rules
Triage rules determine what happens to each category of email automatically:
IF: email from [known contacts list] AND subject contains scheduling
THEN: extract proposed times → cross-reference calendar → draft availability reply
IF: email from unknown sender AND body contains "partnership"
THEN: flag as "cold outreach" → add to review queue → low priority
IF: email from [investors list]
THEN: route to Human-only queue → notify immediately → do not draft
IF: email from [customer domain list]
THEN: categorise by topic → draft initial response → high priority queue
You don't need to write code. In Vega, these rules are configured through a natural language brief: "Always route emails from my investors list to me without drafting anything. For scheduling requests from contacts I know, draft a reply with three available times. For cold outreach, archive unless it mentions a company I've been researching."
Step 4: Run Supervised Mode for Two Weeks
For the first two weeks, don't let anything send automatically. Every draft goes into a review queue that you process in two 10-minute sessions per day (morning and afternoon).
During this phase, you're doing three things:
- Approving good drafts (which teaches the system what good looks like)
- Editing okay drafts (which teaches it where it's missing your voice)
- Declining bad drafts (which teaches it what categories it's not ready for)
Most founders see output quality jump significantly in the first week as the voice calibrates. By week two, 60–70% of drafts need zero edits.
Step 5: Graduate to Autonomous Execution
After two weeks of supervised mode, you have real data on where the AI is reliable. Now you can flip specific categories to autonomous:
- First to autonomous: Scheduling and logistics (lowest risk, most consistent)
- Second: Routine customer FAQ responses (if quality is consistent)
- Keep supervised: Anything involving relationship judgment, money, or complexity
The rule: automate the task type, not the email type. "Schedule meetings with vendors" is a task type. "Emails from Mark at Google" is not — Mark at Google might send you a scheduling request one day and a partnership proposal the next.
What Vega Handles End-to-End
Vega, Veqiro's AI executive assistant, handles all of this within a single system:
- Inbox monitoring — continuous scan, real-time categorisation
- Priority briefing — every morning, a 5-minute summary of what needs your attention and why
- Draft queue — responses ready for review with one-click approve/edit/decline
- Calendar integration — pulls your availability, blocks focus time, schedules proactively
- Follow-up tracking — flags emails that haven't been responded to after your usual response window
The result most founders report: a 90-minute daily inbox processing habit compressed into 15 minutes of review.
"I used to start every day in my inbox. Now I start with my briefing. It's the same information — but curated, prioritised, and ready to act on. Not just a firehose."
Inbox automation isn't about removing yourself from your communications. It's about removing yourself from the mechanical parts — so what's left when you show up is judgment, relationship, and intention.