AI email assistant trained on your inbox history and knowledge base. Grounded replies, not generic templates. Gmail and Outlook integration.
- Analyze
- Automate
- Monitor
monthly retainer
Who this is for
Founder or owner who spends 2+ hours per day on email. Inbox zero is a fantasy, canned responses embarrass you, and the mental cost of context-switching between threads eats the morning before real work starts.
The pain today
- 2+ hours daily on email that is mostly routine but still requires your attention
- Canned responses sound like a chatbot; customers notice and trust drops
- Every reply needs context from previous threads, policies, or pricing docs
- Email backlog lets good customer relationships quietly decay
- Hiring a VA costs $30-50/hour and still misses company-specific nuance
The outcome you get
- Drafts written in your voice, trained on emails you actually sent
- Grounded in your knowledge base: policies, FAQs, pricing, escalation rules
- One-click send-as-drafted or quick edit before send
- Sensitive threads stay local; no cloud training on private data
- Founders typically reclaim 45-60 minutes per day after 4-6 weeks of calibration
How the drafting pipeline works
When a new email arrives, the system runs three steps before you see it. First, triage: the email is categorized (routine inquiry, scheduling request, complaint, deal-related, escalation) and routed accordingly. Second, retrieval: relevant chunks from your knowledge base are pulled and attached to the drafting context. Third, drafting: the LLM writes a reply in your voice, grounded in the retrieved facts, matching the thread tone.
Triage is where most of the time savings come from. A founder's inbox is 60-70% routine. Scheduling confirmations, standard product questions, document requests, follow-ups that just need a date update. The AI handles the drafting for all of these; you spend 10 seconds reviewing and hitting send. The remaining 30-40% needs judgment, and that is where you stay in the loop.
Escalation routing is a first-class feature, not an afterthought. Threads tagged as legal, financial, complaint, or deal-sensitive bypass AI drafting entirely and land in a flagged queue for your direct attention. You define the rules. I build them in.
Drafts written in your voice, trained on emails you actually sent
Drafts, not auto-send — and why that matters
I default to drafts for every new customer. The reason is straightforward: a hallucinated fact or a tone mismatch in an auto-sent email is harder to recover from than a 10-second review step. Brand risk and relationship risk compound quietly.
That said, auto-send works well for a specific slice of email. Scheduling confirmations where your calendar is the source of truth. Standard document requests where the document is the only correct response. Acknowledgment emails on inbound leads where speed matters and content is fixed. For those categories, I can configure auto-send once the draft quality is confirmed over a 2-week sample.
Most customers find that after 4-6 weeks of calibration, around 90% of drafts are ready-to-send with no edits. The review step costs maybe 10 seconds per thread. The time savings are substantial even with that approval gate in place.
+500%: Lead base growth.
Knowledge base and voice training
Drafts sound like you and say the right things because the system is trained on two layers.
Voice layer: the LLM is prompted with 50-200 examples of your past sent emails. It learns your sentence length, formality level, how you open threads, how you close them, and how you handle pushback. You correct drafts that miss; those corrections feed back into the prompting.
Content layer: a retrieval system (RAG) indexes your knowledge base and pulls the most relevant chunks per incoming email. Sources I commonly wire in: FAQs and policy documents, pricing sheets, product documentation, CRM notes on specific accounts, and past thread summaries for recurring contacts.
Both layers update weekly as new emails accumulate and as you add documents. Over time the system gets sharper, not stale. I built a similar AI knowledge pipeline for Norte Web Digital, where Claude AI powered the CRM's outreach logic across 9,000+ companies in their pipeline. The same retrieval architecture adapts well to email drafting contexts.
Gmail and Outlook integration
Gmail: Google Workspace API with OAuth. Drafts are created inside the thread, in the native Gmail compose window. A notification reaches your phone or desktop when a draft is ready. No custom email app.
Outlook and Microsoft 365: Microsoft Graph API, same drafting pattern. Drafts appear in the standard reply composer. Mobile works through the standard Outlook app.
The integration adds no new interface to learn. The AI draft appears where you already expect to write. A lightweight browser extension handles inline review for customers who prefer editing in the web customer. The only thing that changes is that the blank compose window is no longer blank.
Setup and OAuth configuration take 1-2 days. Knowledge base wiring and voice calibration run in parallel during weeks 1-2 of the retainer.
Privacy and data handling
Email carries sensitive data and I treat it that way. A few decisions I make by default on every engagement.
Model choice: Anthropic Claude or OpenAI Enterprise, both with signed data processing agreements and explicit no-training-on-your-data commitments. Neither vendor uses your email content to train public models.
Excluded threads: flag any thread category (legal, HR, M&A, personal) and the AI never touches it. Exclusion is rule-based, not opt-in each time.
Knowledge base scope: the retrieval system only pulls from documents you explicitly designate. It does not crawl your Drive, Slack, or calendar without explicit permission.
PII handling: SSNs, payment card data, and health information are detected and redacted before any LLM call. For the highest-sensitivity customers, on-device draft generation is an option for specific thread categories, keeping data off cloud infrastructure entirely.
Pricing and break-even
AI email response automation fits the AI Automation retainer at $3,999/mo. First version timeline: 3-4 weeks to train voice, wire the integrations, and tune draft quality to a point where review is low-friction.
The retainer continues through ongoing refinement: voice calibration as your writing evolves, knowledge base additions, new category rules, edge-case handling. LLM API costs run $100-500/mo depending on email volume and model selection.
Break-even math: if you spend 2 hours daily on email and reclaim 60 minutes after calibration, that is 22 hours per month. At a conservative $150/hour loaded cost for a founder or senior executive, the reclaimed time is worth roughly $3,300/month. Most customers land above that after the calibration period, and many weight the mental clarity gains higher than the hourly arithmetic.
If your email volume is under 30 minutes per day, the tool probably does not justify the retainer. I will tell you that before I start.
Recent proof
A comparable engagement, delivered and documented.
A custom CRM that turned Google Maps into a lead machine
Built a purpose-built CRM for a digital agency that captures leads from Google Maps, reaches them via official WhatsApp (Twilio + Meta API), and uses AI to suggest replies and standardize templates. The system scaled the lead base by over 500%, with 250 new leads entering the pipeline every day.
Read the case studyKeep reading
Frequently asked questions
The questions prospects ask before they book.
After a 2-4 week calibration period, drafts match your voice closely. The system trains on 50-200 examples of your past sent emails: sentence length, tone, how you open and close threads. You correct the ones that miss; those corrections tighten the prompting. Customers consistently say they stop noticing the difference after 3-4 weeks of use.
Gmail via Google Workspace API and Outlook via Microsoft 365 Graph API, both supported natively. Drafts appear in your regular compose window. No custom email client required. Mobile works through your standard Gmail or Outlook app. Web clients get an optional lightweight extension for inline editing.
Integration and OAuth configuration take 1-2 days. Voice training and knowledge base wiring run in parallel during weeks 1-2. By week 3-4 the draft quality is usually good enough that review feels low-friction. Full calibration, where 90%+ of drafts are ready-to-send without edits, typically lands around the 5-6 week mark.
Drafts are drafts. You review before sending, so errors get caught before they reach anyone. When I see recurring errors of a specific type, I retune the prompts and knowledge base grounding to close that pattern. Over time accuracy improves because the system learns from your corrections and from new knowledge base additions.
Yes, and scheduling is one of the highest-return categories. Integration with Google Calendar or Outlook Calendar gives the system real-time availability. It drafts meeting proposals with specific times, handles reschedule requests, and confirms bookings. This one category alone often deflects a large share of daily email volume.
If you spend under 30 minutes per day on email, the retainer probably does not pay back on time savings alone. The sweet spot is 90+ minutes daily. At 2 hours per day, reclaiming even half of that delivers meaningful ROI at the $3,999/mo retainer rate. I run a quick volume estimate before any engagement to make sure the math works for your situation.
Yes. Each team member can have a separate drafting profile trained on their individual voice, while shared inboxes get a collective drafting profile. The knowledge base is shared across the team; individual voice settings stay distinct. Multiple reviewers can access drafts in a shared queue and send from the appropriate account.