My AI Agent Answered 47 Emails Overnight — Full Breakdown

Last updated: March 26, 2026

I run an e-commerce business selling home organization products. Customer emails come in all day — questions about sizing, materials, shipping times, return policies, and stock availability. I was spending 90 minutes every morning just clearing my inbox.

Two weeks ago, I connected my Tommy.app AI agent to my support email. The first night, it answered 47 emails while I slept. Here's exactly what happened, what the agent got right, what it struggled with, and whether I'd recommend this to other business owners.

How I set up email automation with my AI agent

The setup took about 10 minutes. I told my Tommy.app agent: "Connect to my support email and start answering common questions. Here's what customers ask about most: shipping times, return policy, product materials, sizing, and stock availability. Sound friendly but professional. Sign emails with my name."

The agent asked for my email credentials (I used an app-specific password for security). It connected to my inbox, read through the last 50 emails I'd sent to understand my tone, and confirmed it was ready.

I set one rule: draft responses for review instead of sending automatically. I wanted to check what the agent was writing before it went out to real customers. That rule stayed in place for the first three days.

What the agent answered on night one

I went to bed with 23 unread emails in my support inbox. By morning, I had 47 new customer emails and 47 draft responses waiting in my drafts folder.

Here's the breakdown by category:

• 15 emails — Shipping timeline questions ("When will my order arrive?")
• 12 emails — Product availability ("Is the large drawer organizer back in stock?")
• 9 emails — Return and refund questions ("Can I return this if it doesn't fit?")
• 6 emails — Sizing and material questions ("What's the shelf made of?")
• 5 emails — Order status updates ("Where's my package?")

Every response was accurate. The agent pulled shipping times from my website's FAQ, referenced my return policy correctly, checked stock status from the product pages, and answered material questions using product descriptions.

I reviewed all 47 drafts. I didn't change a single one. I hit "send" on all of them at 7:42 AM.

How did the agent learn my voice?

I didn't give the agent a style guide or tell it specific phrases to use. It learned my voice by reading the last 50 emails I'd sent to customers. The agent noticed patterns:

• I start most emails with "Thanks for reaching out!" or "Great question!"
• I use exclamation points but not excessively
• I include specific details (exact shipping days, not vague timelines)
• I sign off with "Let me know if you need anything else — happy to help!" followed by my first name

The agent mirrored all of this. When I read the drafts, they sounded like me. A few were slightly more formal than I'd normally write, but nothing felt robotic or off-brand.

One customer replied to an agent-written email thanking me for the "fast and helpful response." She had no idea an AI wrote it.

What the agent struggled with

There were two emails the agent flagged for me instead of drafting responses:

• A refund request from an angry customer — The agent recognized the tone was frustrated and flagged it with a note: "This customer seems upset. I recommend you handle this personally." Smart call.
• A custom order inquiry — Someone asked if I could create a custom size for a specific drawer dimension. The agent didn't have enough information to answer and flagged it for me.

Both were the right move. I don't want my AI agent negotiating with angry customers or making promises about custom work without my approval.

Out of 47 emails, 2 required my direct attention. That's a 96% automation rate on night one.

When I switched to auto-send mode

After three days of reviewing drafts with zero changes needed, I told my agent: "You can send responses automatically for standard questions — shipping, returns, product details, and stock. Flag anything about refunds, complaints, or custom requests for my review."

The agent started sending responses immediately. My inbox went from taking 90 minutes every morning to 10 minutes — just handling the flagged emails and skimming the sent folder to make sure nothing weird slipped through.

I've been running auto-send mode for 11 days now. The agent has answered 214 customer emails. I've had to correct exactly one — it gave a shipping estimate for international orders that was slightly off. I told the agent the correction, and it hasn't made that mistake again.

What customers don't know

I don't tell customers that an AI is answering their emails. The responses come from my support email address, they're signed with my name, and they're written in my voice. There's no "This email was generated by AI" footer.

From the customer's perspective, I just have really fast email response times now. Several have commented on it. One said, "I love how quickly you respond — I've emailed other stores and waited days." That email came in at 2:47 AM. My agent answered it at 2:49 AM.

Ethically, I'm comfortable with this. The information is accurate, the tone is mine, and complex issues still get routed to me. If a customer specifically asked "Are you a human or an AI?", I'd tell them the truth. But no one has asked.

How much time and money this saves

Before the AI agent, I spent about 90 minutes per day on customer support emails. That's 10.5 hours per week, or roughly 45 hours per month.

Now I spend about 10 minutes per day reviewing flagged emails. That's 1.2 hours per week, or about 5 hours per month.

I'm saving 40 hours per month. At my hourly rate of $75 (what I'd pay a VA), that's $3,000/month in time saved. Tommy.app costs me $49/month on the Starter plan.

Even if I valued my time at minimum wage, this would be a massive win. At my actual rate, it's absurd how cost-effective this is.

Would I recommend this to other business owners?

Yes — with conditions. This works best if:

• You get a lot of repetitive questions (shipping, pricing, availability, policies)
• You have clear, documented answers to those questions (website FAQ, product pages, policy docs)
• You're willing to review drafts for the first few days before enabling auto-send
• You set clear rules about what the agent can send vs. what it should flag for you

If your customer emails are mostly complex, emotional, or require nuanced judgment, this won't eliminate your workload. But it will still cut it significantly by handling the easy stuff automatically.

For e-commerce, SaaS support, coaching businesses, or any service with a high volume of FAQ-style emails, this is a no-brainer. You're already spending hours answering the same questions. Let the AI do it.