Automation
The email your team keeps rewriting
Look at your sent folder. Half of it is the same six emails in different words. Refund explanations. Scheduling back-and-forth. Status updates. "Got your message, will be in touch." An email responder is not glamorous AI. It is the one that pays for itself fastest.
4 min read · published May 17, 2026
Why a template folder is not the answer
Templates fail because no two emails are actually the same. The customer is angry, or polite, or confused, or in a hurry. A template that opens "We apologize for the inconvenience" sent to a five-year customer reads as a slap.
An AI responder writes a fresh reply every time, in your voice, tuned to what the person actually wrote. The structure is consistent. The words are not.
What it reads before it writes
Three things. The incoming message. Your business context: what you sell, your refund policy, your scheduling rules. And the tone you have asked for: warm, direct, sincere, brief.
- Customer message, verbatim
- Business context: policies, hours, who you are
- Voice: the words you use, the ones you do not
Direct vs. drafts
Two modes. In draft mode the AI writes the reply and drops it in your inbox or Slack. A human reads it, edits if needed, hits send.
In direct mode the AI sends straight back when confidence is high. Typically scheduling, order status, simple policy answers. Anything emotional, anything involving money, anything tagged "escalation" gets routed to a human. You set the line.
What it costs to run
A typical reply costs about a tenth of a cent on Claude. A thousand replies a month runs you a dollar in model fees. The setup is the work. The running is the easy part.
- Per reply: about $0.001 in model fees
- Setup: $1,000-$1,500 one-time
- Monthly: $99-$299 depending on volume and channels
- Vs. a part-time inbox manager: roughly one fifteenth the cost
How we know it works
The responder on the demos page is the same engine. Paste a customer email. Watch the reply. We use it on info@dewgo.org for first-pass triage. The number of times we have had to override it in the last month: twice.