Email automation with AI for companies

I build email systems for companies in which AI sorts the inbox, drafts replies for review and runs sequences that react to what each person did, on domains prepared to reach the inbox. It is for sales and operations teams in Mexico and the United States that depend on email and cannot afford to land in spam.

Email automation with AI is a system that prepares a domain to reach the inbox, sorts what arrives, drafts what goes out and measures what people did with each message. A company needs it when its sales or its requests arrive by email, when replies wait for hours because someone has to read and route them, or when its messages have started to land in spam.

What I build

  • Deliverability, first. A sending subdomain apart from the main domain, mailboxes warmed on a slowly rising schedule, and SPF, DKIM and DMARC checked and corrected.
  • Inbox triage. A classifier that reads each incoming message and sorts it: interested, not now, wrong person, a question.
  • Drafts with review. Replies and first messages written from the facts on record, which a person approves, edits or discards.
  • Sequences that react. Follow-ups that depend on what the person did, and that stop the moment someone answers.
  • Measurement. Deliveries, opens and replies per message, per segment and per sequence, read from real inboxes.
  • The handoff. A reply that matters reaches the right person, with its context, in the channel the team already uses.

What to check before automating anything

Automation multiplies whatever the domain already is. Four checks come first:

  • Does the domain publish SPF, DKIM and DMARC, and do they pass? Without them, a receiving server has no reason to trust the message.
  • Does outbound share a domain with invoices and support? If it does, one bad campaign can hurt the mail the business lives on.
  • Where did each address come from? One published by the company, or verified with its mail server, can be written to. One guessed from a naming pattern cannot.
  • Can the company say today how many messages were delivered and how many were answered? If not, nobody can tell an improvement from an accident.

A no to any of these is the first piece of work. Better copy on a domain with no reputation changes nothing, because the message is filtered before a person sees it.

How the work runs

  1. Audit the sending setup. I read the records of the domain and the headers of real messages to see who runs the mail, how it authenticates and how receivers treat it.
  2. Separate and warm. A subdomain with its own mailboxes warms for six to eight weeks. A reputation reading sets the daily limit of each mailbox: up when it holds, down when it dips.
  3. Define the classes and the rules. With the team, I write down the kinds of message that arrive and what should happen to each one, including which replies may go out without review.
  4. Write the sequences as experiments. A first message in plain text, short, with no links. What follows depends on behavior: a nudge for a message nobody opened, a different message for one that was opened and not answered, and nothing more for a person who replied.
  5. Measure and change. Targeting, volume and copy change with the numbers, and a change that does not move them is rolled back.

What you receive

The sending subdomain and its mailboxes, warmed and authenticated. The classes, the rules and the sequences in writing. A shared view where the team sees replies as they arrive, and the measurements behind each decision. Consent and opt-out rules sit in the path of every send.

What this does not include

I do not sell volume. I do not send to bought lists or to addresses inferred from a pattern, because that is how a warmed domain loses its reputation in a week. I do not promise a reply rate. And I cannot yet show, across enough companies, how replies turn into signed contracts: the sales team owns that last step, and those numbers are still coming in.

Where the evidence is

The method is public. email-intel is an open-source library, command line tool and MCP server that detects who runs the mail of a domain, grades its SPF, DKIM and DMARC posture from A+ to F and reads message headers. warming.email is the product I built so that a new domain earns the inbox before it uses it: a pool of domains that send real mail to each other on a rising schedule, with readiness and health reported. The case study on outbound engineering describes the three layers, and the essay of the same name walks through them in plain language. What I can show with confidence is deliveries, opens and replies, because the system measures them.

Common questions

Why do our emails go to spam?

Most often because of the sender, not the text. Gmail and Outlook decide where a message lands from the reputation of whoever sent it, and a new domain has none. Missing or failing SPF, DKIM and DMARC records make it worse, so I check those before changing a word of the message.

How long does it take to warm up a domain for email?

Six to eight weeks for a new set of mailboxes on a separate subdomain. During that time the system measures reputation and uses the reading to set how much each mailbox may send that day. It is a long wait, and it is the difference between reaching people and reaching filters.

Can AI answer our emails automatically?

It can sort every incoming message and prepare a reply. For a narrow set of cases it can also send one, such as sharing the information a person asked for or offering a link to book a meeting. Everything else goes to a person as a draft, and someone who has answered never receives a scripted follow-up.

Do you send cold email from our main domain?

No. Outbound runs from a separate subdomain with its own mailboxes. If something goes wrong there, the invoices and the support replies of the business keep arriving where they always did.

Do you work with companies outside Monterrey?

Yes. I work remotely from Monterrey with companies across Mexico and in the United States, in Spanish and in English. The sending subdomain, the mailboxes and the prospect data belong to your company, so nothing about the work needs me on site.

Related questions

Open to senior AI engineering roles and contract engagements, remote in the US, Mexico and Latin America.