AI Email Outreach System Built on Google Maps Business Discovery

The client sold website builds to local businesses and wanted the top of his funnel automated: find companies on Google Maps that either have no website or an obviously outdated one, gather their published contact details, have an AI write something personal to each of them, send it, and report back on what happened.

The system

We built a custom Python command-line system that does exactly that, end to end.

• Discovery — campaigns are defined in YAML files specifying city, business categories, minimum rating and per-day send limits. Businesses are found through the Google Places API (New).
• Contact collection — each candidate's own website is scraped for a published email address.
• Writing — Gemini writes a per-business email rather than a mail-merged template.
• Sending — SMTP delivery rotated across the client's three mailboxes, with daily limits, spacing between sends for domain warm-up, and unsubscribe handling.
• Reporting — CSV lead exports covering name, type, address, rating, review count, phone, email, Maps link and contacted status, plus an offline searchable HTML leads dashboard.

The command surface is deliberately small: --find, --preview, --send, --report and --dashboard.

Deliverability groundwork

An outreach tool is worthless if the mail lands in spam. We set up the client's Google Cloud and Gemini API access, and configured his Namecheap Private Email DNS — MX and SPF records, with DKIM and DMARC attempted — before any campaign ran.

Making it his, not ours

The client is not a developer, so the handover mattered as much as the build. We walked him through installing Python on Windows and running campaigns himself, and shipped four double-click .bat launchers (SETUP, FIND, SEND, SEE RESULTS) alongside written documentation: WALKTHROUGH, README, HANDOVER, COMMANDS and a CITY-PACK, plus a private handbook covering 58 cities and over 100 business categories.

Results

We ran live London campaigns from the system: 140 businesses collected, 70 email addresses, 136 phone numbers and 20 emails sent. The client then ran it himself for London, Manchester and Birmingham and pasted his own run output back into the chat. He released the milestone with a bonus and left a review calling the work professional and delivered on time.