• Oct 17, 2026
  • 4 min read

The Outbound Pipeline, End to End: Maps Listing to One Honest Email

The rules came first (what the law allows, and why outbound lives apart from inbound). This is the machine that respects them: from “businesses of type X in town Y” to a short list of companies, each with one problem I measured on its own website.

The flow

flowchart LR
    Q[Query: type + town] --> S[1. Scrape<br/>maps listings]
    S -->|raw.json| E[2. Enrich]
    W[Business's own website] --> E
    CH[Company register API] --> E
    E -->|leads.csv| B[3. Audit batch]
    B -->|one finding + PDF| C[send-ready CSV]
    C --> X[4. Sending tool<br/>separate domain]
    X -->|reply| R[Send the PDF report]
    E -.->|no email / not a company| D1[dropped]
    B -.->|site is fine| D2[dropped]

Each stage is a script that reads the previous stage’s file and writes its own into the same dated run folder (data/<date>-<type>-<town>/), so any stage can be re-run without paying for the ones before it.

1. Scrape: listings, nothing else

A hosted Google Maps actor (Apify) returns name, website, address, postcode, category. Everything else is switched off: no reviews, no photos, no reviewer data, no people. It’s cheaper and there’s less personal data to justify. A run of 30 listings costs cents.

Kept: one raw.json. Dropped: listings without a website.

2. Enrich: two questions, both answered at the source

  1. Is there a business email on its own website? The script fetches the homepage and a few contact paths with the standard library, reads mailto: links, plain text and Cloudflare-obfuscated addresses, and keeps only addresses on the business’s own domain, preferring role addresses (info@, hello@).
  2. Is it a company? The UK company register’s free API, matched strictly by postcode or identical name. Only active Ltd, LLP or PLC pass.

Kept: leads.csv with the trading name, the email, where the email was found, and the register evidence. Dropped: no email, sole traders, uncertain matches.

3. Audit batch: one finding, or no email

The same engine as my public website audit runs per site: Lighthouse on mobile plus security and SEO checks. A ranked list picks one concrete finding; a speed finding is measured twice and rounded down. Sites with no real problem are dropped. For the rest, the batch renders a one-page PDF report in the outbound brand.

Kept: a send-ready CSV (company, email, finding sentence, subject, register evidence) and one PDF per company. Nothing is sent from here.

4. Send: the only stage that talks to people

The CSV is imported into a sending tool on a separate domain, with its own warmed-up inboxes and sender name. Plain text, no tracking, one follow-up, an opt-out honoured permanently. The PDF goes out only to people who reply.

The numbers from a first run

StageCount
Listings scraped30
With a website26
Business email on own site16
Confirmed company7
With a real finding7

Thirty listings became seven emails. The narrowing is the point: every remaining row can answer why this business, why this address, why this claim.

Why there’s no crawler in it

I’d experimented with an AI crawling service (Firecrawl) on another project, and left it out here on purpose:

  • The pages are simple. Small business homepages and contact pages are plain HTML; the standard library fetches them in milliseconds.
  • The data is tiny. One email address and a company number, not page content to summarise.
  • Fewer processors. Every external service that touches the list is one more thing to name in the privacy notice and the data inventory.
  • Cost. It would be a third paid service for work a few lines of Python do.

If a later market has mostly JavaScript-rendered sites, a headless fetch is the first upgrade, not an LLM.

Where the data lives

  • The run folders are git-ignored; the code is in a private repository, the data never is.
  • API keys are in a local .env, never in the repo.
  • Runs are deleted after 90 days (make clean-old); the opt-out list is the one thing that’s kept.

Takeaways

  • One file per stage, in one dated folder: re-runnable and auditable.
  • Ask the source: emails from the business’s site, company status from the register.
  • Every stage drops more than it keeps, and that’s the design.
  • Only the last stage talks to people, from a separate domain.
  • No crawler until the sites need one.