Mozart — Workflow Overview Docs

Plain-language, shareable explanations of what we built for Foundation, plus the reusable template.

Foundation · Live doc

Subscriber Miner — How It Works

In one sentence: Every morning, this automation looks at everyone who signed up for Foundation's newsletter in the last 24 hours, figures out which ones are real business opportunities, writes each a personalized intro, and lines them up in the right outreach campaign — ready for a human to review before anything sends.

When it runs: Automatically, once a day at 7:00am Central (12:00 UTC).

Right now it is in review mode: leads are being staged into draft campaigns only. Nothing is emailed to anyone until the team reviews and switches it on.

The steps, start to finish

1. Start the run

Kicks off and opens a logbook entry so we have a record of every run.

2. Pull the new subscribers

Grabs everyone who joined the newsletter in the last 24 hours, along with the details they gave at signup.

3. Filter out the personal emails

Drops anyone using a personal address (Gmail, Yahoo, etc.) because we can only pursue business contacts. Example: on a recent run, 44 signups came in and 31 were personal, leaving 13 business contacts.

4. Group the rest by company

Bundles the remaining people by their company (using their email domain), so each business is looked at as a whole rather than one person at a time.

5. Research and sort each company (the smart part)

For every company, the system researches who they are and sorts them into one of a few buckets:

6. Check we don't already know them

Cross-checks each contact against Foundation's existing systems so nobody who's already a lead or customer gets a duplicate cold intro.

7. Write a personalized intro and line them up

For the real opportunities, the system writes a tailored opening line based on what the company is up to, then places the contact in the correct campaign for the right rep (e.g. David or Julia). Example intro it generated: "Caught the Qflex launch, smart move giving automotive drying that energy-flexible edge."

8. Save progress and send a summary

Records what happened, remembers where it left off so tomorrow's run doesn't repeat anyone, and sends a short recap to the team.


What a daily recap looks like

44 subscribers → 12 business contacts (9 companies) 2 ideal customers, 0 partners, 7 set aside 5 people lined up in draft campaigns

The tools involved, in plain terms

ToolWhat it's for
CloudflareThe engine everything runs on, plus the memory/logbook
KitWhere newsletter signups come from
AI (OpenRouter)Does the sorting and writes the personalized intros
InstantlyWhere the outreach campaigns live (currently drafts)
HubSpotFoundation's CRM, used to avoid duplicates
TelegramWhere the daily recap is sent

Everything runs inside Foundation's own Cloudflare account. No data leaves Foundation's systems except the tools listed above, which are already part of Foundation's stack.

Foundation · Live doc

Visitor Miner — How It Works

In one sentence: The moment a real person browses Foundation's website, this automation identifies who they are, researches their company, checks we don't already know them, writes them a personalized intro that references what they were looking at, and lines them up in the right outreach campaign — ready for a human to review before anything sends.

When it runs: All day, continuously. A visitor is identified the instant they land, and the automation processes new arrivals in small batches every 30 minutes.

Right now it is in review mode: qualified leads are staged into draft campaigns only. Nothing is emailed to anyone until the team reviews and switches it on.

The steps, start to finish

1. Catch the visitor

Foundation's visitor-ID pixel (RB2B) spots a real, named person on any page of the site and instantly sends their details over. The automation records it and gets out of the way — no waiting.

2. Hold and batch

New arrivals wait in a short queue. Every 30 minutes the automation picks up the latest batch. This keeps costs down and lets it treat someone who viewed five pages as one person, not five.

3. Skip anyone already in the flow

Checks its memory. Anyone who's already been added to one of our campaigns is skipped for good, so no one is ever pitched twice, even if they come back months later.

4. Drop the non-business visitors

Anyone on a personal email (Gmail, Yahoo, etc.) or with no company is set aside — we only pursue business contacts.

5. Fill in the gaps and check the email

If the pixel didn't hand over a work email (or company size/revenue), the system looks the person up to fill those in, finds the most likely email, and confirms it's a real, deliverable address.

6. Check we don't already know them

Cross-checks each person against Foundation's existing systems so nobody who's already a lead or customer gets a duplicate cold intro.

7. Research and sort each company (the smart part)

For every visitor's company, the system researches who they are (using the company size and revenue as a read on whether they fit Foundation's budget sweet spot) and sorts them into a bucket:

8. Write a personalized intro that uses what they browsed

For the real opportunities, the system writes a tailored opening line, and because it knows which pages they looked at (pricing, a case study, the AI-visibility service), the intro can speak to exactly what they came for. Then it routes them to the right campaign:

9. Save progress and send a summary

Records what happened, remembers who it has handled, and sends a short recap to the team.


What a recap looks like

🟡 DRAFT (review mode, nothing sent) 18 identified visitors → 4 ICP, 1 partner, 9 set aside 4 already known, 5 staged into campaigns

The tools involved, in plain terms

ToolWhat it's for
RB2BIdentifies the real people browsing the website
CloudflareThe engine everything runs on, plus the memory/logbook
ApolloLooks up the person to fill in email, company size, and revenue
HunterFinds a work email when Apollo and the pixel don't supply one
Email List VerifyConfirms an email is real and deliverable
InstantlyWhere the outreach campaigns live (currently drafts)
HubSpotFoundation's CRM, used to avoid duplicates
AI (Claude via OpenRouter)Sorts each company and writes the personalized intro
Exa + FirecrawlLive research on each company
TelegramWhere the recap is sent

Everything runs inside Foundation's own Cloudflare account. No data leaves Foundation's systems except the tools listed above, which are already part of Foundation's stack.

How it's different from Subscriber Miner

Subscriber Miner wakes up once a day and works a list of newsletter signups. Visitor Miner is always on — it reacts the instant someone lands on the site, and its personalization uses the pages they actually browsed as the intent signal.

Reusable · Internal template

Overview Template (the blank we fill in for each workflow)

This is the reusable skeleton. Every new workflow doc starts here, then the placeholders in {{ }} get filled in. Kept plain so a non-technical person on the customer's team understands exactly what we built.


In one sentence: {{ONE_SENTENCE_WHAT_IT_DOES}}

When it runs: {{SCHEDULE_IN_PLAIN_TERMS}}

Current status: {{LIVE / REVIEW MODE / DRAFT}} — say plainly whether anything is sending or acting on real people yet.

The steps, start to finish

1. {{STEP_TITLE}}

{{WHAT_HAPPENS_IN_PLAIN_WORDS}}

2. {{STEP_TITLE}}

{{WHAT_HAPPENS}}

Add one block per meaningful step. Fold pure housekeeping (logging, saving state) into a single simple step so the list stays readable.

What a run looks like

{{EXAMPLE_OUTPUT_OR_RECAP}}

The tools involved, in plain terms

ToolWhat it's for
{{TOOL}}{{PLAIN_PURPOSE}}

Everything runs inside {{CUSTOMER}}'s own {{ACCOUNT}}. No data leaves their systems except the tools listed above, which are already part of their stack.


Writing rules