We built NodeCo's outreach engine the same way we build client systems: no vendor platforms, no enterprise CRM, no sales team. Six workflows. A business directory URL goes in one end, and personalized Gmail drafts come out the other. The whole cycle runs in under five minutes.

This is the system we use. If it works for an agency being built alongside a full-time job, it works for you.

The manual cost nobody talks about

Cold outreach eats time in predictable ways. You find a business directory. You open each profile. You visit their website. You decide which service fits. You write an email. You set a reminder to follow up. You check your inbox to see if they replied. You repeat this dozens of times.

For a solo founder or small agency, that's three to five hours a week spent on work that doesn't require your judgment until the very end. The bottleneck isn't writing the email. It's everything that happens before you write it.

Business directories give you company names and websites, but rarely email addresses or owner names. Evaluating a website to figure out what they actually need takes five to ten minutes per prospect. Writing a personalized email, even with a template, still requires enough customization to avoid sounding like a mail merge. Follow-up timing is easy to forget. Most outreach fails not because the first email was bad, but because there was no second one.

Reply detection means monitoring an inbox and manually updating a spreadsheet when someone responds. Miss that step and you send a follow-up to someone who already said yes.

What we built

We needed a system that handled the entire cycle: scrape the directory, enrich the data, evaluate the business, write the email, save it as a draft, schedule follow-ups, and detect replies. Six workflows, end to end.

Workflow one: directory scrape

You submit any business directory URL through a form. A web scraping tool extracts every member profile link from the directory page in one call. The system loops through each profile, scrapes the individual page, and pulls the real business website, phone number, email address, and contact name. Extracted data gets deduplicated and written to Airtable with a timestamp that triggers the next workflow.

Workflow two: AI service recommendation

A webhook triggers when the first workflow finishes. The web scraping tool pulls the prospect's actual website. That content goes to an AI model with a structured prompt that evaluates the business and recommends exactly one of three services: business process automation if they operate manually with no visible systems, website refresh if the site is outdated or poorly structured, or advertising if everything looks solid but there's no visible lead generation. The AI returns the recommended service and a two-to-three sentence explanation specific to that business. Both get written to Airtable.

Workflow three: email draft generation

Another webhook triggers. The AI model generates a personalized cold email using the company name, website, recommended service, and reasoning as context. The prompt enforces peer-level tone, a specific observation about the business, no generic openers, under 150 words, and a natural call to action. Subject line, body, and signature get parsed and stored in Airtable.

Workflow four: Gmail draft save

The email gets passed to Gmail via OAuth2 and saved as a draft in the hello@nodeco.ai folder. The Gmail thread ID is captured and stored in Airtable for reply detection later. No email is sent automatically. Every draft is reviewed and sent manually.

Workflow five: follow-up sequence scheduler

A daily trigger runs at 7:30 AM. The system queries Airtable for records where the follow-up send date matches today. For each match, the AI model generates a contextually aware follow-up using the original email as input and saves it as a Gmail draft. Follow-up one triggers three days after the initial email is sent. Follow-up two triggers three days after that. Both are saved as drafts, never sent automatically. The sequence stops after follow-up two.

Workflow six: reply detection

A trigger runs every fifteen to thirty minutes. For every active record, the system searches Gmail for replies to the original thread using the subject line as the search key. If a reply is detected from an address other than hello@nodeco.ai, the record updates to Replied and no further follow-ups are created.

What changed

A 14-business directory is scraped, enriched, and converted to personalized Gmail drafts in under five minutes. Follow-up emails generate and save automatically with no calendar reminders. Reply detection runs every fifteen to thirty minutes and stops sequences immediately. The system saves an estimated three to five hours per week compared to the equivalent manual process. It runs continuously at approximately thirty dollars per month in API costs.

Every email references something specific about the prospect's business. Service recommendations are grounded in actual website content. Email tone and structure are enforced by prompt rules across every record. Human review is maintained at every send point. Full audit trail in Airtable links every email, follow-up, and reply to its prospect record.

What this means for your business

Most outreach automation is built for funded sales teams with dedicated headcount and multi-thousand dollar platform budgets. For a solo founder or small agency watching every dollar, that model doesn't transfer.

This system demonstrates what becomes possible when you build for your actual constraints. It requires no CRM subscription. It adapts the service recommendation and email tone to each prospect based on actual website content. It scales with directory volume without additional headcount. It can be extended or modified in a visual workflow builder without developer involvement. It costs a fraction of traditional sales automation platforms.

We built this for ourselves first. If you're spending hours on outreach admin, that's the call to have. Thirty-minute discovery, no pitch. We'll walk through where your time goes and whether a system like this makes sense for your operation.

Book at nodeco.ai/contact.