The system that runs the day-to-day GTM motion is a Node.js application deployed on Render. It runs 24/7 as the autonomous intelligence layer beneath the entire sales motion. No third-party agent frameworks. Built from scratch with minimal, well-maintained dependencies.
The operating principle is constrained autonomy: the system proposes, the operator approves. Every external write action, whether a CRM update, a sequence enrollment, an outbound message, or a calendar invite, goes through an HMAC-signed Proceed/Cancel approval gate delivered through Google Chat. The system can do anything the operator could do, but never without an audit trail and a sign-off.
What the system does, in order of how it touches a typical deal:
Outbound enrollment
The system runs a 12-step multichannel sequence in Lemlist over 30 days, with LinkedIn touches interleaved with email, two voice notes, and a calendar link reserved for the break-up step. ICP filtering happens before enrichment spend. An industry qualifier check via Serper and Claude Haiku runs first; PDL (People Data Labs) enrichment runs only if the contact passes. A deliverability monitor auto-pauses any campaign on a bounce rate above 3% and alerts via Google Chat. 200 contacts are currently enrolled in the active track; 830 additional qualified contacts are staged for the next batch.
Intent scoring
Website-visitor identification feeds an intent scorer that tiers traffic into three tiers plus an enterprise flag. Tier 1 (pricing page, contact page, ecosystem overview, or two or more product pages in one session) triggers enrichment, ICP qualification, sequence enrollment, and an immediate Google Chat alert. Tier 2 routes to product-specific sequences. Tier 3 routes to nurture. A separate Sales Navigator signal monitor watches saved accounts for new-hire-in-role and promotion signals; either triggers enrichment and a contextual sequence enrollment.
Reply intelligence
Inbound replies to outbound sequences land pre-classified into five categories: Hot Demo Interest, Warm Soft Interest, Objection, Out of Office, Unsubscribe/Not Interested. Classification runs on Claude Haiku; drafting runs on Claude Sonnet. The Objection category routes through a playbook with response frameworks for the eight recurring objection types the team has seen. A wrong-person handler enriches the correct contact and drafts a warm intro to them, linked to the right CRM record. An OOO handler logs the return date and schedules a follow-up activity for return-plus-one-day.
Lead scoring
A composite 0 to 100 lead score runs daily at 6am ET across four dimensions: ICP Fit (40 points), Signal Strength (25 points), Engagement (20 points), Timing (15 points). Deterministic. Zero LLM cost. The score drives morning-briefing prioritization and triggers deal-coaching nudges.
Operating rhythm
A 9am ET daily briefing posts to Google Chat: new leads, hot replies, website visitors, stale deals (no activity in 14+ days), campaign health (sent, open, reply, click, bounce, interested rates against thresholds). Before the briefing sends, a Daily Action Plan job runs a full pipeline sweep, assembling per-deal context in parallel from the CRM, Gmail (90-day thread lookback), the outbound tool, the intent tool, and an observation snapshot log. One reasoning call per deal produces a prioritized, draft-ready action queue with Proceed/Skip gates. The operator walks into the day with the work already staged.
Pre-demo prep
90 to 105 minutes before any scheduled demo, the system pushes a pre-call brief to Google Chat with contact context, firm details, pain points pulled from prior interactions, and talking points based on the product pages the contact visited.
Post-demo workflow
Triggered automatically when a deal moves to Demo Completed in Pipedrive. The system pulls the demo transcript via the demo-recording tool's API, extracts pain points, product interest, objections, key account numbers, current tools, next steps, and competitor mentions, writes the extraction back to CRM notes, drafts a follow-up email in the operator's voice referencing specific call moments, and pre-populates a proposal template based on the demo's interest signals. Both drafts land in Google Chat for review.
Proposal generation
Four proposal templates are wired for one-click generation via Gamma. A Google Form submitted post-demo triggers an Apps Script POST to a webhook on the sales OS. Claude Sonnet drafts Executive Summary, Product Roadmap, Next Steps, and the final follow-up email in parallel. Gamma generates the proposal from the appropriate template. A Chat alert delivers the proposal URL and the final email draft. End-to-end: approximately one minute.
Inbound and intake
A Google Drive intake folder workflow (intake/, processing/, completed/, failed/) handles any CSV or Excel dropped in. Triage runs first (blacklist plus ICP filter, no API spend), an approval message goes to Chat with row count and estimated enrichment cost, and on approval the full enrichment pipeline runs. The pipeline is source-agnostic with column mapping in the UI, fuzzy auto-match, header-row detection across the first 20 rows, ICP filtering before enrichment spend, and an AI agent that extracts emails from scraped content where needed.
Event capture
A Chrome extension plus a mobile web capture page using Sonnet vision, writing to the CRM with a Google Sheets backup, makes conference and event motions reusable rather than ad hoc.