Richie Reach is a signal-led outbound engine that captures public LinkedIn engagement, filters it to your ICP, classifies it with AI, and runs the signal-to-pipeline engine for your outbound team, every week.

Stage 1 — Signal Monitor

The engine runs every 4 hours.

A cron-based monitor polls public LinkedIn engagement around client-defined keyword themes. Each captured event includes the profile URL, engagement type, post context, and timestamp. Deduplication runs on a LinkedIn URN hash. The same person appearing across multiple posts in the same week is merged to one record with engagement count incremented. The monitoring engine is built and maintained in-house — no third-party listener, no shared account pool, no platform dependency.

{
  "monitor_id": "mon_a4f2",
  "client_id": "clt_7x9k",
  "themes": [
    "AI infrastructure",
    "LLM cost benchmarking",
    "replacing offshore dev",
    "internal AI tooling",
    "vector database selection"
  ],
  "engagement_types": ["comment", "reaction", "share"],
  "recency_window_hours": 168,
  "dedup_method": "linkedin_urn_hash",
  "cron": "0 */4 * * *",
  "noise_filter": {
    "min_engagement_depth": 2,
    "exclude_job_posts": true,
    "exclude_self_promotion": true
  },
  "output": "icp_gate_queue"
}

Stage 2 — ICP Gate

Enrichment starts here. Not before.

Most teams enrich everything and filter later. That wastes enrichment credits and degrades deliverability. The gate runs before enrichment. Records that fail hard gates on title, geography, or company size are dropped immediately. Records that fail soft gates are scored lower and may fall below the delivery threshold later.

Records that fail hard gates are dropped immediately. Records that fail soft gates are scored lower and may fall below the delivery threshold.

Stage 3 — AI Classification

What did they actually say, and what does it mean?

OpenAI (via Clay) reads the engagement text and classifies behavioral intent. This is not sentiment scoring. It is a behavioral label that determines how claude sonnet 4.6 will write the outreach copy in Stage 4.

The classification also triggers which campaign type and copy framework claude sonnet 4.6 will use in the next stage.

Stage 4 — Copy Generation (claude sonnet 4.6)

Every email is written to the signal, not the job title.

Clay sends the scraped website content, the ICP description, and the engagement context to Claude Sonnet via the Anthropic API. The system prompt is around 2,500 tokens and contains the full decision logic for campaign selection, offer framing, and email structure.

14 writing rules are enforced in the system prompt: lowercase subject lines, 60-90 word body, soft CTA questions, no jargon, no exclamation marks, Grade 8 reading level, first sentence cannot start with "I".

Stage 5 — Composite Score

Records are scored before delivery. Not all signals are equal.

A comment from a VP of Engineering at a Series B SaaS company asking "has anyone benchmarked X" scores differently from a reaction from a Marketing Coordinator at a bootstrapped 5-person firm. The score is deterministic and auditable.

Stage 6 — Enrichment Waterfall

Three tools. One verified email. $0.87 per run.

Only records that pass the gate and score at or above 0.52 enter the waterfall. Prospeo runs first. If no match, Apollo. If no match, Hunter. If all three miss, the record is delivered with the LinkedIn URL only and email set to null.

83% cost reduction from filtering before enrichment.

Stage 7 — Outreach Infrastructure (ZapMail)

Mailboxes provisioned separately from the primary domain.

ZapMail provisions and manages the outreach email mailboxes. These are secondary domains, kept separate from the client's primary domain. ZapMail handles provisioning. Lemlist handles execution.

Stage 8 — Lemlist Multichannel

Email first. LinkedIn second. One coherent sequence.

Every qualified, scored, enriched record is pushed into Lemlist via Clay webhook. Lemlist handles the multichannel coordination. Email from ZapMail mailboxes, LinkedIn profile visits, connection requests, and direct messages. Reply detection pauses the sequence automatically at any step.

Stage 9 — HubSpot CRM

Every record is traceable back to its signal.

All active records sync to HubSpot with the full signal metadata attached. Attribution from a closed deal back to the signal theme, engagement class, and composite score at time of capture is available in the CRM. This is the feedback loop that improves the scoring model over time. Every closed-won deal traces back to the signal theme and engagement class that triggered it — which is how the scoring model gets better over time, not just for reporting.

{
  "contact_id": "ct_9f2a",
  "signal_theme": "LLM cost benchmarking",
  "engagement_type": "comment",
  "engagement_class": "evaluating",
  "composite_score": 0.81,
  "icp_fit_score": 0.74,
  "enrichment_source": "prospeo_primary",
  "email_verified": true,
  "sequence_id": "seq_7f3k",
  "sequence_stage": "linkedin_connection_sent",
  "reply_status": null,
  "signal_capture_date": "2026-03-14T09:22:00Z",
  "delivery_date": "2026-03-14T18:44:00Z",
  "deal_stage": "sequence_active"
}

What arrives each week

The delivered record. Every field, every week.

This is what lands in your Clay table, CSV, or webhook endpoint after the full pipeline runs.

Delivered every week. Routable into Clay, any CRM, or webhook. Format adjusts to your stack.