# Omnichannel AI SDR & Revenue Operations Engine OS ==================================================================== MODULE 1: OMNICHANNEL CADENCE TOPOLOGY & TOUCHPOINT STATE MACHINE ==================================================================== - Omnichannel Cadence Strategy & Journey Architecture: * Target ICP: VPs of Infrastructure, Heads of Platform Engineering, and Directors of Cloud Operations at high-growth SaaS firms ($20M-$100M ARR). * Multi-Touchpoint Sequence Topology: - Touch 1 (Day 1 - 09:30 AM EST): Contextual Cold Email (Observation hook regarding Kubernetes cloud cost anomalies). - Touch 2 (Day 3 - 02:15 PM EST): LinkedIn Profile Touch + Peer Connection Request (Referencing recent Kubernetes scaling post). - Touch 3 (Day 6 - 10:00 AM EST): Email Follow-Up (Case study demonstrating 28% reduction in AWS EKS cluster spend). - Touch 4 (Day 9 - 11:30 AM EST): 30-Second Personalized Conversational Voice Note on LinkedIn or phone task. - Touch 5 (Day 13 - 03:00 PM EST): Final Permission-Based Exit Email (Closing loop without guilt-tripping). - Deterministic State Transition Matrix: | Current State | Trigger Event | Guard Condition | Next State | System Action Taken | | :--- | :--- | :--- | :--- | :--- | | `INGESTED` | Lead imports from Clay/CRM | Passed ICP verification & verified email | `QUEUED` | Assigns dedicated sending mailbox pool | | `QUEUED` | Cadence start schedule reached | No active suppression tag | `TOUCH_1_DISPATCHED` | Emits Email 1; schedules Day 3 LinkedIn touch | | `TOUCH_1_DISPATCHED` | Inbound reply detected | Positive or neutral sentiment | `HANDOFF_PENDING` | Instantly freezes LinkedIn & Voice touches; alerts AE | | `TOUCH_1_DISPATCHED` | No reply after 48h | Uncontacted on LinkedIn | `TOUCH_2_DISPATCHED` | Fires LinkedIn connection request via API | | `ANY_ACTIVE_STATE` | Unsubscribe keyword detected | Reply contains "opt out" or "stop" | `GLOBAL_SUPPRESSED` | Cancels all pending touches; syncs DNC to CRM | ==================================================================== MODULE 2: LINKEDIN SAFETY GOVERNANCE, PACING & ANTI-FATIGUE RULES ==================================================================== - Safe Outreach Limits & Activity Pacing Thresholds: | Channel / Action | Daily Safe Ceiling | Minimum Delay Gap | Anti-Fatigue Policy | | :--- | :--- | :--- | :--- | | LinkedIn Connection Requests | 18 requests / day / account | 240 to 480 seconds (Randomized) | Maximum 80 pending requests at any time | | LinkedIn Profile Views | 30 visits / day / account | 120 to 300 seconds | Staggered over 8 business hours | | Cold Outbound Emails | 25 emails / day / mailbox | 180 to 360 seconds | Multi-mailbox rotation across 6 domains | - Multi-Channel Global Suppression Architecture: * Prior to dispatching any touchpoint, the engine checks a Redis central cache: if `prospect_domain` exists in `crm:active_opportunities` or `crm:customers`, the lead is immediately purged with reason `ACTIVE_DEAL_PROTECTION`. ==================================================================== MODULE 3: DYNAMIC CONTEXTUAL MESSAGING & MULTI-TIER PERSONALIZATION ==================================================================== - Multi-Channel Copy Generation Matrix (Verbatim Production Assets): * Asset 1: Cold Email Thread ```text Subject: {{company_name}}'s Kubernetes compute costs Hi {{first_name}}, Saw {{company_name}}'s engineering team recently published a teardown of your multi-region EKS deployment on your engineering blog. Most Platform Engineering leaders we speak with mention that as cluster nodes scale, unallocated memory overhead and spot instance terminations quietly drive AWS bills up by 25-35%. We developed CloudScale to provide real-time container rightsizing with zero pod disruption. Last quarter, it helped Datadog and Scale AI cut EC2 compute waste by 28% within 3 weeks of deployment. Open to reviewing a brief 2-minute Loom teardown comparing our resource allocation against standard Karpenter scaling? Best, Sarah Jenkins CloudScale Systems CloudScale Inc., 100 Montgomery St, San Francisco, CA Unsubscribe: Reply "opt out" to be permanently removed. ``` * Asset 2: LinkedIn Connection Suite ```text // Connection Request (218 characters): Hi {{first_name}}, noticed your team's work scaling Kubernetes clusters at {{company_name}}. Enjoyed the recent engineering team post on container telemetry. Would love to connect and follow your architectural updates here. // Follow-Up Message (Triggered 6 hours after connection accepted): Thanks for connecting, {{first_name}}. Saw you are managing infrastructure growth as {{company_name}} scales. We recently published a benchmark report on optimizing unallocated EKS memory without compromising SLOs. Happy to send the PDF your way if relevant to your roadmap this quarter? ``` * Asset 3: 30-Second Conversational Voice Note Script ```text [Tone: Natural, casual peer, zero sales urgency. Clear audio, background office ambient.] "Hey {{first_name}} — Sarah here from CloudScale. Just saw the engineering post on how {{company_name}} is scaling out the platform team this year. [Pause - 0.5s] Was just looking at how teams running multi-region clusters handle unallocated node overhead without forcing devs to manually tweak CPU requests. [Pause - 0.5s] Sent a quick note to your inbox with our latest Kubernetes benchmarking data. No need to call me back here, just wanted to put a face to the name. Cheers!" ``` ==================================================================== MODULE 4: INBOUND INTENT CLASSIFICATION & AE HAND-OFF ROUTER ==================================================================== - Production Intent Classification Engine (`src/intent_classifier.py`): ```python import re from typing import Dict, Any from pydantic import BaseModel, Field class IntentResult(BaseModel): category: str = Field(description="POSITIVE_MEETING, OBJECTION_MORE_INFO, REFERRAL_REDIRECT, HARD_OPT_OUT") confidence: float extracted_email: str = "" action_required: str def classify_inbound_reply(reply_body: str) -> IntentResult: cleaned = reply_body.lower().strip() # 1. Hard Opt-Out Check if re.search(r"\b(unsubscribe|opt-out|remove me|stop emailing|not interested|take me off)\b", cleaned): return IntentResult(category="HARD_OPT_OUT", confidence=0.99, action_required="SUPPRESS_GLOBALLY") # 2. Positive / Meeting Intent if re.search(r"\b(send over|loom|calendar|book|demo|chat next week|open to|tuesday|pricing)\b", cleaned): return IntentResult(category="POSITIVE_MEETING", confidence=0.95, action_required="ALERT_AE_SLACK_AND_CREATE_DEAL") # 3. Referral / Redirect Check referral_match = re.search(r"\b(?:reach out to|contact|speak with)\s+([a-zA-Z\s]+?)(?:\s+at|\s+on|\.|$)", cleaned) if referral_match: return IntentResult(category="REFERRAL_REDIRECT", confidence=0.88, extracted_email=referral_match.group(1).strip(), action_required="ROUTE_NEW_PROSPECT") # 4. Default Objection / Clarification return IntentResult(category="OBJECTION_MORE_INFO", confidence=0.75, action_required="QUEUE_SDR_REVIEW") { "channel": "c-sdr-inbound-wins", "blocks": [ { "type": "header", "text": { "type": "plain_text", "text": "🎯 High-Intent Omnichannel Reply Detected!" } }, { "type": "section", "fields": [ { "type": "mrkdwn", "text": "*Prospect:* Sarah Connor\n*Company:* Cyberdyne Systems" }, { "type": "mrkdwn", "text": "*Channel:* Email (Touch 1)\n*Intent:* POSITIVE_MEETING" } ] }, { "type": "section", "text": { "type": "mrkdwn", "text": "> \"Sure, send over the 2-minute Loom teardown. We're actually evaluating Karpenter right now.\"" } }, { "type": "actions", "elements": [ { "type": "button", "text": { "type": "plain_text", "text": "Open in HubSpot" }, "url": "[https://app.hubspot.com/contacts/99214/contact/88412](https://app.hubspot.com/contacts/99214/contact/88412)" } ] } ] } from fastapi import FastAPI, Header, HTTPException, Request from src.intent_classifier import classify_inbound_reply import httpx import redis app = FastAPI(title="Omnichannel SDR Webhook Router") r = redis.Redis(host="localhost", port=6379, db=0) @app.post("/webhooks/inbound-reply") async def handle_inbound_reply(request: Request, x_webhook_secret: str = Header(None)): if x_webhook_secret != "sec_sdr_live_auth_token_99": raise HTTPException(status_code=401, detail="Unauthorized webhook secret") data = await request.json() prospect_email = data.get("email") reply_text = data.get("reply_body", "") # 1. Classify Sentiment intent = classify_inbound_reply(reply_text) # 2. Halt All Multi-Channel Sequences Immediately r.set(f"cadence:suppressed:{prospect_email}", "1", ex=86400 * 30) # 3. Call Smartlead / Instantly API to pause campaigns async with httpx.AsyncClient() as client: await client.post( "[https://server.smartlead.ai/api/v1/campaigns/lead/pause](https://server.smartlead.ai/api/v1/campaigns/lead/pause)", json={"email": prospect_email, "api_key": "YOUR_SMARTLEAD_KEY"} ) return {"status": "SUCCESS", "intent_classified": intent.category, "cadence_paused": True}