{
  "video": {
    "id": "ltv-L5oMPIs",
    "title": "Build the AI GTM Agent That Knows the Buyer - Dr. Sajjan Kanukolanu, Position2 (Position Squared)",
    "duration": 1587,
    "upload_date": null,
    "channel": "AI Engineer",
    "source": "AI Engineer",
    "event_format": "Online"
  },
  "analysis": {
    "video_id": "ltv-L5oMPIs",
    "title": "Build the AI GTM Agent That Knows the Buyer - Dr. Sajjan Kanukolanu, Position2 (Position Squared)",
    "one_liner": "Position2's VP of global ops argues that bolting AI onto an existing GTM stack can't scale, and walks through the three-layer architecture — signals, buyer intelligence, action — plus the connected agents and live dashboards they run so a buyer is identified and personalized to the moment they land on the site.",
    "summary": "Sajjan Kanukolanu argues that B2B buyers now arrive late-stage with the decision mostly made, so a chat window asking \"how can I help you?\" pushes them backwards. He lays out three problems GTM teams must solve simultaneously — AI that can't know the buyer on its own, broken integrations, and an architecture that wasn't built with AI at the core — and maps them onto three layers: signals (CRM, enrichment, LinkedIn), buyer intelligence (knowledge base, de-anonymized identity, ICP scoring, context builder, routing logic) and action (personalized chat, rep alerts, CRM context, sequence triggers). He demos Position Squared's system, \"Intelligence,\" showing de-anonymized visitors and LinkedIn engagement dashboards, then names the four things that break in production: ICP drift, alert fatigue, the identity ceiling and the human approval bottleneck.",
    "key_points": [
      "Position Squared has launched 75+ AI agents for clients, powered by 18+ vertical knowledge bases, doing 800+ runs per month (month-to-date June 2026).",
      "Buyer stats cited: 94% of buyers use GenAI as primary research (Forrester 2026), 67% of B2B buyers prefer a rep-free experience, 80% of deals go to vendors on the pre-contact list, and only 17% of total buying time is spent talking with potential vendors.",
      "Three problems must be solved together — AI alone can't determine the buyer's role, history or intent; legacy GTM stacks can't capture intent signals and CRM context; and the underlying architecture must be rebuilt with AI at the core. Solving one and two but not three still leaves you stuck.",
      "Three-layer architecture: signals (CRM deals/contacts/owners, enrichment of anonymous visitors, LinkedIn engagement and job changes); buyer intelligence (knowledge base of product, ICPs, personas, titles, playbooks and success criteria; visitor identity; ICP scoring that separates fit from buying stage; context builder; routing logic deciding message, channel and timing); action (personalized chat, rep alerts, CRM context updates, sequence triggers).",
      "The context graph is a per-buyer connected record linking persona, account and deal — a person's signals roll up to their account so the buying committee surfaces, and deal-level data shows whether sales or marketing has already reached out. Without it you can't tell which accounts to prioritize.",
      "Implemented as connected agents: a visit kicks off an identification job using more than one identity source (consolidated and deduped), then enrichment, then an ICP filter agent that checks the knowledge base and deletes wrong-industry/geo/size/persona contacts into nothing while qualified ones go to a clean table, then a match agent that checks CRM for warm/hot, then an action agent that decides per person or per buying committee: Slack alert with context and a draft email, email send, or LinkedIn outreach.",
      "Demo of their system \"Intelligence\": the chat greets the speaker by name and links back to his earlier conversation rather than asking generic questions; the GTM dashboard shows ~3,000 identified visitors from ~280 accounts with information technology as the top vertical, hot signals blended with HubSpot conversation history; the LinkedIn intelligence card shows ~100 visitors from 73 companies engaging with 8 recent posts, sortable by C-suite, VP, director, manager and IC.",
      "Four failure modes: ICP drift (retrain agents every quarter on closed-won/closed-lost); alert fatigue (flag everything hot and reps stop acting and stop trusting the system); the identity ceiling (visitor-ID tools are ~70%+ accurate on company but only 15–20% on individual — a structural limit, so keep testing tools and plan around the numbers); the human bottleneck (an approval queue that feels like a chore gets backlogged and ignored).",
      "A specific latency threshold for human-in-the-loop: if editing an AI-drafted email takes longer than 30 seconds, reps will write their own and the initiative is dead."
    ],
    "takeaways": [
      "Stop opening the on-site chat with \"How can I help you?\" — by then the buyer has already researched and shortlisted; identify them, address them by name, and resume where they left off or serve content matched to their intent signals.",
      "Run more than one visitor-identification tool in parallel and dedupe the results, and set expectations against the identity ceiling: ~70% company-level accuracy, 15–20% individual-level.",
      "Score fit and intent as separate dimensions — conflating them sends the wrong message to the wrong person — and use the knowledge base plus context graph to suppress alerts so only accounts that genuinely matter reach a rep.",
      "Retrain agents every quarter against closed-won and closed-lost opportunities, and feed every send, reply, win, loss and deferred deal back into the knowledge base so the flywheel compounds.",
      "Build the policy engine so a GTM team member — not a developer — can audit which agent or workflow step broke and fix it, and shrink draft-to-send friction to a one-click, under-30-second edit.",
      "Treat LinkedIn engagement and job changes as first-class automated signals: when an exec sponsor leaves for a company not on your list, add that company and carry the champion across."
    ],
    "topics": [
      "agents",
      "go-to-market",
      "b2b-sales",
      "knowledge-bases",
      "identity-resolution",
      "context-graph",
      "crm",
      "human-in-the-loop"
    ],
    "tools": [
      "Position2 (Position Squared)",
      "Intelligence (Position Squared's internal GTM system)",
      "HubSpot",
      "LinkedIn",
      "Slack",
      "Forrester (2026 research)"
    ],
    "quotes": [
      {
        "text": "The uncomfortable truth is that by the time the buyer reaches you, the decision is mostly made.",
        "at": "00:06",
        "url": "https://www.youtube.com/watch?v=ltv-L5oMPIs&t=6s"
      },
      {
        "text": "only 17% of the total buying time is spent talking with potential vendors.",
        "at": "04:36",
        "url": "https://www.youtube.com/watch?v=ltv-L5oMPIs&t=276s"
      },
      {
        "text": "I'm using more than one source here running together, and that's because no single tool catches every visitor.",
        "at": "13:54",
        "url": "https://www.youtube.com/watch?v=ltv-L5oMPIs&t=834s"
      },
      {
        "text": "But if the time taken to edit an email that the AI writes is, let's say, longer than 30 seconds, then this is a dead initiative because the sales would then rather focus on drafting their own email as opposed to trusting what the system gives them.",
        "at": "24:11",
        "url": "https://www.youtube.com/watch?v=ltv-L5oMPIs&t=1451s"
      }
    ],
    "words": 4675
  },
  "summary_url": "/#ltv-L5oMPIs",
  "transcript": {
    "html": "/transcripts/ltv-L5oMPIs.html",
    "txt": "/transcripts/ltv-L5oMPIs.txt",
    "vtt": "/transcripts/ltv-L5oMPIs.vtt"
  }
}