An AI GTM system needs signals, buyer intelligence, and action connected through an architecture designed around AI.
2
Visitor identity, CRM history, intent, ICP fit, and social activity should be resolved before the first chat message.
3
The system fails when ICP data drifts, every alert becomes urgent, visitor identity is weak, or sales approval takes too long.
Summary
Dr. Sajjan Kanukolanu argues that most AI GTM deployments start at the wrong point. They add a chatbot to an existing stack, then ask visitors what they need even though those buyers have already researched vendors and may have shortlisted suppliers. His proposed system starts by resolving identity and gathering signals from the CRM, enrichment tools, LinkedIn, and site activity. A knowledge base defines products, personas, ICP rules, and playbooks. A context graph connects each person to accounts, buying committees, deals, and prior interactions. Agents then score, match, route, personalize the website conversation, alert sales, update the CRM, and trigger outreach. The implementation has limits. ICP rules need regular updates, identity tools identify companies more reliably than individuals, excessive alerts reduce trust, and sales will reject drafts that take too long to edit. The system must also feed replies, wins, losses, and deferred deals back into its knowledge base.
The buyer often reaches the vendor after most of the decision is made
Kanukolanu opens with the claim that modern B2B buyers research, compare, and shortlist vendors before contacting them. He cites Forrester 2026 research saying 94% of buyers use GenAI for primary research, 67% prefer a rep-free experience, and 80% of deals go to buyers on a pre-contact list. Buyers spend only 17% of their total buying time talking with potential vendors. That makes a generic website greeting poorly timed. Asking, "How can I help you?" or "What would you like to know?" forces a researched buyer to repeat an earlier stage of the process.
AI, integration, and architecture have to be solved together
Kanukolanu describes three connected problems. A standalone AI system cannot understand a buyer's role, history, or intent. Existing GTM systems may fail to capture and combine the signals that AI gathers with CRM context. The underlying architecture also needs to change so AI is at its core. Solving only the AI and integration pieces leaves the deployment unable to scale. His approach treats these as one design problem rather than placing a language model on top of unchanged processes and workflows.
Signals provide the raw material for buyer intelligence
The signals layer combines CRM records about deals, contacts, owners, and sales representatives with enrichment data about anonymous visitors. Kanukolanu also puts weight on social signals, especially LinkedIn engagement and job changes. He gives the example of an executive sponsor who leaves a current account for another company. Capturing that move can identify a new account and preserve the relationship with the same champion. These signals help determine who to target and when, provided the enrichment data is accurate.
A knowledge base and context graph turn signals into decisions
The buyer intelligence layer stores product context, ICPs, personas, titles, playbooks, success criteria, visitor identity, and scoring rules. It separates fit from buying stage and research activity. A context builder then connects the signals into a graph. Each person's signals link to an account, while account records include multiple people, buying committee activity, and deal-level sales and marketing history. This lets the team prioritize high-intent accounts and contacts instead of treating every visitor as an equal opportunity.
Connected agents identify, qualify, match, and route visitors
When an anonymous visitor arrives, an identification job uses multiple sources because no single tool catches every visitor. The system consolidates and deduplicates the results before enrichment. An ICP filter agent checks the knowledge base and removes contacts that fail industry, geography, company-size, or persona rules. A match agent checks the CRM to see whether people are known, warm, or hot. An action agent then decides whether to alert sales in Slack, prepare or send an email, use LinkedIn outreach, or act across a buying committee.
Personalization starts before the visitor sends a message
In the demo, the website recognizes a returning visitor and offers to continue an earlier conversation instead of opening with a generic question. The dashboard also groups anonymous visitors by account and industry, then surfaces hot contacts and hot accounts using signals that include HubSpot conversations. A separate LinkedIn view shows visitors who arrived through LinkedIn, the companies they represent, the posts they engaged with, and their seniority. The team can sort these records by industry, seniority, and engagement.
Operational limits can destroy trust in the system
Kanukolanu names four failure points. ICPs drift, so the knowledge base and agents need retraining each quarter using closed-won and closed-lost opportunities. Alert fatigue appears when everything is marked hot, causing salespeople to stop acting. Identity tools are structurally limited, with company identification described as about 70% accurate and individual identification at about 15% to 20%. There is also a human bottleneck. If editing an AI-written email takes longer than 30 seconds, salespeople will prefer writing their own.
The system needs control and feedback from real outcomes
His final design requirements are to start with visitor identity, keep fit separate from intent, and build an auditable, adjustable policy engine. GTM teams should be able to find which agent or workflow step failed and change it without waiting for a developer. The system should also feed every send, reply, closed deal, loss, and deferred sale back into the knowledge base. Those outcomes update future decisions and give the agents a way to improve from actual GTM activity.
"If everything is flagged as hot to a sales rep, they would stop acting because it just gets overwhelming for them."22:03
Who should watch
You are adding an AI chatbot to a B2B website and need to connect it to identity, CRM records, intent signals, and sales routing.
Your GTM team has visitor data and alerts but cannot tell which accounts deserve attention or how to personalize the first interaction.
You need to set practical limits for identity accuracy, alert volume, ICP updates, and human approval time.
