Developing an AI-based Next Best Action recommendation system for sales managers
Next Best Action (NBA) is a recommendation system that tells the seller, at any given moment, "Here's what to do next with this deal." Not general advice, but specific actions with justification.
NBA system architecture
State Machine: Each deal has a specific state. State is determined by: CRM stage + activity (whether there's been a recent contact) + deal signals (new LI posts, company news) + time (approaching the end of the quarter).
Action Library: Possible actions include: call, send a case study, offer an extended trial, request a meeting with a decision maker, offer a discount, or ask for a referral. Each action has an expected value based on historical data.
Recommendation Engine: LLM + structured rules. LLM generates a contextual explanation ("It's recommended to call because: 12 days without contact, the company has posted a vacancy for your position, the quarter ends in 3 weeks").
Personalization: We take into account the specific manager's style—what actions they perform well and what they perform less frequently. Recommendations are tailored to suit each manager's needs.
Integration
CRM data + email tracking + LinkedIn activity monitoring + calendar. Recommendations appear in the CRM deal card and/or Slack/Teams notifications.