Intermediate

Next Best Action

Next Best Action (NBA) engines determine the single most valuable action to take for each customer at each moment. Using reinforcement learning and contextual bandits, NBA systems optimize for long-term customer value rather than short-term engagement.

NBA Decision Framework

ComponentDescriptionExample
StateCurrent customer context and historyJourney stage, engagement level, recent behavior
ActionsAvailable marketing actionsSend email, show offer, trigger call, do nothing
RewardBusiness outcome to maximizeConversion, LTV, engagement, satisfaction
PolicyStrategy for choosing actions given statesLearned mapping from customer context to optimal action

Algorithm Approaches

🎲

Contextual Bandits

Select the best action based on current context, balancing exploitation of known good actions with exploration of potentially better ones.

🧠

Deep Q-Networks

Learn action values over multi-step horizons. Optimize for long-term customer lifetime value rather than immediate click or conversion.

📊

Propensity Models

Predict the probability of each action succeeding for each customer. Rank actions by expected value (probability x value) to find the best one.

🛠

Constraint Optimization

Select optimal actions while respecting business constraints: budget limits, inventory, channel capacity, and contact frequency rules.

Building an NBA Engine

  1. Define Action Space: Enumerate all possible marketing actions (offers, content, channels, timing options) the system can recommend
  2. Define Reward: Choose the business objective. Weight short-term conversion against long-term LTV and customer satisfaction
  3. Feature Engineering: Build customer state representations from journey history, engagement patterns, and real-time context
  4. Train Models: Use historical interaction data to train propensity models or initialize RL policies via offline learning
  5. Deploy with Exploration: Start with high exploration rates (20-30%) and gradually reduce as the system learns what works
  6. Monitor and Iterate: Track action distribution, reward trends, and edge cases to ensure the system behaves as intended
✅
Pro Tip: Include "do nothing" as an explicit action in your NBA system. Sometimes the best next action is no action at all. Aggressive marketing to every customer at every opportunity leads to fatigue. An NBA system that can choose restraint is more effective long-term.

Ready to Go Deeper?

Live instructor-led courses from our partners. Affiliate disclosure.

🧠

AI & ML Courses - 30% Off

Live instructor-led AI, machine learning, data science, and cloud courses for working professionals. Use code Limited30 at checkout.

Edureka
📊

DataCamp - AI & Data Science

Hands-on Python, machine learning, and AI courses with interactive exercises and real projects.

DataCamp
🎓

edX - Top AI Courses

University-level AI courses from MIT, Harvard, Stanford. Earn certificates that employers recognize.

edX