How to use AI for next-best-action marketing?

The Short Version

Next-best-action (NBA) marketing is the practice of using AI to determine which action—offer, message, channel, or timing—will most likely drive the desired customer outcome. Rather than running static campaigns to broad segments, AI analyzes individual customer behavior, preferences, and lifecycle stage to recommend the single best action in real time.

How Next-Best-Action Marketing Works

The Core Framework

Next-best-action marketing operates in three connected layers:

1. Data Collection & Unification — Consolidate customer data from all touchpoints (web, email, CRM, purchase history, support interactions, social behavior) into a single customer view. This is where a CDP (Customer Data Platform) becomes essential.

1. Predictive Modeling — Use AI to identify patterns: Which customers are likely to churn? Who's ready to upgrade? Which segment responds best to video vs. email? Which offer drives the highest lifetime value for this specific customer?

1. Real-Time Decisioning — When a customer triggers an action (visits your site, opens an email, abandons a cart), the AI engine instantly recommends the next best action and executes it automatically.

Why This Matters for CMOs

Traditional marketing sends the same message to everyone in a segment. Next-best-action marketing recognizes that Customer A might need a discount code, while Customer B (in the same segment) needs social proof, and Customer C needs a product recommendation. This personalization at scale drives measurable lift:

• 15-40% increase in conversion rates (depending on baseline maturity)
• 20-30% improvement in email engagement when recommendations are personalized
• Reduced marketing spend waste by eliminating irrelevant offers
• Faster customer progression through the funnel

Building Your Next-Best-Action Stack

Step 1: Audit Your Data Foundation

Before implementing AI, you need clean, unified customer data:

• Identify all customer data sources (CRM, marketing automation, analytics, e-commerce, support, social)
• Assess data quality: Are customer records duplicated? Is behavioral data complete? Is consent tracked?
• Define your customer identifier strategy (email, user ID, device ID, hashed phone number)
• Establish a single source of truth for customer attributes

Step 2: Choose Your Platform

You have three main options:

Option A: CDP + AI Engine (Recommended for most CMOs)

• Segment, mParticle, Tealium, or Treasure Data (CDPs with built-in AI)
• Salesforce Einstein (if you're already in Salesforce ecosystem)
• Adobe Real-Time CDP (if you're in Adobe ecosystem)
• Cost: $50K-$500K+ annually depending on data volume and features
• Timeline to first campaign: 3-6 months

Option B: Marketing Automation Platform with AI

• HubSpot with predictive lead scoring
• Marketo with AI-driven engagement scoring
• Klaviyo with predictive analytics (for e-commerce)
• Cost: $10K-$100K annually
• Timeline: 1-3 months (faster, but less sophisticated)

Option C: Custom AI + Existing Stack

• Build predictive models using Python, R, or cloud ML services (AWS SageMaker, Google Vertex AI, Azure ML)
• Integrate with your existing CRM and marketing automation
• Cost: $100K-$500K+ for development plus ongoing data science resources
• Timeline: 6-12 months
• Best for: Large enterprises with dedicated data science teams

Step 3: Define Your Next-Best-Actions

Start by mapping the actions you want to recommend:

• Offers: Discount code, free trial, upgrade incentive, loyalty reward
• Content: Product recommendation, educational article, case study, testimonial
• Channels: Email, SMS, push notification, in-app message, web personalization
• Timing: Immediate, delayed, optimal send time, lifecycle stage trigger
• Frequency: How often should this customer receive recommendations?

Pro tip from practitioners: Start with 3-5 high-impact actions rather than 20. You can expand once you see results.

Step 4: Build Your Predictive Models

Your AI engine needs to predict which action drives the best outcome for each customer:

Essential models to build:

• Propensity models: Which customers are likely to convert, churn, or upgrade?
• Affinity models: Which products/offers is this customer most likely to purchase?
• Channel preference models: Does this customer prefer email, SMS, or push?
• Timing models: When is this customer most likely to engage?
• Lifetime value models: Which customers are worth the most investment?

Most modern CDPs and marketing platforms include pre-built models you can activate immediately. Custom models require data science expertise.

Step 5: Implement and Test

Launch in phases:

1. Pilot phase (Month 1-2): Test NBA on one segment (e.g., high-value customers) or one channel (e.g., email)
2. Measure baseline: Track conversion rate, engagement rate, revenue per recipient, and customer satisfaction
3. Expand gradually: Once you see lift, expand to additional segments and channels
4. Continuous optimization: Retrain models monthly as customer behavior evolves

Key metrics to track:

• Conversion rate lift vs. control group
• Revenue per email/message
• Customer satisfaction (NPS, CSAT)
• Unsubscribe rate (watch for fatigue)
• Time to conversion
• Customer lifetime value

Real-World Implementation Example

E-Commerce Use Case

An online retailer uses next-best-action to recommend actions based on browsing and purchase history:

• Customer A (browsed but didn't buy): Recommend product with 10% discount code via email (high propensity to convert with incentive)
• Customer B (repeat buyer): Recommend complementary product via email (high affinity, no discount needed)
• Customer C (at-risk churn): Recommend exclusive loyalty reward via SMS (urgent, high-touch channel)
• Customer D (new visitor): Show social proof/testimonials on-site (needs trust-building, not a discount)

Result: 32% higher conversion rate vs. sending the same offer to all four customers.

Common Pitfalls to Avoid

• Insufficient data: Don't launch NBA with incomplete customer data. Garbage in = garbage out.
• Too many actions: Overwhelming your team and customers. Start small.
• Ignoring frequency caps: Customers get fatigued. Set limits on how often they see recommendations.
• Not accounting for consent: Ensure you have proper permissions for each channel and use case.
• Treating it as "set and forget": Models degrade over time. Retrain monthly and monitor performance.
• Prioritizing technology over strategy: The platform doesn't matter if you don't know what actions drive business value.

Timeline and Budget Expectations

For a mid-market CMO (500K-5M customer database):

• Technology investment: $100K-$300K annually
• Implementation timeline: 4-6 months to first campaign
• Team investment: 1-2 FTE (data analyst, marketing technologist)
• ROI timeline: 6-12 months to positive ROI

For an enterprise CMO (10M+ customers):

• Technology investment: $500K-$2M+ annually
• Implementation timeline: 6-12 months
• Team investment: 3-5 FTE (data scientists, engineers, analysts)
• ROI timeline: 3-6 months to positive ROI (higher volume = faster payback)

Bottom Line

Next-best-action marketing is no longer a competitive advantage—it's becoming table stakes for sophisticated CMOs. Start by auditing your data foundation, choose a platform that matches your maturity level (CDP for scale, marketing automation for speed), and launch with 3-5 high-impact actions. Measure religiously, optimize continuously, and expand gradually. Most CMOs see 15-30% conversion lift within 6 months of implementation, making this one of the highest-ROI marketing investments available.