From Segmentation to Agents: Scaling Enterprise Personalization with Reinforcement Learning

Transforming business strategy as AI takes center stage: traditional customer segmentation is being challenged by new technologies. In this conversation, Paul Meinshausen, CEO of Aampe, and Dr. Tanwa Arpornthip, Senior Advisor at SCB 10X, analyze the future of enterprise personalization at AI-VOLUTION, emphasizing the shift toward an “Agentic Workflow” powered by Reinforcement Learning to deliver truly precise and individually tailored customer outcomes.

1. The end of “rule-based” — the rise of “agents”

Paul begins by highlighting the limitations of traditional tech stacks, which typically operate in a hierarchical, rule-based manner—such as “if the customer does A, send message B.” This approach lacks flexibility and cannot truly respond to the complex, dynamic context of each individual user.

Aampe offers a new concept: “Agentic Infrastructure”—an agent-driven backbone. The core idea is one AI agent per user. If an application has 1 million users, it will have 1 million agents, each learning that specific user’s behavior, history, and context. The agent then decides what message (notification) to send or what screen to show to create the greatest value for that user.

2. Compared to self-driving cars: context is everything

Paul uses a compelling analogy to compare traditional A/B testing with agentic personalization:

• Traditional approach: Like driving a car using only a map and traffic rules—turning left or right at predetermined points.
  
• Agent-based approach: Like an autonomous vehicle that must make decisions based on real traffic conditions in the moment—pedestrians crossing, traffic jams, unexpected obstacles.
  

The agent continuously learns and adjusts its actions according to real-world context, not just pre-programmed rules.

3. Strategy stays with humans, tactics go to agents

Many organizations worry about losing control, but Paul explains that roles are clearly divided:

• Business teams: Define the strategy and objectives, such as which inventory needs to be cleared, or which product categories should see higher sales.
  
• Agents: Handle the tactics—deciding how to communicate with each individual user to achieve those goals.
  

Example: An e-commerce platform has excess stock of athleisure but, at the same time, professional workwear is trending.

• The business team sets the reward values—how important it is to sell each category.
  
• The agent then decides, for each user (e.g., Mr. A), whether to show professional wear (because he is more likely to buy it) or to try promoting athleisure (because selling it would be more valuable for the business).
  

4. Tackling scale and coordination

When running an enormous number of agents (for example, one app in Southeast Asia runs 80 million agents per month), the challenge isn’t just infrastructure—it’s coordination.

Aampe’s agents are designed to support cross-user learning: they learn from the behavior of similar users so that each agent becomes smarter without having to learn everything from scratch. At the same time, they preserve each user’s privacy and unique context.

5. Breaking down organizational silos

A major obstacle to adopting this technology is not technical—it’s about people and organizational structure.

Traditionally, Marketing, Product, and Sales teams may use different tools and measure success differently, causing fragmented customer experiences. Agentic Infrastructure allows all teams to share the same “brain,” with a single agent looking after the same user across all channels.

Paul suggests that the best way to start is to educate teams, then run a pilot with one team first. Once clear results are measured, the organization can scale up. Agents can also log and report their own decisions more accurately than legacy systems.

Summary

The shift from segmentation to agents is not just a software upgrade—it is a fundamental change in how we build customer relationships. We move from broad “casting the net” over segments to truly caring for individuals, using AI that learns and grows together with each customer.