Our Head of Data Science Erik Mathiesen-Dreyfus explains, why foundational customer models bridge the gap between generative AI and predictive analytics and how they can reshape the way businesses make decisions.
AI for businesses: GenAI & predictive analytics #
With the widespread development and spread of AI-solutions, two different AI applications have gained popularity among business cases.
On the one hand, you have predictive analytics* offering narrow solutions for very specific topics such as churn prediction, lifetime value calculation, fraud detection, lead scoring and more.
*By collecting and processing historical data, predictive analytics generate probable outcomes (just like your daily weather report) for future events.
At the same time, generative AI, mostly in form of large language models (LLMs), create summaries, translations, transcripts and more for marketing, sales, support and content creation. With its many applications (chatbots, email generation, product descriptions, search results), GenAI has become almost ubiquitous in our business and private lives.
However, these two powerful technologies often operate in silos. If they were combined, they could create a holistic view of the customer and the business.
The next level: the foundational customer model #
Foundational modelling – which is, for example, used in LLMs to train on vast datasets – can bridge the divide when it is applied to customer behavior as a whole instead of content.
The foundational customer model is a general-purpose AI system developed to understand and simulate customer behavior across time. Rather than building a separate model for different business questions (e.g. churn, LTV, conversions), a single, unified model is trained on large-scale anonymized behavioral data.
The foundational customer model captures patterns in customer behavior, for a vast variety of touchpoints along the customer journey:
- Subscription
- Upgrades
- Churn
- Re-engagement
- Incentives
The model is not trained to answer a single, predefined question. Quite to the contrary, it learns a generative process: how customer behavior evolves under different conditions and business interventions.
How does it work? #
The foundational customer model operates similar to an LLM. Instead of learning the flow and structure of language, it learns the flow and structure of lifecycle behavior: how people join, upgrade, churn, respond to incentives, and move through the customer journey. Once trained on general data, it can be adapted for individual businesses.
- A standard approach starts with pre-training on general behavior patterns that can be observed across many businesses.
- The model is then fine-tuned to a specific industry.
- Finally, the individual’s company’s specific dataset is introduced to create unique context (e.g. pricing structure, plan tiers, customer lifecycles, and product analytics).
This results in a high-fidelity simulation engine that projects customer behavior over time for both existing/observed and hypothetical conditions.
This foundational approach enables the modelling of customer journeys instead of isolated touchpoints and provides a flexible, integrated system for strategic decision making.
- Test “what-if”-scenarios for pricing or product changes
- Forecast key metrics such as revenue, MRR, and churn under different conditions
- Understand behavior-drivers
- Personalize precise interventions like retention campaigns
Who should use foundational customer models? #
Especially industries that can gather large amounts of customer behavioral data can profit from foundational customer models. Additionally, any business that gets significant gains from small improvements to retention or upsell rates, can compound its results if it can base its optimizations on precise data predictions.
By being able to simulate even single customer’s behaviors over time, businesses not only get insights into what is happening but also get information on what could happen (and why), making it a lot easier to plan ahead.
The future of business AI #
Foundational customer models are still emerging and cutting-edge. But their advantages for actionable insights based on both general and specific datasets have a huge potential.
Instead of implementing AI first and defining a use for them later, these models immediately solve existing issues and challenges
Frisbii likes to live on the cutting edge: our Revenue Insights Labs provide you with both performance and predictive analytics to monitor, control and plan your business strategy. As part of our recurring revenue management platform, they allow for an all-in-one solution to automate, manage and optimize your recurring billing, payments and customer data.
Book a meeting with one of our experts and see how it works in a live demo.
Our Author: #
Erik is a former quant and two-time founder with a PhD in mathematics. He’s led data science at fast-scaling startups like Streetbees and Attest and now heads Frisbii’s AI strategy.
At Frisbii, he helps our customers turn raw billing data into predictive, explainable insights – powering smarter pricing, churn prevention, and LTV growth.
Connect with Erik on LinkedIn
