At FunnelStory, we've been thinking a lot about the challenges facing sales and customer success teams today and what we hear from current customers and prospects. New sales are down across industries, and organizations are scrambling to improve retention and expansion. But here's the problem: most retention and churn prediction is still based on guesswork and intuition. Post-sales teams lack the data-driven tools they need to learn from previous opportunities and make informed decisions.
That's why we've developed our Revenue Protection AI. Our goal? To provide truly data-driven revenue predictions that go beyond surface-level insights. We wanted to create a solution that could not only predict outcomes but also answer the all-important question: "Why?"
The problem with health scores
Many teams start their customer health monitoring journey with basic health scores. Without sophisticated tools, they often resort to manually setting Red/Yellow/Green statuses for their accounts. This approach, while a start, has significant drawbacks.
It's primarily based on intuition rather than data
The health score becomes heavily dependent on each account owner's personal opinion and experience, leading to inconsistencies and potential blind spots.
Some organizations attempt to improve this by implementing data-driven health scores, where the Red/Yellow/Green status is determined by specific metrics. However, this approach brings its own set of challenges. Teams struggle with
Deciding which metrics to use
How to weigh their importance
Setting appropriate thresholds for these metrics
How these vary based on customer segment
Moreover, even when an account's health score is marked as red, indicating a potential problem, the work has only just begun. Teams then have to manually investigate to understand why the account is at risk, a process that can be time-consuming and may not always yield comprehensive insights.
FunnelStory's Revenue Protection AI
Our Revenue Protection AI is designed to address these challenges head-on. Unlike traditional systems, it doesn't require you to select metrics, set thresholds, or perform any complex manual setup. Instead, FunnelStory automatically discovers what characteristics good and bad accounts have in common. It identifies the factors and metrics that truly matter, determines how much each factor contributes to the overall health, and establishes various thresholds dynamically.
The AI's consideration is comprehensive and nuanced. It takes into account a wide range of factors including product usage patterns, feature adoption rates, engagement trends over time, license utilization, specific product activities, frequency and quality of meetings, conversation content, support ticket history, and much more. In practice, our AI can easily analyze over 100 different factors to provide a holistic view of account health.
But our AI doesn't stop at prediction. Once it makes a prediction about an account's health or likelihood of churn, it goes a step further by diving deep to uncover the "whys" behind its assessment. For instance, if the AI flags an account for low product usage, it can answer in-depth questions such as:
Has the usage always been low, or is this a recent development?
Have users reduced their activity over time?
Are there any mentions of bugs or issues in recent meetings or chat logs?
This level of detailed analysis provides actionable insights that teams can use to intervene effectively and improve outcomes.
AI + ML
But how did we go about building this? It’s not just AI, and it’s not just machine learning - it’s both. Let me break it down a bit.
Machine learning is all about algorithms and techniques. It’s the nuts and bolts of probability and statistics, regression, clustering, forecasting, and language models. AI, on the other hand, is about problem-solving, decision-making, and natural language processing and synthesis.
We’ve brought together tried-and-tested ML techniques with state-of-the-art Large Language Models (LLMs) to create something truly powerful. Our prediction models use algorithms and techniques like random forests and regression, while our in-depth analysis is powered by agentic AI and LLMs.
What really made this possible was the strong foundation we already had in place. Our data ingestion layer allows us to pull in both structured and unstructured data from databases, warehouses, and apps. And our Customer Activity Model, a knowledge graph of your customer data, gives our AI a rich context to work with.
ML savvy readers may be curious about how our prediction models handle large numbers of factors. How do we handle overfitting, multicollinearity, etc.? One of the primary techniques we leverage is ensemble learning. Instead of one ML model, we use a mixture of several algorithms and models trained on different subsets of data. Together this ensemble votes to come up with a prediction. The dozens of individual components come together to build a much stronger ensemble.
An ensemble that’s greater than the sum of the parts
Our Revenue Protection AI is powerful because it has access to our Customer Activity Model knowledge graph combined with sophisticated machine learning tools and LLMs. It’s not just regurgitating information and providing surface-level insights; it’s analyzing, predicting, doing a deeper diagnosis by going multiple levels deep and asking “why?” It’s not just about predicting churn - it’s about understanding why it happens and what you can do about it.
Perhaps what I’m most excited about is the UX achievement. This is the first time we’ve introduced entire UI components driven by LLM. It’s a big step forward in making AI truly accessible and useful for our users.
The best part? All of this comes with minimal configuration required on your end. We’ve done the heavy lifting so you can focus on what matters: making data-driven decisions to improve your customer retention and revenue.
We’ve released this feature to a few early adopters, and are energized by seeing their eyes light up and hearing their positive feedback! If you’re interested in learning more or trying it for your own organization, please contact us.
