Latest / SaaS Business with Fexingo: Software-as-a-Service Companies, ARR, and Recurring Revenue / How SaaS Companies Use AI to Predict Customer Health
Transcript
- Lucas: Look, if you run a software company today, you probably have a dashboard that tells you how many logins happened last week. But here is the problem: that dashboard is a rear-view mirror. It shows you what already happened. The real edge right now belongs to companies that use artificial intelligence to predict what will happen next month. Luna: Right, because seeing a drop in logins is obvious, but knowing which specific user is about to cancel three weeks before they do? That changes the entire playbook. Lucas: Exactly. And we are not talking about sci-fi level predictions. We are talking about fairly simple machine learning models that ingest your existing data—login frequency, feature adoption, support ticket sentiment—and output a health score. A recent report from a major cloud infrastructure provider showed that firms using these predictive scores saw an eighteen percent reduction in churn compared to those relying on manual reviews. Luna: Eighteen percent is massive in this space. But does that model actually capture the human element, or is it just counting clicks? Lucas: That is the critical question. If the model only counts clicks, it is useless. You can have a power user clicking everything while secretly hating the product. The good models weight behavioral depth over raw volume. They look at whether a key decision-maker has logged in recently, or if the primary workflow tool is being used consistently. Luna: So it is less about activity and more about intent. How do you even start building something like that without a billion-dollar budget? Lucas: You start with what you already have. Most SaaS platforms record every event. The trick is labeling them correctly. Take a mid-market analytics firm I tracked recently. They didn't hire a new data science team. They took their existing SQL database, added a simple algorithm that flagged accounts where two out of five key users stopped logging in for ten days, and triggered an alert for their customer success team. Luna: Just a flag. Not an automated email, just a nudge to a human being. Lucas: Precisely. And that human touch is where the magic happens. The alert says, 'Hey, account X looks risky.' The success manager calls them. Turns out, their main champion was on vacation, so the other four users weren't logging in. It wasn't churn. It was just absence. If they had relied on a rigid rule, they might have fired that account anyway. Luna: That is such a simple fix, but it saves so much revenue. Why do so many companies miss this obvious nuance? Lucas: Because they treat data as a reporting tool rather than an operational one. They build dashboards for the boardroom, not triggers for the frontline. The gap is usually cultural. Leadership wants to see pretty charts. Operators want actionable alerts. Bridging that gap is where the efficiency gains live. Luna: I guess it comes down to trust. If the sales team doesn't trust the algorithm, they ignore it. Lucas: Trust is earned through transparency. The best systems don't give a black-box score. They explain why. 'Your health score dropped because login frequency fell by forty percent and no feature upgrades occurred in thirty days.' When you show the math, people believe the result. Luna: That makes sense. It turns a scary number into a diagnostic tool. What about false positives though? You don't want your team chasing ghosts. Lucas: False positives are the enemy of adoption. If the system screams wolf too often, everyone stops listening. The solution is tiered alerts. Low-risk flags go to a general inbox. High-risk flags ping the account owner directly. You calibrate the sensitivity based on your average contract value. Enterprise clients get tighter triggers; SMB clients get broader ones. Luna: So you customize the noise level based on who you are selling to. That is smart resource allocation. Lucas: It is. And it forces you to think about your customer segments. A startup using your software might behave totally differently than a Fortune五百 enterprise. One size fits all rarely works in predictive modeling. Luna: I imagine segmenting the data is half the battle. How do you handle that initial setup complexity? Lucas: Start small. Pick one segment. Pick one metric. Let's say you sell marketing automation software. Look at email send volume. If a user who sends ten thousand emails drops to one thousand, that is a red flag. Test that single variable against actual churn events. See if it correlates. Luna: Correlation is not causation, though. Just because they stopped sending emails doesn't mean they are leaving. Maybe they switched channels. Lucas: True. Which is why you layer the data. Combine usage drops with support ticket spikes. If usage goes down AND they ask for a refund, that is a strong signal. You are looking for patterns, not isolated incidents. The more dimensions you add, the clearer the picture becomes. Luna: It sounds like a detective story. You are gathering clues to solve the mystery of why someone leaves. Lucas: I love that analogy. And the best detectives don't wait for the crime scene. They patrol the neighborhood. Proactive intervention is the only way to win the retention war. Reactive measures are just damage control. Luna: Damage control is expensive. Prevention is cheaper. But how do you measure the ROI of these health scores? It feels abstract. Lucas: It is measurable if you track interventions. Record every time a success manager reaches out due to a health alert. Track the outcome. Did they renew? Did they expand? Calculate the lifetime value saved versus the cost of the intervention. Usually, the margin is huge. Luna: So it is an investment in relationship capital. You spend a little time saving the account, you keep the recurring revenue flowing. Lucas: Exactly. And it scales. Once you have the model trained, it runs in the background. Your team focuses on high-value conversations instead of guessing who needs help. Luna: It shifts the job description from firefighter to gardener. Tending to the plants before they wilt. Lucas: Perfect metaphor. Gardeners know when to water. Firefighters just put out flames. Which role do you prefer? Luna: Gardener any day. Less stress, better results. But it requires patience. You can't rush growth. Lucas: And you can't fake it with vanity metrics. Real health is quiet. It is consistent usage, steady engagement, happy support interactions. It doesn't always make for exciting headlines, but it builds a durable business. Luna: Durable businesses beat viral ones every time. Stability is the ultimate moat. Lucas: Agreed. So if you are building a product, stop asking how many people signed up yesterday. Ask who is still active tomorrow. That is the only number that matters. Luna: A powerful way to end the thought. Focus on the tomorrow, not the yesterday. Lucas: Before we wrap, I want to mention that listener support keeps this network running. If these insights help your operations, consider buying me a coffee at buy me a coffee dot com slash fexingo. Even a small amount helps cover the research costs and keeps the episodes ad-free. Luna: Yeah, it really makes a difference. We appreciate anyone who can chip in to keep the conversation going. Lucas: Thank you. Now, back to the topic. One final thought on those health scores: remember that technology amplifies strategy. If your strategy is weak, your AI will just predict failure faster. Fix the product first. Luna: Technology is a multiplier. It makes good things great and bad things worse. Choose wisely. Lucas: Exactly. Build a product people love, then use data to protect that love. That is the sustainable path forward. Luna: Love the product, protect the relationship. Simple principles, hard execution. Lucas: Hard work, yes. But worth it. Thanks for tuning in to this deep dive on predictive health scores. Luna: See you next time. Keep building.