I remember the exact moment I decided to leave my first bank. I was 22, just starting my career, and I got hit with a $35 overdraft fee for a purchase that was off by less than a dollar. I called customer service, explained the situation, and was met with a robotic, unhelpful script. They wouldn't budge. I closed my account the next day and never looked back. It wasn’t about the money; it was about how they made me feel. Just another number in their system.
That experience stuck with me. Now, as an investor in over 200 companies, including some of the biggest names in AI like Anthropic and OpenAI, I see the same problem from a different angle. Customer churn is a massive, expensive problem for banks. And for the longest time, they've been trying to solve it with outdated methods. But that’s all changing.
Why Customers Walk Away
It’s rarely one single thing that makes someone switch banks. It’s a death by a thousand cuts. It could be high fees, a clunky mobile app, better interest rates elsewhere, or a single bad customer service experience like mine. For years, banks have been reactive. They wait until a customer is already halfway out the door before they do anything.
Think about it. The traditional approach is to look at lagging indicators. A customer starts withdrawing large sums of money or stops using their debit card. By the time the bank notices, it’s often too late. The decision has already been made. They’re just playing defense.
The AI Crystal Ball
This is where AI flips the script. Instead of looking backward, we can now look forward. We can predict, with a startling degree of accuracy, which customers are likely to churn before they even know it themselves.
At one of my portfolio companies, we built a churn prediction model for a large retail bank. The amount of data we had was staggering. We weren’t just looking at account balances and transaction history. We pulled in everything:
- Demographics: Age, income, location.
- Product Usage: How often they use the mobile app, which features they use, how many products they have with the bank (checking, savings, mortgage, etc.).
- Customer Service Interactions: Transcripts from chat logs, sentiment analysis of phone calls.
- Web Behavior: Pages they visit on the bank's website.
- External Data: Even public social media data can be a goldmine.
We fed all of this into a series of machine learning models. We started with simpler models like logistic regression to get a baseline, but the real magic happened when we moved to more complex models like gradient boosting and neural networks. The models started to find patterns that no human analyst would ever spot.
For example, we found a strong correlation between customers who stopped using bill pay for their utilities and a high likelihood of churn within 90 days. Or customers who called customer service twice in one month about fees, even if the issue was resolved, were three times more likely to leave. It’s like the AI could sense the frustration building up.
This is where the work of companies like Scale AI and Hugging Face becomes so critical. You need massive amounts of clean, well-labeled data to train these models effectively. It’s not just about having the data; it’s about having the right data, structured in a way that the model can understand.
From Prediction to Action
This is the part that really excites me. A prediction is useless without action. Knowing a customer is a churn risk is one thing; doing something about it is everything.
Once the model flags a customer as a high churn risk, the bank can be proactive. Instead of waiting for the angry phone call, they can reach out first. And not with a generic, "We value your business" email. But with a highly personalized offer based on that customer's specific situation.
- High fees? The model can trigger an offer for a new account type with no monthly fees.
- Low interest rates? A targeted promotion for a high-yield savings account.
- Bad customer service experience? A personal call from a senior relationship manager to address the issue and offer a goodwill gesture.
This is where conversational AI, like the models being developed by Anthropic and OpenAI, can be a game-changer. Imagine a customer service bot that doesn’t just follow a script but understands the customer's frustration, empathizes with them, and is empowered to solve their problem on the spot. That’s the future.
We did something similar at my last company, RemoteTeam. We used data to predict which employees were at risk of leaving. We looked at things like their communication patterns, their engagement with company initiatives, and even their promotion history. When the model flagged someone, we didn’t just sit back and wait for them to quit. We took action. Their manager would have a conversation with them, we’d look at their compensation, and we’d make sure they felt valued. It worked. We reduced employee churn by over 30%.
The Unfair Advantage
Banks that embrace this technology will have an almost unfair advantage. They’ll be able to hold onto their best customers, reduce their marketing costs (it’s far cheaper to keep a customer than to acquire a new one), and build deeper, more meaningful relationships.
Those that don’t? They’ll continue to bleed customers, wondering why their old methods aren’t working anymore. They’ll be stuck in the past, making decisions based on gut feelings and outdated reports, while their competitors are making surgical, data-driven moves.
I’m not saying AI is a silver bullet. You still need a great product and a culture that genuinely values its customers. But in a world where customers have more choices than ever, AI is the tool that can help you understand them, anticipate their needs, and earn their loyalty. My 22-year-old self would have been a lot happier—and $35 richer—if my bank had figured that out sooner.
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.