Why B2B Brands Are Using AI for Customer Health Scoring

Lucas: So we talk a lot on this show about using AI to find new customers, but there's a flip side that I think is just as important: keeping the customers you already have. And specifically, using AI to score customer health in real time. Luna: Customer health scoring - I've seen that term pop up more and more. But what does it actually look like in practice?

Are we talking about a simple red yellow green dashboard? Lucas: That's exactly what it used to be. Most companies had a manual health score based on things like login frequency or support ticket volume. The problem is, by the time those metrics turn red, the customer is already halfway out the door.

Lucas: The new generation of AI tools - companies like Gong, ZoomInfo, Totango - they're building models that ingest not just product usage data, but also the language from sales calls, support chats, emails, even Slack messages. They look for signals that predict churn weeks or months in advance. Luna: So it's not just 'did they log in today?' It's 'how did they talk about the product on the call with their account manager?'

Lucas: Exactly. Gong's model, for example, analyzes the sentiment of every recorded sales conversation. If a customer starts using phrases like 'we're not getting value' or 'we're evaluating other options,' the system flags that account as a risk, even if usage numbers look fine. Luna: That's powerful - but also a little creepy.

Are customers okay with their calls being analyzed for churn signals? I mean, they know the call is recorded, but do they know it's being run through a churn prediction model? Lucas: It's a fair question. Most companies disclose that calls are recorded for 'quality and training purposes,' and they typically include a clause about analytics in the terms of service.

But I think the bigger issue is trust. If a customer feels surveilled, that could actually accelerate churn. Lucas: The best approach I've seen is transparency: tell the customer you're using AI to improve their experience, and then use the insights to proactively solve problems. If you detect frustration early, you can reach out with a solution instead of a renewal scare.

Luna: Right, so it's less about spying and more about being proactive. Give me an example of a company that's doing this well. Lucas: ZoomInfo recently rolled out a feature called 'Churn Risk Score' that combines data from their own platform - intent data, firmographics - with the customer's product usage. If a key champion at the account hasn't logged in for two weeks, and the company's website traffic drops, the score goes up.

Lucas: Then the system automatically triggers a workflow: the customer success manager gets a notification, a personalized email goes out to the champion, and maybe even a discount offer is generated if the risk is high enough. Luna: So it's not just a score - it's an action trigger. That's the key difference from the old red yellow green dashboards, which just sat there. Lucas: Exactly.

The AI doesn't just tell you someone is at risk; it suggests the next best action. Totango has a model that even predicts the likelihood of a customer expanding their contract. So you can focus your best reps on the accounts most likely to grow. Luna: That's smart.

It's like lead scoring, but for existing customers. I can see why this is becoming a standard part of revenue operations. Lucas: And the numbers back it up. A 2025 study from Gainsight found that companies using ai driven health scoring reduced churn by an average of 25% within six months.

For a B2B SaaS company with a million-dollar annual contract value, that's huge. Luna: Twenty-five percent - that's not incremental. That's a game changer for any subscription business. But I wonder about false positives.

If the model flags an account because a champion had a bad week, you could end up wasting resources on accounts that are actually fine. Lucas: That's a real risk. The best models use multiple signals, not just one. If a champion misses a login but the rest of the team is active, the score stays low.

It's when you see a pattern across several signals - declining sentiment, dropping usage, support tickets spiking - that the confidence goes up. Lucas: And you can set thresholds. A 70% risk score might trigger an automated email, but an 85% score triggers a personal call from the VP of Customer Success. You don't want to cry wolf too often.

Luna: Makes sense. So what's the biggest challenge in implementing this? Is it the data, the model, or the culture? Lucas: I'd say culture is the hardest part.

You need sales, customer success, and product teams to actually act on the insights. If the AI says 'this account is at risk' and nobody follows up, you've wasted the investment. Lucas: There's also the data integration challenge. You need to pull in data from the CRM, the product analytics tool, the support platform, and the call recording system.

That's a lot of pipes to connect. Luna: Right. And for smaller teams without a dedicated data engineer, that could be a barrier. Are there out-of-the-box solutions that make it easier?

Lucas: A few. Totango and ChurnZero both offer pre-built integrations with Salesforce, HubSpot, and Zendesk. But you still need clean data. Garbage in, garbage out.

Luna: I've been thinking about something you said earlier - about using language from sales calls. How do you handle privacy? Especially in Europe with GDPR. Lucas: It's tricky.

Most companies anonymize the call data before feeding it into the model. They extract sentiment and keywords without tying it to a specific person. And they make sure the model doesn't store raw audio. But you're right to be cautious.

Lucas: Some companies are even building opt-in models where customers can choose to share their call data in exchange for a better experience - like proactive support or personalized recommendations. Luna: That seems like the right balance. Give people a choice. And honestly, if it means I get a heads-up before my subscription lapses, I might opt in.

Lucas: Exactly. And that's the vision - not just reducing churn, but improving the customer experience. If the AI helps you solve a problem before the customer even knows they have it, that's a win-win. Luna: So where do you see this going in the next year or two?

I'm thinking about July 2026 right now - what's the next frontier? Lucas: I think we'll see more predictive models that go beyond churn to predict revenue expansion. Imagine a score that tells you not just which accounts are at risk, but which ones are ready to buy more, and what product they'd likely add. Lucas: Some companies are already experimenting with that.

They look at product usage patterns - like which features a customer uses most - and then recommend upgrades or add-ons automatically. It's like Amazon's 'customers who bought this also bought' but for B2B. Luna: That would be incredibly valuable for account managers. Instead of guessing which upsell pitch to make, they'd have data-backed recommendations.

Lucas: And it all ties back to the same principle: use AI to make the customer's life better, not just to extract more revenue. If you solve their problems, the revenue follows. Luna: I love that framing. It's customer-centric, but it's also smart business.

Before we wrap, I want to touch on something. These conversations we have - they take a lot of research and prep. And we keep them ad-free because we believe that's the right way to do it. If you get value from the show, you can support that choice at buy me a coffee dot com slash fexingo.

Lucas: Yeah, it's a small way to keep this going without selling your attention. Appreciate anyone who chips in. Luna: Alright, back to the topic. One thing I'm curious about: how do you measure the ROI of a customer health scoring system?

Is it just churn reduction, or are there other metrics? Lucas: Churn reduction is the big one, but you can also look at net revenue retention. If the model helps you identify expansion opportunities, that shows up in higher upsell rates. Some companies also track time to resolution for support tickets - if the AI proactively flags an issue, it might get solved faster.

Lucas: But the simplest metric is: are you saving accounts you would have lost? If you can point to a specific customer that would have churned but didn't because the AI alerted you, that's a win. Luna: That's a story you can tell the CFO. 'We saved a million-dollar account because the model caught a signal three weeks early.'

Lucas: Exactly. And that's the kind of concrete impact that makes this more than just a nice to have. It's becoming a core part of how B2B companies think about revenue. Luna: Great episode.

Thanks, Lucas. Lucas: Thanks, Luna. Talk next time.