The customer success manager at a growing SaaS company squints at her dashboard. Three hundred accounts, color-coded by some algorithm's definition of "healthy." Green mostly. A few yellows. She clicks through to the yellows, but nothing jumps out—usage looks fine, support tickets minimal. Two months later, one of those accounts doesn't renew. The exit interview reveals they'd already been evaluating competitors for six weeks.
This is the gap that Userlens, a Helsinki startup from Y Combinator's Spring 2026 batch, is trying to close. Not with better dashboards or more frequent check-ins, but with AI agents that essentially never sleep—watching every account, every day, for the subtle shifts that precede churn.
The company's pitch is straightforward, perhaps deceptively so: an "AI CSM that never misses churn." Co-founders Hai Ta and Ankur Dahama aren't first-time founders taking a flyer on a trendy problem. They previously built Wudpecker, an AI meeting assistant that reached something in the neighborhood of 25,000 users and over 1,000 customers. That experience, they say, taught them that customer success teams are structurally overwhelmed—too many accounts, too much data, and a reactive playbook that catches problems only after they've metastasized.
Userlens launched its AI agent product this past spring. The timing matters. After years of hype around AI capabilities, companies are now hunting for specific, measurable use cases. Customer success—long a cost center that rarely gets the tooling budget of sales or engineering—suddenly looks like fertile ground.
Always Watching, Rarely Sleeping
Here's how it works in practice. Each customer success manager gets a dedicated AI agent. Not a dashboard refresh or a weekly report, but something closer to a tireless junior analyst who monitors adoption patterns, feature usage, and engagement signals across the entire book of business.
When a power user goes dark or a core feature gets quietly abandoned, the agent pings the team in Slack. Sometimes weeks before a renewal comes up for discussion. Sometimes months, if the signal is subtle enough.
The platform scores account health daily—not just a traffic-light color code, but explanations in plain language about what's changed and why it might matter. It benchmarks accounts against similar customers, surfaces talking points for upcoming calls, and even drafts quarterly business review materials automatically. Setup reportedly takes about a day, and the system can ingest up to three months of historical data during initial pilots.
Userlens plugs into the usual suspects: product analytics platforms like PostHog, Amplitude, and Mixpanel; CRMs including HubSpot and Salesforce; support systems such as Intercom; and billing tools like Stripe. The connective tissue of B2B SaaS, essentially.
What the company emphasizes—what it believes sets it apart in a suddenly crowded field—is that the agents learn from corrections. When a CSM overrides an alert or adjusts a recommendation, the system doesn't just log the exception. It treats that feedback as a new skill, one that then propagates across the entire team automatically. "Correct it once, it remembers forever," the company claims on its product page.
Whether that learning loop actually works at scale, and whether it becomes genuinely smarter or just accumulates edge cases, remains to be seen. Early customers seem optimistic, at least according to Userlens's own case studies.
Early Adopters, Early Signals
Homie, one customer, reportedly monitors 100% of trial accounts without anyone manually clicking through them. They've built workflows around multiple specialized agents: one for health monitoring, another for onboarding support, a third for meeting prep, and a fourth dedicated to spotting upsell opportunities.
Luminovo's RevOps and customer success teams are using the AI health score alongside more than two years of historical data—a test of whether the system can make sense of messy legacy information. Quartr's account executives track trial engagement through the platform to build what they call "usage-backed business cases" during sales conversations.
These are small samples, the kind you'd expect from a startup still in its early phase. But they hint at something interesting: different teams are finding different uses for the same underlying system. Some treat it as an early warning system. Others use it for expansion plays. A few are experimenting with it as a sales tool rather than purely post-sale support.
The platform doesn't limit itself to product analytics, either. It reads support ticket sentiment and incorporates CRM signals—contract dates, account ownership, past interactions—into its risk assessments. The theory is that usage data alone misses crucial context: a frustrated user might still log in daily, and a happy customer might go quiet for legitimate reasons.
The agent also hunts for expansion signals. Power users bumping into feature ceilings. Teams outgrowing their seat counts. Accounts that have mastered features in ways that suggest they're ready for upsells. In an April blog post, the company tied these "proactive alerts" to net revenue retention strategy, arguing that the same infrastructure catching churn can identify growth opportunities. That makes sense, though whether CS teams have bandwidth to act on both simultaneously is another question.
The Agentic Era (Or At Least the Pitch for It)
Userlens is far from alone in chasing this opportunity. The customer success software market has spent years selling dashboards and health scores; now everyone's adding "AI agents" to their vocabulary.
Gainsight, the category incumbent, announced support for something called the Model Context Protocol in April, framing customer success as entering an "agentic era." ChurnZero has rolled out "AI & Agents" features and even launched an agent marketplace. Totango promotes AI-driven churn prediction. Cleeng, focused on media subscriptions, launched cross-platform AI agents to reduce churn in May.
The category is heating up, in other words, and Userlens is entering a race that's already underway. What they have going for them is founder credibility—Ta and Dahama know the problem firsthand—and the YC stamp, which still opens doors in SaaS circles. The company has self-reported that "MRR grew 27% on the first week of YC," though this claim lacks independent verification.
Userlens raised $761,000 in seed funding, reported in late March, though investor names haven't been disclosed. For context, the team previously raised €330,000 for Wudpecker back in October 2023, led by Trind Ventures with participation from Accelerace and Sofokus. The team remains lean—seven employees as of late spring, according to LinkedIn. That's small, even for an early-stage startup, and it raises the question of how quickly they can move against better-funded competitors.
The Human-in-the-Loop Question
Userlens is careful to emphasize that CSMs stay in control. The agent suggests; humans decide. It's a "human-in-the-loop" model, in the current parlance, where AI surfaces insights and drafts actions but doesn't fire off emails or cancel accounts on its own.
That's probably the right call, given how messy customer relationships can be. An algorithm might flag declining usage without knowing the customer's engineering team is on parental leave, or that they're in a quiet period before a big product launch. Context matters, and encoding all possible contexts into an AI system is—well, it's the kind of problem that keeps getting harder the more edge cases you encounter.
Still, there's something appealing about the idea of an agent that actually learns from feedback rather than requiring a data science team to retrain models. If it works as advertised, it could lower the barrier to adoption significantly. CSMs aren't going to fine-tune machine learning models, but they will correct a Slack alert that got it wrong.
Pricing starts at $500 per month for a Growth plan that monitors up to 50 companies with unlimited tracked users and events. Enterprise pricing is custom and includes single sign-on, Okta integration, and an assigned account manager. Y Combinator companies get three months free—a smart growth hack for a product trying to build a reference customer base in the YC network.
The company says it's working toward SOC 2 Type 2 certification and claims GDPR compliance, though no completion timeline for SOC 2 has been provided. That's standard for early-stage B2B startups; the certifications take time and money, and customers at this stage are often willing to accept "in progress" as an answer.
Betting on the Quiet Signals
For customer success teams drowning in data and losing sleep over silent accounts, Userlens is making a specific bet: that always-on AI agents, ones that learn from corrections and improve over time, can make churn prediction reliable enough to act on.
It's a bet that depends on several things going right. The agents need to be accurate enough that CS teams trust them. The learning loop needs to actually work, not just accumulate noise. And critically, the product needs to deliver value before larger competitors with deeper pockets ship similar features.
Whether Userlens can pull that off remains an open question. For now, it's a fresh take on a problem that has defied easy solutions for years—the gap between when a customer starts drifting and when anyone notices. If nothing else, it appears the company is putting AI to work on a problem that genuinely costs businesses money, rather than chasing hype for its own sake. That's something.
