Vision AI Rewrites Product Analytics Playbook

At 6 a.m., tired but focused, 20-year-old Amogh Chaturvedi described how his team went from a bootstrapped e-commerce accounting tool to a fresh startup riding vision AI. Human Behavior, launched a few months ago by Chaturvedi and co-founders Skyler Ji and Chirag Kawediya, closed a $5 million seed in two days after participation in Y Combinator.

Their bet is simple and bold: use computer vision to watch session replays and answer the questions traditional analytics struggles with — not just what users do, but why they do it. Instead of instrumenting every click and button with events, Human Behavior’s models parse video to summarize feature use, surface bugs, and identify churn signals.

How it works

Human Behavior processes thousands of hours of session footage with modern vision models that are finally accurate enough to parse UI interactions at scale. The output is human-readable insights: which features drove conversions, where users encountered friction, and which patterns precede churn — delivered in daily summaries rather than raw clickstreams.

Why product teams care

Engineers at fast-moving startups often spend days or weeks wiring event trackers. Even with data, teams still struggle to understand context and user intent. Human Behavior pitches a shortcut: stop writing instrumentation for every button and let models watch the session replays instead.

• Reduce engineering time spent on event instrumentation.
• Surface behavioral context — not just clicks but intent and frustration.
• Automate QA and embed support from the same replay dataset over time.
• Deliver daily, actionable summaries that product and ops teams can act on immediately.

The founders have tangible chops: their prior startup, Dough, was sold for six figures and the team’s YC admission assumed a pivot — which they did after customer conversations revealed a consistent ask: explain why customers buy or churn. That shift led them to build Human Behavior and attract investors including General Catalyst and Y Combinator.

They’re also positioning themselves against established analytics players like Mixpanel and PostHog. The edge, they claim, is architectural: products built for event-based analytics may find it hard to pivot to video-first replay analysis without major reengineering.

There are real challenges ahead: privacy, storage and compute costs, and ensuring model accuracy across different UIs. But the payoff is significant — a new layer of behavioral understanding that could power feature prioritization, faster bug triage, and even automated support.

For fast-moving Series A and B startups that value speed over elaborate instrumentation, Human Behavior promises a different route to insight. Their long-term vision is expansive: become the Datadog of session replay and spin out products from the same core dataset.

As vision models keep improving, watching the video may be the most direct way to understand users. For teams deciding whether to rewire event pipelines or trial replay-first analytics, the Human Behavior story is a reminder: sometimes the simplest data source — the user’s session — still holds untapped value.