Inside Val’s Product Development: Building AI Features That Actually Work

In the rush to add AI features, most companies fall into a familiar trap: prioritizing speed over accuracy. But when Val CEO Andy Berman set out to build AI-powered meeting tools, he took a radically different approach.

In a recent episode of Category Visionaries, Andy revealed how Val’s experience with computer vision at Nanit shaped their approach to building reliable AI features that users actually trust.

The Foundation: Learning from Video Analytics

Val’s journey with AI didn’t start with meetings. “At Nanit, we very much focus on providing parents with analytics from video. So we track sleep, movement, breathing. We do it all from a video feed. We use computer vision and machine learning,” Andy explains.

This experience proved crucial. While many companies were rushing to add AI features as an afterthought, Val approached the problem with deep expertise in extracting meaningful insights from video.

The AI Accuracy Challenge

When Val launched their AI features two months before the interview, they faced a crowded market of competitors claiming similar capabilities. But Andy noticed a critical gap: “A lot of people actually talk about AI. They talk about a summary, they talk about instant notes from the meeting. But from what I’ve seen in terms of the products out there on the market, no one actually does it accurately.”

Instead of just adding AI features because they could, Val focused obsessively on accuracy. The goal wasn’t just to generate meeting summaries – it was to generate summaries people could actually trust and use.

Speed Meets Precision

Val’s approach combines two seemingly contradictory goals: speed and accuracy. “The meeting ends and seven to 10 seconds later you get the notes or the summary of what happened in the action items and they’re formatted and they’re ready to go. And maybe it takes you 30 seconds to clean it up, but it’s 85 95% accurate,” Andy notes.

This level of accuracy didn’t come from simply using existing models. As Andy explains, “We build a lot of custom AI ourselves, and we’ve just really focused on this problem from day one. So it’s not something and it’s not an add on.”

The Impact of Accurate AI

The results speak for themselves. Users describe Val as time travel because “you could log into after being out on vacation for a week and read the TLDRs from six or seven meetings and know exactly what happened.”

This reliability has driven remarkable growth: “Our user base has, I think, tripled over the last 40 days,” Andy shares.

Beyond Simple Summaries

Val’s AI doesn’t just summarize meetings – it makes meetings more actionable. After each meeting, “you automatically get a summary, the action items, the next steps from the video feed,” Andy explains. The platform also integrates these insights with existing workflow tools: “We’re rolling out our Zapier integration. We’re rolling out a whole host of other native integrations to the collaboration tools like Notion and Jira that you use.”

Looking to the Future

Val’s vision extends beyond just making meetings more efficient. Looking three to five years ahead, Andy sees AI becoming more proactive: “We’re pushing the information to you. So any given moment at any given time will tell you what’s relevant and give you in the moment the knowledge you need, whether it’s in the meeting or after the meeting, to be a superhuman.”

For B2B founders building AI features, Val’s journey offers several key lessons:

1. Start with a foundation of domain expertise
2. Focus on accuracy over feature quantity
3. Build custom solutions when necessary
4. Integrate insights into existing workflows

In a market flooded with AI features, Val shows that success comes not from being first to market, but from being the most reliable. As Andy puts it, “It compares to the magic that people have seen with Chat GPT” – but with the crucial difference that it actually works as promised.