Richard White, Fathomin perustaja ja toimitusjohtaja – Haastattelusarja

Richard White, Founder & CEO of Fathom, is a repeat founder and product-focused entrepreneur best known for turning personal frustrations into category-defining software. Prior to Fathom, he founded and led UserVoice for nearly 13 years, growing it into a profitable feedback management platform used by thousands of companies ranging from startups to enterprises like Microsoft, while also pioneering the now-ubiquitous website “Feedback” tab. Earlier in his career, he built and ran SlimTimer entirely solo for over a decade, led influential open-source projects such as AjaxScaffold in the Ruby on Rails ecosystem, and worked as Product Design Lead at Kiko (YC S05), experiences that collectively shaped his philosophy around usability, customer empathy, and building tools that quietly but meaningfully improve how teams work.

Perustettu 2020, syli reflects that same ethos by tackling a universal pain point: the cognitive overload of taking notes while trying to have real conversations. The platform automatically records, transcribes, and summarizes meetings—most notably on Zoom—allowing users to highlight moments in real time, share short clips instead of raw notes, and preserve the nuance that’s often lost in written summaries. As Fathom has matured, it has evolved beyond simple transcription into a lightweight system of record for conversations, designed to help teams retain context, learn from customer calls, and collaborate asynchronously without adding friction to the meeting itself.

You’ve spent the past 15 years building companies that reshape how people communicate — from UserVoice to Fathom. What was the moment that pushed you to found Fathom, and how did your engineering and product-design roots shape the company from day one?

My inspiration for founding Fathom came in early 2020. It was pre-pandemic, but I was doing extensive user research for a product and suddenly was sitting through 15 or 20 back-to-back Zoom meetings a day. Six weeks of that made me keenly aware of how painful the experience was. I can’t talk and type at the same time — I’d look at my notes two weeks later and not remember which conversation was which. The biggest problem was that I’d do all this research and then share a few bullet points with my team and it just wouldn’t land. Everything got lost in translation. It was a ‘stub your toe’ moment for me: Something that, if it happens once a month, you ignore. You stub your toe on something every day, multiple times a day, you very quickly try to fix it.

My engineering and design background both informed the choices I made while building Fathom. I’ve always approached problems by taking concepts that already exist and making them radically more usable for a much larger audience. With Fathom, I had this insight that transcription technology was becoming commoditized — there was a proliferation of off-the-shelf solutions that didn’t exist five years earlier. So, transcription was part of the solution, but it wasn’t the solution itself.

From a product design perspective, I realized that transcripts can be valuable for the people who were on the call. But they’re really not helpful to people who weren’t there. What we found far more impactful was showing you the 30-second video clip of the customer objecting on price or asking that technical question. We use the transcript almost like a table of contents to find the actual audio-video clip. That product thinking — understanding the jobs to be done, not just the technology — came directly from my design roots.

Fathom was created in 2020, well before most companies were thinking seriously about AI-native workflows. What advantages did building with AI at the core — rather than retrofitting it — give you early on?

The key advantage was architectural freedom. We could design every system, from data pipelines to user experience, assuming AI would be a fundamental layer and not a bolt-on feature. Most competitors in 2020 and 2021 were hiring linguistics experts and ML specialists to build their own models. We took the opposite tack because we believed the winners in the space would be those who could apply AI effectively to solve real problems, not those who built the models themselves. That contrarian view let us stay nimble with a smaller team and to focus our engineering resources on the hard infrastructure problems — reliable recording across platforms, viral distribution mechanics, real-time processing at scale.

Here’s the thing about starting in 2020: AI wasn’t good enough yet. We knew that. But we also knew that if we waited for AI to mature before building the company, we’d be two to three years too late. The door would be wide open, and everyone would flood in. So we built everything else first — the infrastructure, the distribution channels, the user experience — with the explicit expectation that when AI got there, we’d drop it in like a new engine in a car. That decision paid off massively. When GPT-4 and Claude arrived in 2022-2023, we could immediately integrate them. Competitors who’d spent years building custom NLP pipelines suddenly had to rethink their entire stack. We just upgraded our models and kept shipping.

Building AI-native also fundamentally changed our product development process. Traditional software has a pretty linear roadmap: You decide what to build, you build it and you ship it. With AI, we use what I call a “Jenga model.” Each block represents a potential AI capability. If we push on a block and get resistance because the models aren’t good enough yet, we try a different one. We know that in six months, the technology will improve and we can come back to it. This keeps us from forcing features before they’re ready, while ensuring we’re always shipping value.

The other advantage was credibility. Yes, investors told me not to put “AI” in our name in 2020, but being early gave us authenticity. We weren’t jumping on a trend; we were betting on a thesis before it became obvious. That positioned us as builders, not fast followers.

You’ve described meeting conversations as one of the most overlooked data sources inside organizations. What convinced you that this was the next major frontier for AI?

I realized that I’d never met a salesperson who’s got eight hours a day to listen to all their team’s meetings, let alone make decisions and coach their team based on what they’ve heard. Meetings generate incredibly valuable data, but it’s completely inaccessible at scale. With traditional meetings we throw away 99% of the content, while the last 1% of notes go into the CRM. Then we try to reverse extrapolate from there what’s going to happen with our business. It’s an absurd process. The information that actually matters — the tone of a customer’s voice, the specific objection they raised, the competitive mention that came up — all gets filtered through someone’s hastily typed notes and loses all context.

What convinced me this was the next frontier was recognizing that this “conversational dark data” is actually the richest signal of what’s happening in an organization. You’re getting real-time insight into customer pain points, product gaps, competitive threats and training needs — all in people’s own words. When a customer explains why they need a feature, that’s way more valuable than a sales rep’s paraphrase in a CRM field.

The breakthrough with AI is that we can finally harness this data at scale. When we first launched Ask Fathom, it could answer questions about individual meetings. Then we enhanced it to handle small groups of meetings. Now it’s smart enough to understand your entire company’s set of meetings. Sales leaders can ask, “What competitors are trending up most recently? Show me some clips.” Engineering teams can query, “Tell us the history of transcription engines at Fathom” and get a six-page synthesized document pulling from four years of engineering meetings.

It’s beginning to be a much bigger brain that really understands what your business is doing and the conversations it’s having. You can imagine a world soon where an AI can tell you what features you should build next based on what would help close the most deals, or what competitors are coming up, or what training gaps exist across your team. There’s this amazing data source that AI is mining to give you inputs into your next strategy meeting or roadmapping process.

Many users cite Fathom as transformative for staying present during meetings. How do you balance automation with preserving the natural flow of human conversation?

This has been core to our design philosophy from the beginning. The goal isn’t to have AI tell you what to do in a meeting, but rather to give you insights that help you be more present and effective in your conversations.

We’re careful about what we automate and what we don’t. We won’t launch features until we know we can do them really well. This sometimes means we’re not first to market with certain capabilities, but when we do launch something, it works and delivers genuine value. We’ve been cautious about pursuing things like phone call recording or certain in-room meeting capture despite frequent requests. We’d rather excel at what we do than roll out a mediocre experience that disrupts the natural flow of conversation.

Ultimately, our users tell us we’re striking the right balance: They say they’re saving 6+ hours per week and moving 3× faster from insight to next steps; 95% report that Fathom keeps them present in meetings. This affirms that we’re augmenting human capability, not replacing it.

Fathom attracted more than 1,300 user-investors in its Series A — a rare sign of product-level trust. What do you think resonated so strongly with everyday users?

For one thing, we give away a genuinely robust free product: unlimited meetings, five AI summaries per month. Two-thirds of our users never pay us a dime, and we’re completely fine with that. It’s not a typical SaaS play. Our users see that we’re not trying to extract value from them at every turn. We’re focused on making individual contributors’ lives better for free, and we monetize by selling management tools to their bosses — coaching dashboards, cross-meeting intelligence and competitive insights. The product just works, and it keeps working whether you pay or not. That creates genuine trust.

Our growth is almost entirely word-of-mouth — we’ve grown more like a social media platform than traditional B2B software. Our users are our advocates and distribution channel. Letting them become investors just acknowledges what’s already true: They’re partners in this mission.

I also think there’s a deeper resonance around the problem we’re solving. Everyone has experienced the pain of being in a meeting, trying to be present and watching someone frantically type instead of engaging. Everyone has needed information from a meeting they weren’t in and gotten a useless two-line summary. The problem is universal, and the solution feels almost magical when it works well. Users invest because they want this future to exist — not just for themselves, but for everyone they work with.

Your background includes building UserVoice, which helped define how companies manage customer feedback. How did that experience influence your thinking about organizational memory and AI-powered knowledge flows?

UserVoice taught me that the most valuable information in companies is often the most scattered. Customer feedback was everywhere. It was buried in support tickets, forwarded emails and random sales conversations. Companies would have thousands of data points about what customers wanted, but no way to synthesize it into strategic decisions. We built infrastructure to aggregate that feedback at scale and make it accessible to the people making product decisions.

The parallel with Fathom is clear, but the problem space is more profound. Meetings are exponentially more scattered than customer feedback. Every organization has hundreds or thousands of hours of conversations happening every week. What I learned from UserVoice is that capture is necessary, but it’s not enough. You can’t just aggregate information; you need to build intelligence about what matters and route it to the right people. With UserVoice, we built voting systems, trending algorithms and admin dashboards so product teams could separate signal from noise. With Fathom, we’re building AI that understands context across conversations and can proactively surface insights: “Five customers mentioned this use case this month,” or “Your team keeps getting stuck on this objection.”

The other lesson was about democratization. UserVoice made it possible for any customer to provide feedback, not just the loudest ones who could get executives on the phone. With Fathom, we’re democratizing access to meeting intelligence. In our case study with Netgain, their operations manager was spending 7.5 hours a day just answering basic questions about what was happening in sales calls. That’s insane. The information existed, but it was trapped in people’s heads and scattered notes.

The future of organizational memory is moving from these isolated knowledge silos — CRM, docs, feedback systems — to connected, conversational intelligence. That’s the logical evolution of what we started building with UserVoice, but AI makes it possible to do it with the full fidelity of human conversation, not just structured data.

Zoom-based AI tools exploded after 2020. In your view, what differentiates a truly helpful AI assistant from one that just adds noise?

I always tell people there are only two things that can really sink an AI meeting assistant: if the product isn’t reliable, or if the AI output is garbage. I think there was a lot of marketing AI in the previous generation where it was easy to promise magical stuff, but then the reality came out as gibberish. We’ve always tried to make sure we have a high-quality, reliable product that does what it promises. Our key differentiators are:

• Transcription accuracy. Fathom is considered the most accurate transcript out there today. Most tools leverage a third-party transcription service, whereas we built our own proprietary transcription technology in-house. If your transcript’s bad, everything from an AI component is absolutely trashed because it all comes off the transcript.
• Reliability and infrastructure. When you join a meeting, you’re often in a hurry or stressed. A lot of these other tools would have bots join meetings but then wouldn’t record, or the recording would fail. We exist almost at a real-time system level — you’re working on something that’s one step behind avionics. If it doesn’t work twice, the user’s gone. It’s not like traditional SaaS where you can be down occasionally.
• AI that understands nuance and context. Business language can be very subtle. I remember running the sales team at UserVoice and reading people’s notes, thinking, “I need to hear how they actually said this.” The AI needs to capture not just what was said, but the tone, the hesitation and the excitement (or lack thereof). That’s why we link every summary point back to the actual moment in the recording.
• Customization without complexity. The AI should adapt to your business, not the other way around. Sales teams should be able to modify templates to match their specific methodologies — MEDDIC, Challenger, SPICED, whatever they use. But this can’t require a data science degree. It needs to just work.

Fathom turns meeting content into actionable knowledge. How close are we to AI systems that function as real workflow engines — connecting conversation, decisions, and downstream tasks automatically?

I think we’re closer than most people realize, but there are still important steps to take. Right now, we’re moving into a world where Fathom does more and more of the work for you. The first step is just getting the information where you want it to go. The next step, which isn’t far away, is having the AI actually do the work for you.

We’re already seeing early versions of this. Our Asana integration takes action items from meetings and automatically creates trackable tasks. Fathom does not want to come up with a task management solution — there are lots of great ones out there, like Asana. So we’re building integrations that push meeting outcomes directly into the tools people already use for getting work done.

On the CRM side, we automatically push structured fields — pain points, timelines, key decision-makers — into Salesforce and HubSpot. In one case study, this saved 20 to 30 minutes per deal status update and led to near-perfect month-end forecast accuracy. That’s a workflow engine in action: Conversation happens, AI extracts the key business data and then it flows automatically into your system of record without anyone typing anything.

But I think the real breakthrough is coming with what I call semantic-based alerts and intelligent routing. Imagine being a manager or sales leader and getting a daily highlight reel where the AI has found every pricing discussion that went sideways, or every product blocker that came up in a renewal call. If you’re an engineering manager, you’d see every heated debate among your engineers. The AI can understand tone and nuance now, not just keywords, so it knows what moments you actually care about.

As companies scale, they struggle with distributed knowledge and information decay. How do you see AI addressing the gap between what teams discuss and what actually gets executed?

This is one of the most critical problems we’re solving. There are two groups we can really help: people in the meeting trying to take notes and be present, and the management, leadership and founders who aren’t in the meeting but are running teams and trying to understand what’s happening. That second group is where the distributed knowledge problem really hits.

The core issue is visibility. When anyone in a company wants to know the status of a deal or what’s happening with a customer, traditionally there’s no place to easily find that information. They call the sales team, forcing reps to spend 20-30 minutes digging through notes. During peak periods, some operations managers get 15 requests daily — that’s 7.5 hours spent on information retrieval instead of value-adding activities.

AI can start connecting dots across conversations that no human could track. That kind of pattern recognition across distributed conversations is how you prevent knowledge decay and actually turn conversations into strategic intelligence.

Looking ahead five years, how do you anticipate meeting intelligence evolving — and what role do you see AI playing in the future of organizational memory, decision-making, and collaboration?

Five years from now, I think we’ll look back at today’s meeting intelligence tools the same way we now look at early smartphones: impressive for their time, but primitive compared to what became possible.

The first major evolution is moving from note-taking to true workflow automation. We envision a future where just saying something in a meeting can will it into existence, without the post-meeting work. Right now, if you say in a meeting, “Let’s create a spec for this feature and schedule a follow-up with engineering next week,” you still have to manually create that doc and send that calendar invite. In five years, the AI will do all of that automatically. You speak it, and it happens. With AI creating the tasks, specs and docs, people can focus on the work that actually requires human creativity and judgment.

The second evolution is expanding from customer-facing to all meetings. Right now we focus on external meetings: sales, customer success, agencies meeting with clients. But our goal over the next 12 to 18 months is to make Fathom the platform you can use across your entire organization, not just customer-facing teams. We’re building botless recording that can capture any conversation, including Slack huddles and in-person meetings. It’s evolving into being able to capture any conversation you’re having at your company, no matter what the medium is.

The companies that rise to the top will be those that treat conversational data as a first-class citizen — as important as their CRM data, analytics and documents. Because ultimately, the most important knowledge in any organization isn’t in the systems; it’s in the conversations. AI is finally making it possible to harness that.

Thank you for the great interview, readers who wish to learn more about this note taking app should visit syli.