The Moment I Knew I Had to Build This
It was 2023. I had just spent two years at NielsenIQ building consumer goods forecasting models that went from zero to $7 million ARR. Before that, six years leading AI at Deutsche Bank, where the team I built hit $10 million ARR in two years. By any measure, a solid corporate AI career.
And I kept watching the same thing happen over and over again: a brilliant piece of software—a library, a developer tool, an infrastructure product—would launch on GitHub, earn genuine love from its users, rack up stars, and then completely fail to convert any of that love into a business. The founders were spending their nights answering the same support questions. The people with real buying intent were leaving silently because nobody was there to catch them.
I'd seen this pattern from both sides. At the bank, I'd evaluated hundreds of vendor products and watched small teams lose deals not because their technology was worse, but because they couldn't respond fast enough. At NielsenIQ, I'd watched our own internal tooling get underutilised because nobody had time to answer the support burden. Revenue was being left on the table—consistently, predictably, and unnecessarily.
The gap wasn't a product gap. It was a presence gap.
What I Learned at Deutsche Bank About Enterprise Revenue
When I joined Deutsche Bank in 2019, I was tasked with building an AI team inside their Securities Services division. We were selling AI products internally and to institutional clients. The lesson I learned quickly was counterintuitive: enterprise buyers don't buy the best product. They buy the product they trust.
Trust, at that level, comes from responsiveness. From the ability to answer their specific question—their compliance question, their integration question, their security question—at the moment they ask it. Not two days later in an email. Not in a 30-minute scheduled demo three weeks out. Right then.
We built systems to help our team respond faster. We trained AI on our documentation. We automated the first-touch qualification. And it worked: conversations that would have stalled became deals. That is where the $10 million ARR came from—not from a better product alone, but from being genuinely present when prospects had questions.
I filed that lesson away. At the time, I didn't know I was accumulating the thesis for a company.
The Documentation Bot That Became a Business
Fast forward to 2023. I had left NielsenIQ, I was thinking about what to build, and I started with a simple observation from my own developer experience: I was always looking up the same documentation. Reading the same README sections. Asking the same setup questions in Slack communities.
So I built a documentation search engine—a bot that could answer questions about a library by reading its docs. Simple, personal, useful for me.
I shared it with a few friends who maintained open source libraries. Within days, they were all using it. Within weeks, they were reporting back that their Discord channels were calmer, their GitHub Issues had fewer “how do I” questions, and they were spending more time shipping features. Bereket Engida, who maintains Better Auth, told me his Discord activity had increased 10x since deploying the bot—because people were finally getting answers fast enough that they stuck around and kept asking.
That was the moment. I wasn't just solving a documentation problem. I was solving a presence problem. A revenue problem.
The Pattern Nobody Was Talking About
Once I started looking for it, I saw it everywhere. Across the open source communities I knew, the SaaS products I used, the developer tools I evaluated in my corporate roles—the same dynamic played out constantly.
A prospect arrives. They have a real question. The question is the first signal of intent. But the only options are: send an email (which feels like too much effort) or submit a contact form (which goes into a queue). So they close the tab. They didn't bounce because the product wasn't right. They bounced because the friction of engaging was higher than the benefit of staying.
According to Stack Overflow's 2024 Developer Survey, 75% of developers report answering the same questions repeatedly once they gain any traction. That is not a support problem. That is a scaling problem masquerading as a support problem. Every one of those repeated questions is a person who needed an answer and couldn't get one fast enough without taking the founder's personal time.
And the numbers behind the lost pipeline are staggering. 60% of revenue-bearing conversations happen outside business hours. 78% of buyers go with the first vendor to respond. Average B2B response time is 42 hours. These three statistics, taken together, describe a leaky bucket that every small team is pouring traffic into.
Why I Named It an AI Inbound Conversion Engine
I deliberated for a long time about how to describe what Clarm is. “AI chatbot” was too small—it implied a widget sitting in the corner answering FAQs. “Support automation” missed the revenue dimension entirely. “Lead capture” sounded like a form.
The frame that finally clicked was what I had seen at Deutsche Bank: a revenue desk. A function within a bank responsible for being present at the moment a client needed something revenue-relevant—a trade, a question about pricing, a risk inquiry. The desk exists so that no client signal goes unnoticed and no conversation dies waiting for a human to become available.
That is exactly what I wanted to give every technical founder: the equivalent of a senior revenue desk staffed 24/7, across every channel they operate on, that does not cost $150,000 a year and does not need managing.
An AI Inbound Conversion Engine. Clarm.
What the First Deployments Taught Me
The early results validated the thesis faster than I expected. GiveLegacy, a healthcare company, deployed Clarm on their website. In 90 days, their inbound conversations went from roughly 760 email inquiries to 4,624 AI-handled conversations—a 6.1x lift from the same traffic. 25.2% of those showed buying intent. 60% happened outside business hours. The channel went from $0 to their top inbound revenue source in 90 days.
Better Auth grew from 8,000 to 22,000 GitHub stars in three months after deploying Clarm. Not because the product changed. Because the community experience changed—people got answers fast enough to stay engaged, contribute, and tell others.
c/ua, an agent framework company, used Clarm's enrichment to identify which GitHub users were building production systems. That identification led directly to their first enterprise customer.
The pattern held across healthcare, open source, and SaaS. The gap was the same everywhere. The fix was the same.
What I Gave Up (And Why It Was Worth It)
I am sometimes asked whether I miss the stability of a senior corporate AI role. Honestly: not once.
When I was at Deutsche Bank, I was building systems that made a large institution slightly more efficient. When I was at NielsenIQ, I was improving forecasts for consumer goods companies. Both meaningful. Neither directionally what I wanted to spend my working life on.
Building Clarm, I am solving the problem I could not stop thinking about. I am talking to founders who are building things that matter—products that exist because someone cared enough to write the code, get the GitHub stars, and keep the community alive. And I am giving them a piece of infrastructure that means they do not have to choose between building and selling.
I believe the world needs more founders. Not fewer. The accelerating pace of AI means that more problems than ever are solvable. But most of the people trying to solve those problems will fail not because of bad ideas or bad engineering, but because they run out of runway before the revenue catches up with the product. Clarm is my attempt to change that ratio.
Where We Are Now
Clarm is a Y Combinator–backed company (YC X25). We are processing thousands of inbound conversations every week across website chat, Discord, Slack, and GitHub. We are HIPAA and SOC 2 compliant. We support on-premise deployment for regulated industries. We have a team of people who are as serious about this problem as I am.
And I still answer every email that comes to [email protected]personally. Because the irony of running a company whose whole point is that founders should not have to answer every question manually is not lost on me—and because talking directly to the people we serve is still the best way to understand what we need to build next.
If you are a founder sitting on traffic that is not converting, or a team drowning in the same support questions, or an engineering team that is losing enterprise deals because response time kills conversations overnight— I would genuinely like to talk to you. Book 20 minutes with me. No sales script. Just a conversation about the problem.
Further Reading
If this resonated, you might also want to read about the hidden pipeline problem every technical founder ignores or how founder-led sales breaks at scale and what to build instead.