Building Findly: From Seed of an Idea to Y Combinator Demo Day

The idea for Findly didn't come in a flash of inspiration. It emerged slowly, through years of frustration watching brilliant people make crucial business decisions based on gut feeling rather than data—not because they didn't want to use data, but because accessing and understanding it was simply too complicated.

This is the story of how that frustration became Findly, and how we went from a napkin sketch to Y Combinator Demo Day in 18 months.

The Problem We Couldn't Ignore

During my time at Google and Twitter, I had access to incredible analytics infrastructure. Want to know how a feature performed across different user segments? Query BigQuery. Need to understand engagement patterns? Build a dashboard. Curious about anomalies in your metrics? Write some SQL and dig in.

But this luxury was only available because I worked at companies with massive data teams and sophisticated tooling. Most businesses weren't so fortunate.

I watched small business owners make inventory decisions without understanding seasonal patterns in their sales data. I saw marketing teams run campaigns based on hunches rather than customer behavior analysis. Product managers prioritized features without really knowing how users interacted with their existing product.

The gap wasn't just about tools—it was about accessibility. Business intelligence had become the domain of specialists, requiring knowledge of SQL, understanding of data warehouses, and familiarity with complex dashboards. For most people, asking a question about their business data was like trying to perform surgery: theoretically possible, but practically out of reach.

The question that kept bothering me: What if asking questions about your data was as simple as asking questions in a conversation?

From Idea to MVP

In early 2022, I started experimenting with GPT-3's API, exploring how language models could bridge the gap between natural language questions and structured data queries. The early prototypes were rough—imagine a chatbot that could sometimes generate correct SQL queries and sometimes hallucinate completely fictional database schemas.

But even those broken early versions revealed something important: people intuitively understood how to ask questions conversationally, even about complex business problems. "Which customers are most likely to churn?" "What's driving the spike in support tickets?" "How do our new features impact retention?"

The technical challenge was clear: translate natural language intent into precise data operations, then translate the results back into insights that drive decisions. Not just showing charts, but understanding context, asking clarifying questions, and providing actionable recommendations.

I brought the idea to my co-founder Jonathan, and we spent countless evenings in London coffee shops (and later, over Zoom during lockdown) debating the architecture. How do you handle ambiguous queries? What happens when the AI misunderstands a question? How do you build trust in automated insights?

Our MVP was embarrassingly simple: a text interface that could answer basic questions about CSV files. Upload your sales data, ask "What were my best performing products last quarter?" and get a chart with some bullet-point insights. No dashboards, no configuration, no SQL.

We tested it with ten small business owners in London. The feedback was immediate and clear: "This is exactly what I need, but it needs to work with my actual data systems, not just CSV files."

The Y Combinator Journey

Applying to Y Combinator felt like a long shot. We had a working prototype, some encouraging user feedback, and a conviction that conversational AI would transform business intelligence. But we were competing against thousands of teams with more traction, more technical depth, or more obvious market opportunities.

Our application video was recorded in my tiny London flat, with Jonathan explaining our vision while I demoed the product. We kept it simple: show the problem, demonstrate the solution, explain why now was the right time for conversational BI.

The interview was intense. Partners grilled us on technical feasibility ("How do you handle hallucinations in financial data?"), market size ("Is this just a feature that existing BI tools will build?"), and team dynamics ("Why should we bet on you two specifically?").

The question that caught me off guard: "What happens when OpenAI or Google builds this?"

My answer: "The same thing that happened when Google built search ads—the market expanded. We're not just building a better way to query databases; we're creating a new category of business intelligence that makes data analysis accessible to millions of people who can't use traditional BI tools."

Getting the acceptance email was surreal. Suddenly, we were part of the Summer 2022 batch, surrounded by incredible founders working on everything from biotech to crypto to space technology.

What We Learned

Building Findly taught us lessons that no amount of Google or Twitter experience could provide:

Start with the customer problem, not the technology. Our early technical prototypes were impressive, but they solved problems we found interesting rather than problems customers would pay to solve. The magic happened when we started with user pain points and worked backward to the technical solution.

Perfect is the enemy of shipped. Coming from big tech, I was used to extensive testing, code reviews, and gradual rollouts. Startups require a different mindset: ship fast, learn quickly, iterate constantly. Our MVP was objectively terrible by Google standards, but it was exactly what we needed to validate the core hypothesis.

Focus is a superpower. Y Combinator forces you to pick one target customer, one core use case, one key metric. This constraint felt limiting at first, but it enabled rapid progress. When you're trying to serve everyone, you serve no one particularly well.

Distribution matters more than features. We spent months perfecting our natural language processing capabilities, but the real breakthrough came when we figured out how to integrate directly into tools people already used. The best technology is invisible.

The Road Ahead

Today, Findly serves customers across five continents, processing millions of queries per month and helping teams make better decisions faster. We've raised funding, expanded the team, and built partnerships with major data infrastructure providers.

But we're still working on the same fundamental problem: making data analysis as natural as having a conversation. The technology has evolved dramatically—our AI can now handle complex multi-step analyses, generate predictive insights, and even suggest follow-up questions you should ask.

The mission remains unchanged: democratize business intelligence. Not everyone needs to become a data scientist, but everyone deserves access to insights from their data.

The journey from idea to Y Combinator taught us that building a startup isn't just about solving technical problems—it's about understanding human problems and creating solutions that feel magical, even when the underlying technology is complex.

Most importantly, it reinforced our belief that the future of business intelligence isn't about better dashboards or faster queries. It's about making data analysis so intuitive that insights become a natural part of how teams work, not a specialized skill that only experts possess.

That's the future we're building at Findly, one conversation at a time.

Want to try Findly? Visit findly.ai or reach out to me directly at pedromnasc@gmail.com.