We didn’t raise a Series A. We didn’t have a growth team or a head of marketing or a sales org. For most of Speak AI’s history, it was a small group of people building something they believed in and finding customers one at a time, then figuring out how to find them faster.
250,000 users across 100+ countries didn’t happen because of a viral moment or a well-timed launch. It happened because of a few things done consistently over a long time, some decisions that worked out better than expected, and a product built close enough to real user needs that people kept coming back and telling others.
This is what I actually remember about how we got there.
What we saw and why we built it
The original insight behind Speak AI wasn’t about transcription specifically. It was about a broader problem: most of the data that organizations actually care about is spoken. What their customers say, what their researchers hear in interviews, what their teams discuss in meetings. And spoken data is almost entirely wasted. It gets transcribed manually, if at all. It sits in recordings nobody revisits. The insight lives in someone’s memory until it doesn’t.
The people who felt this most acutely were researchers. Qualitative researchers running interviews. UX teams trying to understand users. Market researchers conducting studies. They were spending enormous amounts of time on transcription and manual theme extraction, and the tooling available to them was either generic transcription software or expensive enterprise solutions built for a different workflow entirely.
That’s where we started. A product built specifically for researchers who needed to go from recorded interview to structured insight faster than they could do it manually.
The first 10,000 users
The first real growth came from content. Not social media content. SEO content. Long-form articles about the problems our users were trying to solve: how to analyze qualitative research, how to transcribe interviews efficiently, what the best tools were for different research methodologies. We wrote those articles because we understood the space and had things to say. The traffic that came from them converted well because it was highly targeted.
That’s a pattern worth understanding. The content that works best isn’t content about your product. It’s content about your user’s problem, written by someone who genuinely understands that problem. We weren’t writing to rank. We were writing to be useful to a specific type of person, and ranking was the outcome of that.
The second driver was product-led growth. Speak AI had a free tier that let people upload recordings and get transcripts. Many users started there, found the experience useful, and converted to paid plans when they needed more volume or more features. This wasn’t a sophisticated PLG motion at the beginning. It was just making the product accessible enough that people could try it without a sales call. The conversion happened because the product worked.
What +123% gross volume growth actually means
When I say we hit +123% gross volume growth, that number is real but context matters. Gross volume here refers to the total audio and video processed through the platform. The raw measure of how much work Speak AI is doing for its users. Growing that number by 123% in a year means more than doubling the amount of data flowing through the system.
What drove it wasn’t a single thing. It was a product that had gotten meaningfully better, more accurate transcription, better AI analysis, more integrations, meeting a market that was growing fast because AI adoption was accelerating across the board. Researchers who previously thought automated transcription wasn’t good enough were finding it had crossed the threshold. Teams that hadn’t considered speech analysis as part of their workflow were starting to.
We were in the right place. We’d been building toward it for years before it looked like the right place.
The mistakes
We built features too early that not enough users wanted. That’s the standard startup mistake and we made it. Spent engineering time on capabilities that seemed like logical extensions of the product but weren’t things users were pulling us toward. The clearest signal to build something is users asking for it repeatedly. We occasionally ignored that in favor of what seemed strategically interesting.
We underinvested in customer success for too long. The users who churned usually churned because they didn’t fully understand what the product could do. A better onboarding experience and more proactive outreach to new users would have improved retention meaningfully earlier than it did. We knew this. We didn’t act on it fast enough.
We also spent more time than we should have trying to serve every potential customer rather than getting exceptionally good at serving a specific one. The researchers were our best customers. High intent, clear need, willing to pay for quality, likely to tell colleagues. Leaning harder into that segment earlier would have accelerated everything.
What I’d do differently
Start narrower. The temptation when you’re building something with broad applicability is to try to capture all of it. The reality is that depth beats breadth in the early stages every time. Becoming the obvious choice for one specific type of user is worth more than being a reasonable option for many.
Invest in content earlier and more systematically. The content we built compounded. Articles written in 2021 are still driving traffic and signups. That’s the nature of SEO. It’s slow to start and hard to stop. Every month you’re not building it is a month of compounding you’re not getting.
Find the users who are already doing the thing you’re enabling manually and talk to them constantly. They will tell you everything you need to know about what to build. We did this reasonably well but not well enough. The times we were closest to our users were the times we made the best product decisions.
The Y Combinator chapter
We applied to Y Combinator for the Winter 2023 batch. I wrote up every question in the application and published it, because I think the habit of founders keeping their actual thinking hidden does more harm than good to the ecosystem.
We didn’t get in. What we got was the discipline of articulating every assumption we were making about the business in writing, under constraints, for an audience that would be skeptical of anything that sounded like wishful thinking. That process is valuable regardless of the outcome. If you’re building something and you haven’t written down, as precisely as you can, what you believe about the market, the differentiation, the growth model, and the team: do it. The clarity forces honesty about what you actually know versus what you’re hoping is true.
The interview process, specifically sitting with Michael Seibel and answering questions about the business, taught me something about how I was thinking about Speak AI’s position in the market. The questions YC partners ask are mostly not the questions you’ve been preparing answers for. They’re the questions you’ve been implicitly avoiding because answering them honestly would require updating something you’re attached to. I updated some things after that conversation.
What Techstars actually gave us
Before YC, Speak AI went through the Techstars program. The value of an accelerator is almost never the equity you give up or the check you receive. Those things matter, but they’re not the primary value. The primary value is acceleration of the learning that would have happened anyway, concentrated into a period short enough that you’re forced to absorb it quickly.
The mentor network was the most consistently useful part. Not the formal mentor sessions, which are hit or miss depending on who you’re matched with, but the informal conversations with other founders in the program who were three to five years further along than we were. The advice you get from someone who recently solved the problem you’re currently stuck on is categorically different from advice from someone who solved it decades ago or who has only ever advised on it without having done it.
The Techstars brand did open doors that would have taken longer to open otherwise. That’s an honest acknowledgment, not a boast. Institutional credibility accelerates some conversations that would otherwise require more time to build. Use what you have.
The SEO moat in concrete terms
I want to be specific about what “invest in content earlier” actually means for a B2B SaaS company, because the general advice is useless without the specifics.
The content that drove the most valuable traffic for Speak AI directly answered questions people typed into search engines when they were trying to solve the problem our product solves. Not blog posts about the company. Not thought leadership about where the industry is going. Tactical, specific, useful content: how to transcribe a qualitative research interview, what the best tools for thematic analysis are, how to analyze focus group data at scale.
That content converts better than any other kind because the person reading it is already in the mindset of solving a problem. They’re not browsing. They’re looking for something specific. If your content actually helps them and your product is the next logical step toward solving the underlying problem, the conversion path is short.
The compound effect is real and it runs on a timeline that feels frustrating until it doesn’t. Articles we published in 2021 are still in the top ten for their target queries today. The monthly traffic from those articles has accumulated for four years. The total number of users those articles have contributed to bringing to Speak AI is in the tens of thousands. The work was done once.
The mistake most early-stage SaaS companies make with content is treating it as a short-term channel. They write a few articles, don’t see immediate returns, and redirect the effort toward paid acquisition or other channels with faster feedback loops. Paid acquisition stops the moment you stop paying. Content compounds. The right comparison isn’t this month’s content ROI versus this month’s paid ROI. It’s this month’s content spend versus the next five years of compound organic traffic versus paid acquisition costs over the same period.
The numbers behind the story
250,000 users across 100+ countries. +123% gross volume year over year. +15% registrations month over month as of early 2026. These are the numbers I share publicly because they’re directional and honest without exposing the revenue details I keep private for competitive reasons.
What they represent is a compounding curve that was very flat for a long time before it started moving. The first 10,000 users took much longer than the next 10,000. The next 10,000 took much less time than that. This is not a surprising observation. It’s how compounding works. But experiencing it firsthand is different from understanding it abstractly. The years when the curve was flat felt like failure. Some of them were failure in specific ways that we needed to fix. Most of them were the slow accumulation of the preconditions for faster growth later.
The platform is now used across healthcare, academic research, market research, journalism, and enterprise. That breadth wasn’t designed. It emerged from building something good enough that different types of users found it and decided it solved their problem. The lesson I take from that is that category definition is something you earn through adoption, not something you claim through positioning.
What this actually takes
There’s a version of the bootstrapped growth story that gets told as if it were primarily about cleverness. Finding the hack, the channel, the angle that unlocks growth. Some of that is real. But most of what it actually takes is less glamorous: building something that works, understanding your users deeply enough to keep making it better, and doing the consistent work of telling people it exists over a long enough time horizon that the compounding kicks in.
250,000 users didn’t happen in a moment. It happened in accumulated thousands of moments where the product was useful enough to come back to, and the content was useful enough to find in the first place. That’s the whole story, mostly.
What we got right that we don’t talk about enough
There’s a version of the growth story that attributes everything to strategy and execution. Some of it was luck, timing, and being in the right place when the market shifted.
We built a transcription platform before AI transcription was considered production-ready by enterprise buyers. We spent years with a product that worked well, at a quality level that exceeded what most people expected, while the market’s expectations were still anchored to early-generation ASR. When the market’s expectations updated, when buyers stopped assuming AI transcription was unreliable and started assuming it worked, we had four years of product refinement, customer relationships, and content compounding behind us.
We also built for the segment that cared most about quality. Researchers who are publishing work based on interview data cannot afford transcription errors. The quality bar they held us to forced us to be better than we needed to be for most use cases. That investment in quality for the demanding segment paid dividends when less demanding segments started adopting AI transcription. We were already at a quality level that exceeded their needs, and the pricing premium that came with it reflected genuine differentiation.
Being early in a market that becomes large is less about genius and more about conviction and patience. We were convinced the problem was worth solving and that AI transcription would eventually be good enough that large numbers of people would want it. We were right. It took longer than we expected. Companies that make it through the long period before a market matures are not always the ones that made the best decisions. They’re the ones that didn’t run out of patience or capital first.