Growing a talent marketplace to $2.5M+/yr after two years of losses

David Stepania is the founder of ThirstySprout, a bootstrapped talent marketplace that's bringing in over $2.5M in annual revenue. He's also the founder of ChoppingBlock.ai. And it all started with a beach, an idea, and a lot of credit card debt.

Here's David on how he got here. 👇

Make it work

I'm David Stepania, a two-time bootstrapped founder. I was born in the Republic of Georgia as the Soviet Union was collapsing, attended 13 different schools across multiple countries growing up, and eventually landed in Seattle, where I went to the University of Washington. That upbringing taught me one thing above all: resourcefulness. During some periods, textbooks were a luxury, so "make it work" became my default operating mode — excellent founder training.

Today, I run ThirstySprout, a talent marketplace that places remote AI and engineering talent — mostly from Latin America, Eastern Europe, and Asia — with funded US startups and enterprises. Companies like The Real Real, Hopper, Mailchimp (Intuit), Rover.com, and a wave of VC-backed startups have used us to build their tech teams.

I also built ChoppingBlock.ai, an AI salary and jobs intelligence platform with a 40K+ subscriber newsletter and a podcast, which doubles as our audience engine.

ThirstySprout crossed $2.5M in annual revenue a while back — and our growth since then put us at #247 on the Inc. 5000 in 2024. This year, we aim to double that and push past the $5M mark with a dramatically leaner team than we had the first time around.

Born out of failure

ThirstySprout was born out of failure. I was coming off a sabbatical in Hawaii after my previous venture, nursing a few product startups that hadn't worked out. The realization that hit me on that beach was simple — to build an extraordinary startup, you need an extraordinary technical team, and finding one you can trust is brutally hard. I didn't have a product idea, so I decided to solve the problem I knew firsthand.

We started as a two-person product development agency, and we made every mistake in the book. We took on underbudgeted projects and over-delivered just to earn testimonials. After enough of those, we realized the model didn't scale, and a friend suggested we focus on staffing remote technical talent instead.

The pivot validated itself almost immediately: A single cold email landed us Rover.com during their hypergrowth phase in 2018. A six-figure engagement and a reference from a hot startup changed everything. Suddenly, we could win clients of that caliber repeatedly.

At the time, I had close to $0 in cash reserves. I funded the gap on personal credit cards — we ran at a loss for the first year or two, peaking somewhere between $50K and $100K in the hole, while I paid myself no salary. I don't recommend that path, but stubbornness is a real business asset.

Building the initial product

For years, the "product" was a service run on top of other people's tools. No custom platform, no code. We deliberately refused to build technology until we had 50–100 actively engaged freelancers and proven, repeatable processes — because automating a process you haven't proven by hand just scales your mistakes.

The initial version was WordPress, then low-code tools like Webflow. The real "build" was the supply side: We went country by country looking for deep talent pools, and found our strongest footing in Georgia — the country I was born in. The cultural fit and engineering work ethic were exactly right, and today we're one of the top talent players there, competing directly with Toptal and Turing.

The initial build didn't cost money — it was those two loss-making years on credit cards, and the time to learn an industry I'd never operated in. The people who helped most were early clients who took a chance on us, and the founder communities I leaned on for advice.

Cutting back on the stack

Today, the spine of the company today is Claude — and I mean that literally, not as a buzzword. Candidate screening and ranking, financial analysis, contract drafting, content, strategy: Most of the real work runs through dedicated Claude workspaces, one per business domain, each loaded with custom instructions. For agentic builds, I use Claude Code.

Around that: Google Workspace and Slack (Slack also runs our community), Beehiiv for the newsletter, Riverside for the podcast, Deel for international contractor payments (team across four continents), Calendly, Fathom for call transcription, BetterProposals for proposals, and Attio as the CRM.

We're also building our talent marketplace product on Next.js, Postgres, and Supabase — anonymized candidate cards, a lightweight ATS, and a Slack notification loop.

The biggest stack change we made was subtraction: Earlier this year, I audited our subscriptions and found 50+ tools, including five overlapping AI subscriptions. We cut aggressively. One tool used at 90% depth beats five used at 10%.

Two fee models — and natural expansion revenue

We have two fee models, and their mix reflects the business's current state.

The first is staff augmentation — clients contract vetted engineers through us, and we take a markup of around 30% on the engineer's rate. For context, competitors like Toptal run markups closer to 100%. This gap is our business model. Because we run lean, we can pay engineers more and charge clients less than the big platforms, and still make healthy margins.

The second is direct placement — a contingent fee of 20% of the candidate's annualized base salary, paid only upon a successful, lasting placement, backed by a guarantee period. We exclude signing bonuses, equity, and benefits from the calculation, which keeps it clean and client-friendly.

Historically, staffing engineers has been the bulk of our business, rather than direct placements, though it's currently split about 50/50.

We started charging from day one; as a bootstrapped company, we never had a free-work phase to graduate from.

The model includes natural expansion. First, land-and-expand: a client who hires one engineer and has a good experience often hires more — most of our revenue comes from repeat placements, not new logos. Second, we also run contract/hourly engagements for clients who want flexible capacity rather than direct hires, which smooths revenue between placement fees.

We are building a bigger expansion play around the placement business: ChoppingBlock serves as an audience and data asset (salary data, jobs intelligence, newsletter sponsorship potential) for the supply side. And a marketplace product reduces our cost per placement. Recruiting fees fund the flywheel; the flywheel makes each fee cheaper to earn.

Owned audience is the only moat

For years, we primarily used high-volume cold email for growth. Early on, we sent a few hundred emails daily with open rates in the high 50s. At its peak, we sent thousands daily — but open rates dropped to the low 20s and reply rates to 2–3%, indicating the channel's declining effectiveness. It worked. It built the company. But it is now dying. AI-generated outreach will soon bury every inbox, and you cannot out-automate spam.

Over the last two years, we deliberately rebuilt our engine around inbound. Three channels:

• Community and relationships. We built founder communities (around 6,000 founders at the peak) and joined strategic ones like Hampton. Trust is the entire product in recruiting — word-of-mouth referrals are still our highest-converting source.

• Programmatic SEO. ChoppingBlock employs programmatic SEO for AI salary and jobs data, creating thousands of pages that target long-tail queries our exact buyers search for. It compounds while I sleep.

• LinkedIn content. I post contrarian, data-grounded takes on the AI talent market: salary data, hiring trends, and which roles are rising versus on the chopping block. That positioning as the "data guy" in AI hiring drives inbound conversations that cold email never could.

The lesson: Rented channels (cold email) got us to millions in revenue, but owned audience is the only durable moat in a world where AI makes outreach free.

Challenges and solutions at every stage

The early challenge was survival. Two years operating at a loss, funded by personal credit cards, no salary, in an industry I'd never worked in. The mistakes were classic — underpriced projects, no niche, no systems.

The mid-stage challenge was self-inflicted: we overhired. Growth felt like headcount, so we scaled to 50+ contractors and drifted from the lean operation that made us work. Unwinding that — getting honest that a smaller, AI-augmented team could produce more than a bigger one — was operationally and emotionally hard.

We deliberately made three changes.

1. We restructured around single owners of whole domains instead of teams. One person owns our entire programmatic SEO surface. One person owns community and data infrastructure. Each is heavily AI-augmented

2. We moved the grunt work to AI and kept humans on judgment. AI now handles candidate screening, first-pass matching, call recaps, proposal drafting, financial analysis, and contract drafting, with a human making the final call. Instead of forwarding thirty resumes and hoping, we send clients three to five that are exactly right, each scored against their actual hiring rubric. Fewer people, better output.

3. We changed the revenue engine itself. Cold email at volume required more people for list building, sequencing, and follow-up. Inbound doesn't scale with headcount: Programmatic SEO, LinkedIn content, and our newsletter compound whether or not anyone is working that day. Higher-quality demand, near-zero marginal labor.

As a result, the company runs dramatically leaner and produces more per person than we ever did at peak headcount.

The current challenge is the industry itself: AI agents are commoditizing exactly the part of recruiting (volume sourcing, resume matching, mass outreach) that most firms live on. We've responded by deliberately abandoning that layer and moving up to the judgment layer — deep vetting, fit, retention.

If I started over, I'd niche immediately instead of being a generalist for years — the riches really are in the niches. I'd start building an owned audience from day one instead of renting attention through cold email. And I'd set a 6–12 month window to test hypotheses instead of letting stubbornness fund losses on credit cards for two years.

Books and AI as unfair advantages

Books shaped the foundation. I've listened to over 200 on Audible — most of the business and self-help canon, from Zero to One to Deep Work to Essentialism to Principles. My takeaway after all of them: Many ideas are regurgitated, and you often read an entire book to find the one moment that unblocks you. But that one moment can be a game-changer, and sometimes an idea I picked up five or ten years ago clicks when I'm stuck today. So I can't name a single book that changed everything — cumulative reading provides a library to draw on at the right moment.

But the biggest resource shift in my career happened recently: I stopped reading about leverage and started using it. AI — Claude specifically — has been the single greatest advantage of my founder life. I did CFO-level financial analysis that found real margin issues in our books. I rebuilt our entire legal stack (MSA, contractor agreements) with AI drafting and a lawyer reviewing. I built an AI coach in Claude Code that pressure-tests my decisions before I commit. As a solo CEO, that's the closest thing to a cofounder I've found.

With that said, never automate a process you haven't proven manually. We successfully scaled every process by doing it manually first, painfully, until we understood it. Every automation disaster came from skipping that step.

Niche down, build manually, and build an owned audience

Here's my advice:

• Pick a niche where you have unfair context, and go embarrassingly narrow. I spent years as a generalist agency earning generalist money. The business only grew significantly when we became the people for a specific thing. If you're starting today, niche opportunities in AI are everywhere and mostly unclaimed.

• Do things manually before you build anything. Your first "product" should be you, doing the service by hand, learning what actually matters. Code written before that understanding is usually a waste.

• Start building an owned audience on day one — a newsletter, a community, a data asset, anything people intentionally visit. Cold outreach still works today, but AI will soon make every rented channel worthless. The founders who own audiences will be untouchable.

• Know the difference between stubbornness and strategy. Stubbornness kept me alive through two unprofitable years — but a 6–12 month testing window with clear kill criteria would have led me to the same place faster and cheaper. Persistence on the mission, ruthlessness on the tactics.

What's next?

Our headline goal is to grow ThirstySprout past $10M in annual revenue — and achieve it as we've always done: bootstrapped, with a lean AI-native team, rather than a bloated one.

Beyond this number, we have two tracks.

For ThirstySprout: Productize the marketplace — move from a services company using software to a software-enabled marketplace with service margins, without taking VC money. The AI-native talent niche is exploding, and we intend to own it.

For ChoppingBlock: Become the Levels.fyi of the AI era — the default place people check what AI-era roles pay and which skills are rising or dying. The flywheel involves user-submitted comp data unlocking aggregate insights, wrapped in a newsletter and podcast people enjoy.

And a personal goal. Right now I'm working from the beaches of the Riviera Maya near Cancún, which I'm not complaining about — but the real goal is to someday run all of this from Hawaii. The idea for ThirstySprout came to me there, so building it into something that lets me work from there would close the loop, which means a lot to me.

The overarching goal: Prove that I can build a modern, AI-native company of consequence with a tiny team and zero outside capital.

You can follow along on X, LinkedIn, and my personal site. Or check out The AI Chopping Block podcast.