Finding product-market fit in a fast-paced and competitive market

Emma Lawler is a second-time founder building an AI platform called Velvet.

Velvet has already logged over 60 million requests and is approaching six-figures in ARR. Finding product-market fit in such a fast-paced market is a challenge, but the team keeps momentum by shipping early and often. 

Here's Emma on how they're doing it. 👇

Second-time founder

I was an MBA student and an entrepreneur in residence at Chicago Ventures. I gave myself two years of personal runway to validate ideas, trial co-founder relationships, and work in venture capital. 

My last startup was a developer marketplace called Moonlight, which I grew to $10k MRR with $55k monthly GMV. We sold the company to a Google-backed startup called PullRequest in 2020. Now, I'm working on Velvet.

Velvet is an AI gateway to run model analytics, evaluations, and continuous monitoring. We’re just at the beginning of a huge wave of AI adoption. It’s becoming part of the mainstream tech stack, and Velvet provides the infrastructure to observe and control models in production.

We’ve logged more than 60 million requests for startups building AI features using foundation models like OpenAI and Anthropic, and hundreds of developers trust Velvet to run their LLM operations. We're approaching six figures in revenue with a mix of bottoms-up developer adoption, and annual contracts sold to growth-stage startups. 

Building based on customer feedback

Our first product at Velvet was an AI copilot that enabled text-to-SQL on top of any database. Most of our customers used it to analyze their LLM data from OpenAI and Anthropic. We studied usage and began prototyping AI-specific tooling like request warehousing, caching, and evaluations. 

It was just me and my co-founder, Chris, for the first year, with the occasional fractional engineer. Once we saw signals of real demand, we brought on two founding engineers (Jack and Andy) to move faster. Things accelerated quickly once we had a paying customer using us in production

In spring of 2024, one of our customers requested access to the AI gateway. As heavy users of OpenAI’s batch endpoint, they needed to granularly analyze cost, latency, and other metrics.

We spent the summer building a proxy for them to warehouse request logs from OpenAI to their PostgreSQL database. They were an amazing design partner because they had product-market fit, a senior engineering team, and strong opinions. 

In September, we launched a self-serve version of the product. Now any developer can sign up and start capturing logs in about five minutes. Developers use the platform to run model experiments, continuous monitoring, and fine-tuning workflows.   

Tech stack

We’re a proxy that sits between a company’s application and the AI models they use. Developers can use our hosted service to get started, which is self-serve and takes about five minutes to get set up. For companies at scale, we offer custom deployments.  

Our application is built with Next.js and Node, and we spin up a Neon database per workspace. For paying customers, we can log requests to any database they control. This allows for complete data ownership and compliance. 

We built the first version of our gateway with Vercel. It was great to get started, but it quickly became unsustainable from a cost, scale, and latency perspective. We switched to Cloudflare Workers, KV, Queues, and Analytics Engine, which enable us to handle a huge volume of spikey logs at a reasonable cost.

Product-market fit is hard

Finding product-market fit is extremely difficult.

A startup can either invent something to create a new market, or compete in a market that already exists. WorldWideWeb (later called Nexus) initiated a new market of internet web browsers. In comparison, Mercury competed in a well-established banking market. 

I would recommend reading Clayton Christensen's Innovator's Dilemma for anyone operating a disruptive technology startup. He writes that thought leadership drives adoption, incumbent best practices don’t apply, and iterative failure is required. Demonstrating the value of disruption takes time, and a yet-to-exist market cannot be analyzed by investors. 

We raised money before we had a concrete product in-market, which meant we had 18 months to build a product, find product-market-fit, and show customer traction. The fundraising game also changed while we were playing it — after the pandemic bull run of 2022, the bar for getting to a second round of funding is extremely high. 

For people starting new companies today, I recommend developing a product before raising money. And if you do raise money, prioritize enough cash and personal runway to show meaningful traction in this more risk-averse fundraising environment. 

Experiment with pricing

We’re a pure software business, so our costs are upfront rather than ongoing. This means we can implement value-based pricing based on what each customer is willing to pay. 

We started by billing per request for logs warehoused through our gateway. This pricing is great for bottom-up freemium adoption, but quickly becomes unreasonable at scale. For example, our first enterprise customer quickly went from 0 to 40 million logs.

To control for spikey usage, we recently introduced annual plans with reasonable logging tiers. Paying customers get access to premium features like evaluations, monitoring, versioning, and custom deployments.   

The six most useful growth tactics

We have a two-pronged go-to-market similar to Vercel, Supabase, or Pinecone. We offer free self-serve tooling for developers, then sell annual contracts to CTOs or CPOs at growth-stage companies. 

When there’s a major platform update, we launch on HackerNews, Product Hunt, and forums like Indie Hackers. We also send a weekly newsletter to existing customers with our latest feature announcements and articles. 

Here’s a list of our most useful tactics:

• Weekly newsletter: We subscribe new users to our Beehiiv newsletter, then send a weekly update about new features, articles, or industry trends. This keeps customers engaged, even before they’ve adopted the platform. 

• YouTube: We record demos of new platform features to embed on the website, docs, and product articles. These go a long way in showing the value of our features before someone creates an account. 

• HackerNews: Our best spikes in user signups come from HackerNews. It’s hard to control or plan around, but making it to the homepage drives tens of thousands of page views, hundreds of signups, and meaningful feedback. 

• Product Hunt: I think of Product Hunt as a PR announcement more so than an acquisition tactic. We rarely see a meaningful bump in new users from the site, but it’s a good forcing function to ship a cohesive product and website to demo new features. 

• LinkedIn outbound: Sending direct messages to relevant founders and executives results in sales meetings. Especially if I speak at an event or host a meetup, and then follow up with a targeted campaign. 

• Event demos: Speaking at conferences, developer events, and AI meetups in New York or San Francisco has been a great way to meet new customers. Well-designed business cards are a surprisingly effective tactic since people rarely see them.  

Maintain momentum

Momentum is the only thing that matters when building a 0-1 startup. This means constantly shipping features, talking to customers, and communicating progress. 

I write a weekly memo each Sunday, then send it to the team at 8 AM Monday morning. This sets the tone for the week and keeps things moving. We do a daily standup in Slack, and the entire team meets for a 30-minute work session each day. And we aim to ship new code every week, followed by sharing updates with customers via our product newsletter. 

Operating a startup is overwhelming in every way. If it doesn’t feel chaotic, you’re likely not moving fast enough to disrupt incumbents. Setting timelines, metrics, and a weekly routine creates a culture of accountability — a reason to keep moving forward together.  

The five major decisions of a founder

Life as an entrepreneur is a rollercoaster. When things are going well, it feels like you’re on top of the world. You’re also solely responsible for building the product, attracting customers and investors, and showing up with a smile on your face every single day. It’s a huge commitment, and it takes a toll on your personal life. You absolutely should do it if you’re willing to pour every ounce of your energy into this new venture. 

Your lifestyle will be largely influenced by five major decisions:

1. Bootstrapped/VC-backed: Self funding gives you full control, but you can’t invest in anything until you have revenue. If you raise money, you have 18-24 months to figure something out, and can pay yourself and others to scale your vision quickly.  

2. Solo founder/cofounder: Starting a business alone means you maintain full ownership. You’re also solely responsible for product, revenue, recruiting, and fundraising. A co-founder is a partner by your side through every high and low.

3. Junior/senior employees: Junior employees are affordable and amenable, but are inefficient and require training. Senior contributors are expensive and their opportunity cost is high. They also do high-quality work without micro-management. 

4. Remote/in-person: It’s really hard to get a 0-1 startup off the ground without deep collaboration. The more synchronous time you can spend together, surrounded by other ambitious people and startups, the higher your chance of success. 

5. Lifestyle/Hustle: How much do you like to work? Make sure you collect employees, mentors, and investors who are aligned with your long-term vision and work ethic. Intrinsic motivation is the only thing that keeps your company moving forward.

Resources

Here are a few things I’ve found inspiring in my founder journey: Aspiration, Pattern Breakers, Super Founders, Zero to One, The Struggle, How to Do Great Work, and How to Find Product Market Fit.

And you can follow my journey on X, LinkedIn, or my personal website. And check out Velvet!