Growing the first-ever AI product design tool to $3.5M ARR and getting acquired

Tony Beltramelli built the first-ever AI product design tool, Uizard. He was so early that he had to educate people about AI to create a market. Then he used that content, which was already ranking in the SERP, to surf the AI wave.

In 2024, when the company was at $3.5M ARR, it got acquired. Here's Tony on how he did it. 👇

The world's first AI product design tool

My journey is rooted in a deep passion for AI and machine learning.

I fell in love with computer science through my interest in product design. I pursued graduate studies at the IT University of Copenhagen and ETH Zurich, where I was captivated by the potential of AI and machine learning to empower creativity and solve complex engineering problems.

I cofounded Uizard and led the company as CEO, until it was acquired. Uizard is an AI-powered product design tool that simplifies prototyping and wireframing for teams. It enables customers to design mobile apps, web apps, websites, and interfaces for desktop software.

When we started the company in 2018 and when we launched out of beta in 2021, it was the world’s first AI-powered product design tool — long before the genAI frenzy we’re all familiar with today.

We grew Uizard to serve over 3 million users and reached around $3.5 million in ARR by the time of its acquisition. At its core, Uizard was built to democratize design by making it accessible to non-designers such as entrepreneurs, product managers, developers, and empowering teams to iterate faster and bring ideas to life with ease.

These days, I'm the Head of Product for AI at Miro, where I lead the design and development of AI solutions to enhance collaboration and accelerate innovation workflows our customers love Miro for.

Virality of open source

The idea for Uizard came from my experience working as a front-end web developer, my first job between undergrad and grad school. I was astounded by how manual and inefficient the design-to-development process remained. The workflow — translating design pixels into code — had barely evolved since the 1990s. 

A few years after completing my Master’s degree, I was working as a data scientist and was playing around with AI research in my free time. I was interested in exploring how machine learning and computer vision could automate parts of the front-end dev process, reducing friction and enabling faster iteration for product teams. This was in 2017, still pre-LLM/genAI era so it was very messy, highly experimental, and much more research-oriented than purely hacking.

Sometime in 2017, one of my week-end projects started to work. I was able to feed a screenshot of a UI design into the AI model, and it would generate HTML code for it. I was so excited that I decided to write a research paper describing the algorithm, recorded a demo video, and shared it all online with the code open sourced on Github.

The project — called pix2code — went viral and people started reaching out to ask me if they could use this for work. That’s when I decided that maybe I should build a company to productize the research into a real SaaS solution.

I had saved up enough to be able to run six months without a salary. I took a leap of faith, quit my job, started building the product, convinced the three smarter people I knew to join me as cofounders, and started pitching VCs to raise capital. On the 6th month, the last month where I could pay rent, we finally closed our pre-seed round — led by Evan Nisselson at LDV Capital in NYC.

Building in AI before it was cool

Building Uizard was a labor of love that took 3.5 years from inception to launch — from early 2018 to mid 2021. We developed a user-facing SaaS product with a real-time collaborative canvas, and proprietary AI models to generate UI designs, wireframes, prototypes, and code. The journey from a proof-of-concept to a functional product required countless iterations, user feedback, and refining the technology to align with real-world use cases.

Given how much R&D was required to build an MVP we could start charging customers for, we quickly realized that we would need to raise venture capital to buy us time and hire a team. In total, we ended up raising three rounds of funding at a total of $18.6 million, the latest of which was a Series A led by Insight Partners. By the time we were acquired by Miro in June 2024, we were on our way to raise our Series B.

Capital allowed us to invest in infrastructure, cloud-based GPUs to train our models, and most importantly, enabled us to hire an amazing team to help build the product. Our first two hires were a product designer and a computer vision engineer, as we were four cofounders able to cover AI engineering, SaaS engineering, product, infrastructure, accounting, operations, and marketing on our own.

Finding the right AI tech stack

Our tech stack evolved over time but some of the tech remained roughly the same.

On the product side, we used a Javascript / Typescript stack: React on the frontend and Node on the backend. We used Python for infra as well as model training and serving using TensorFlow, Keras, and PyTorch. Our entire platform was built on AWS.

The introduction of GPT models further enhanced our platform’s capabilities, allowing us to combine proprietary AI with OpenAI’s LLMs and, eventually, Anthropic Claude as well. LLMs and genAI enabled us to “glue” our proprietary models into more complex pipelines and transformed the product from good to great within weeks.

Building with HTML DOM

One key challenge that I am sure few people will realize is that the infinite canvas in the Uizard collaborative editor is actually not using webGL nor the HTML canvas at all. It’s entirely built with HTML DOM! Yes you read that right, our entire infinite canvas is just the plain DOM and built with HTML elements. Why would one do that?!

At some point in our journey, we realized that the only way we would enable our users to be successful is to give them a canvas where they would be able to freely edit things generated by AI, and iterate on the results further by hand or with AI.

At this crossroad, we had two options:

1. Either we would do like Miro, Figma, etc. and build an infinite canvas using the HTML canvas and webGL. It’s great for performance but you basically need to build a ton of software to support drawing, editing, manipulating of UI elements. The engineering complexity could easily explode and it would take a long time to build a comprehensive editor this way.

2. Or we would “draw” things directly in HTML by inserting them into the DOM. Given that we were in the business of building a product to enable teams to design mobile apps, web apps, websites, and interfaces for desktop software, the HTML DOM already natively supported the concept of “buttons”, “checkboxes”, “radios”, “input fields”, etc. As such, by building our editor directly using the DOM, we would get a TON of functions for free directly built in the browser to enable drawing, editing, manipulating of UI elements.

We picked the 2nd solution because time is very precious for any startup, and this is what enabled us to iterate the fastest on product to learn from customers before investing into the first solution eventually. Even though the DOM is already benefiting from a ton of optimizations natively built-in modern browsers, our biggest challenge was performance issues when certain users would build very complex apps with hundreds of individual UIs in a single project. Nevertheless, this solution was more than enough to support 98% of our customers.

We optimized React to its limit to be able to run our infinite canvas in the DOM and our plan was to move to webGL post Series B but we were acquired by Miro before that.

Building trust in AI

One of the toughest challenges was building user trust in AI. Many users were initially skeptical of AI-powered tools, especially in 2021 before ChatGPT. Educating users about the capabilities and limitations of our technology was critical to overcoming this hurdle.

Another major obstacle was waiting too long to introduce pricing, which delayed key validation points for product-market fit.

Looking back, I would have engaged customers earlier in the development process and focused more on distribution than purely product development. The “build it and they will come” doesn’t work and was a lesson learned the hard way — true success came when we paired a strong product with effective distribution strategies, including content marketing and viral loops.

Growth

Word of mouth

Our growth has been largely organic, with 95% of new users coming from word of mouth and community sharing. And this was jumpstarted by early demo videos showcasing Uizard’s AI capabilities, which went viral and attracted our first cohort of users.

Because our product is very visual (AI generating UI designs for apps), a lot of early users and customers would record videos of their experience using our product and share it on social media — this led to viral growth that further attracted new users.

SEO

As the AI space gained momentum with launches like ChatGPT, our content marketing positioned us to capture the surge in demand for AI tools in design. We’ve invested early in SEO to start ranking for long tail keywords about AI and product design — which became a phenomenal tailwind as AI became hyped.

When OpenAI launched ChatGPT and the world started to become interested in AI-first solutions, we already had our entire AI-focused content ranking on Google with a product that was live and ready to use. This was a fantastic tailwind and there were multiple months in 2023 where we were adding $1M of revenue in a single month.

Waitlists

One of the most efficient growth levers we leveraged multiple times was the use of pre-launch waitlists for new products or features we were building.

It’s pretty straightforward: Whenever we were a few months away from the launch date, we would create a marketing video, and have a landing page for people to sign-up to a waitlist to get early access once the feature would go live.

When people would sign up to the waitlist, we would tell them what’s their number in the queue, and they would then get the possibility to climb up the waitlist by inviting other people to sign up. This simple gamified mechanic enabled us to grow our pre-launch signup list for Uizard 1.0 to 100k in a few months.

When we re-applied the same technique to promote the release of Autodesigner, our most advanced AI-generated design engine, we quickly grew the waitlist with 10k or more users signing up every other day.

Throughout 2023 and 2024, we were getting 100k new users signing-up every month, sometimes more than 240k in a single month.

Product-led sales

Uizard operates on a freemium model, with paid tiers offering advanced features and increased usage limits. Adjusting pricing based on user feedback and competitive analysis has helped us optimize revenue while maintaining accessibility for small teams and startups. Here are some of the documented techniques we used to come up with our price plans.

Apart from the self-serve business model, we started to put a product-led sales motion in place in early 2023. We would identify self-served active users that were part of enterprise companies, and we would reach out to understand if we could onboard their entire team to Uizard, increase their feature limits, and give them access to the latest AI improvements.

We were able to close deals with Fortune 500 companies this way, but it was a very early enterprise motion. In terms of scale, by the time Uizard was acquired by Miro in June 2024, our self-serve GTM motion was responsible for 98% of revenue, and enterprise sales 2%.

Focus on distribution

Focus on distribution from day one. It’s not enough to build a great product—you need a strategy to get it into the hands of users. Stick to a narrow target market initially and validate your ideas with real customers.

And don’t shy away from pricing early; it’s a crucial validation point that can guide your product development. Paying users will care more and provide higher quality feedback.

Lastly, the book Hacking Growth by Sean Ellis and Morgan Brown was transformative for us — but don’t invest in growth until you have strong signals of early product-market-fit.

What's next?

Being an employee again is actually a lot more fun than I expected! I guess I am lucky to be part of an organization where I have the trust of the CEO and CTPO to drive things independently and freedom to operate.

So I am excited to continue bridging the gap between cutting-edge AI technology and user-friendly apps at Miro, making advanced tools accessible to a broader audience and enabling users to bring their ideas to life as quickly as possible. We’re serving over 80 million users and we’re on our way to reach $1B in revenue.

You can visit my personal website at tonybeltramelli.com or connect with me on LinkedIn or on X at @tbeltramelli. For more about Uizard, check out uizard.com. For more about Miro, check out miro.com.