Bootstrapping AI SaaS: costs, CAC, and real traction with Todd Hooper @LandingSite

saas.unbound is a podcast for and about founders who are working on scaling inspiring products that people love, brought to you by https://saas.group/, a serial acquirer of B2B SaaS companies.

In episode #46 of season 5, Anna Nadeina talks with Todd, founder of Prerender, one of the first brands we acquired at saas.group and now also a founder of LandingSite.ai, AI-Generated Websites For Every Industry.

I studied computer science at Georgia Tech, spent a short time in the corporate world, then found my way to smaller startups in Boulder, Colorado. Boulder is magnetic for founders — I met my cofounder Andrew there and we kept building stuff together. I like the zero to one part of startups: spotting a real problem, hacking together a solution fast, and finding customers who will pay for it.

The birth of PreRender

Back in 2013 to 2014 we ran into a recurring issue: modern JavaScript web apps like Angular and early React rendered blank pages to search engine crawlers. Search engines and social networks couldn’t index or preview content because they saw an empty page.

One weekend I decided to fix it. Instead of letting crawlers see a blank white page, we loaded the site in a browser, took a snapshot, and served that to bots. That made the content visible to Google and other crawlers. I built the first version quickly, put it online, and it picked up traction through developer communities. Even though the original code was open source, I wrapped a hosted service and support around it and started charging from day one.

“It worked out really well. I started getting people paying right away. I started charging from day one, which is a good thing that I always advocate for.”

Growing PreRender as a solo founder

PreRender was a tiny, very technical business for a long time. It was largely just me until we hit roughly $2 million in annual revenue. At that point I brought my brother on to help, mostly with customer support. The product attracted both hobby devs and big enterprise customers; even Fortune 50 companies ran into indexing issues after major releases and needed a quick fix.

Support was highly technical — developers wanted clear answers — and I spent a lot of time on it. In hindsight, hiring dedicated support earlier would have helped, but the work also kept me close to customers and their problems.

The acquisition and why founder-friendly terms matter

When it was time to consider an acquisition, I gravitated toward a buyer that was upfront and founder friendly. The right fit matters: I did not want PreRender to be swallowed as a tiny product line inside a large organization where alignment would be poor.

On valuation and negotiation I kept things simple: I shared the number I needed to walk away and the acquirer agreed. There was minimal back-and-forth and the process was smooth. Crucially, the acquirer wanted the brand to live on independently and allowed the product to continue growing under a dedicated team.

Transition period and operational wins

During my six month transition I helped the new team reduce operational costs and focus work where it mattered. For example, PreRender’s infrastructure was all on AWS and the bill had ballooned to about $1 million a year. Moving off that setup and optimizing systems reduced it to roughly $200,000 per year — a huge win for margins and sustainability.

Why I started LandingSite.ai

After selling PreRender I took time off — I barely opened my laptop for a year — and then teamed up with Andrew again. We tried a few ideas and landed on LandingSite: an AI-first website builder focused on marketing websites for small businesses. The goal was narrow and practical: if you run a local business like landscaping or moving, we build a website for you, with logo, copy, and images, and make it publishable fast.

Product positioning: AI websites for every industry

LandingSite is deliberately focused. We are not trying to build a general purpose app development platform. Instead we aim to help non-technical business owners who either do not have a website or find tools like Squarespace too difficult or expensive. We produce full HTML websites (not just client-side apps), which helps with discoverability and avoids many indexing issues.

Key product differences:

• Generate a complete website from simple inputs: business type, location, phone, services.
• Create logo, text, and select high-quality images from a licensed library.
• Publish real HTML that search engines can index without extra tooling.

How we find customers and niche down

Niching matters. Instead of targeting “everyone,” we run experiments to find customers who have the biggest pain and will pay for a solution. Early tactics included:

• Targeted SEO pages for specific business types — people searching for “website for moving company” or similar queries.
• Building manually and testing demand: we once generated websites for newly formed LLCs and mailed them a printed mockup of their site to get attention.
• Offering parity pricing by country early on to reach global customers and prove the model works across languages.

That mail experiment was part growth hack, part learning exercise. We only mailed a small set and did not see conversions, but it remains an interesting direct approach to grab attention. We plan to try again now that our sites look better.

Working with AI: band-aids, evaluations, and model swaps

AI is the enabling technology, but it still requires engineering and process work. We built our product to tolerate imperfection and improve over time as models evolve. Some practical lessons:

• Start with band-aids and iterate. Early on we used templates with limited AI text replacement. As models improved, we shifted to fully AI-generated custom layouts.
• Use eval systems. We collect thumbs up and thumbs down and feed problem examples into an evaluation workflow to improve prompts and code.
• Don’t have the same model grade its own output. Using a different model to review generated code or content catches more issues than self-evaluation.
• Plan before you code. Writing a clear markdown spec with one model and then switching to another model to implement code leads to better, more reliable results.

“We generate everything in Sonnet and then we have our open AI do the evaluation on our evals. An AI grading its own generated content doesn’t do as good as a different model grading that content.”

AI and code: how much is AI doing?

Almost all of our code today is AI-assisted. I am not handing production to the models blindly; every line is reviewed and we keep careful guardrails for security and correctness. For the tech stack we use, AI performs extremely well, but you still have to check, iterate, and be ready to fix issues. Code generation has made experimentation cheap: try an approach, review, iterate, and discard if it fails.

Dealing with hallucinations and UI limitations

AI can hallucinate facts or handle spatial/UI tweaks poorly. For non-technical small business customers this can be frustrating: they see a visually perfect preview but sometimes want tiny layout tweaks that AI struggles with. We handle this by:

• Setting expectations early: the product can make mistakes and we offer simple undo/feedback mechanisms.
• Using support to fix high-impact issues. Every negative feedback entry is routed to support and our eval pipeline.
• Continuously improving prompts and components as models improve.

Pricing, costs, and bootstrapping an AI SaaS

People often say you cannot bootstrap an AI product because model costs balloon. It is harder, but possible with the right approach. Key rules I follow:

• Charge from day one. Price based on customer value, not model cost.
• Target business customers who can pay meaningful amounts rather than consumer hobbyists.
• Optimize infrastructure aggressively. From PreRender I knew infrastructure could drive a huge bill. We keep the balance between growth and cost in mind always.

Our pricing evolved over time. We started low and used parity pricing internationally to reach more businesses. Today our plans sit in familiar territory compared to Squarespace and Wix: a low tier for a single homepage and a standard tier for multi-page sites. As we add AI-heavy features like automated SEO or marketing tools, we may introduce add-ons or higher tiers.

Traction and what I consider wins and failures

LandingSite is growing — we are on track to add roughly $1 million in annual revenue this year and expect stronger growth next year. The biggest wins are product-market fit within a clear niche and the ability to move quickly thanks to AI. Failures have mostly been operational: pushing buggy code, dealing with angry customers, and learning where to invest in support early.

Practical hacks and advice for founders

Some practical, repeatable tactics that have worked for me and that other founders can try:

1. Speak to customers early and often. The clearest path to product-market fit is to find people with acute pain and solve it for them.
2. Do things that do not scale to find product hooks. Building bespoke assets and mailing them, or manually creating initial customer experiences, can reveal demand and messaging.
3. Charge based on value, not cost. If your product materially helps a business make money, price it accordingly.
4. Use different AI models for generation and evaluation. Cross-model review catches more problems than self-reviewed output.
5. Write specs first. Build a markdown spec with your AI, iterate on it, then switch to an implementation model to generate code.

Final thoughts

AI has changed the speed and scale at which startups can build. It does not replace product thinking, customer discovery, or operational discipline. If you focus on a tight niche, charge based on value, and build processes that tolerate model imperfections, you can bootstrap an AI SaaS that grows sustainably. The work is still human: talking to customers, fixing bugs, and learning fast. AI just helps you build and iterate far faster than before.