Your Website Should Rewrite Itself While You Sleep

What we talked about.

In this episode of the Sunny Ray Show, host Sunny Ray talks with Aviv Frankel of Shuttle Base about letting AI handle conversion rate optimization for e-commerce sites. Frankel, a former TV journalist and tech reporter in Israel who crossed over to become a founder, explains how his first e-commerce company struggled to convert traffic and relied on slow, guess-driven testing with designers, developers, and analysts. That experience led to Shuttle Base, which scans a site, identifies what is not working, builds test variants including design, copy, and code, runs the A/B test, and updates the site. Humans approve changes at first, especially at brand-conscious enterprises. Frankel shares a story of a luxury footwear brand where a quick add button reportedly drove about $600,000 more than an icon, discusses the company's rename after a naming collision with a Chinese AI lab, and explains why trust-building tests tend to lift sales. He also covers traffic requirements for smaller stores and the future of optimizing for AI agents.

Stop guessing at conversion tests: Shuttle Base's Aviv Frankel explains how AI can find, build, run, and ship website improvements automatically.

The questions, and the answers.

What does it mean for a website to rewrite itself while you sleep?

For an e-commerce site, it means AI should handle conversion optimization instead of a pile of manual work from designers and developers. You've already spent heavily on building the site and driving traffic, so the goal is improving your conversion numbers without the heavy lifting.

What made you move from covering tech to building companies?

I was on the other side of the screen, meeting amazing founders, and eventually thought I could do that too. It was very humbling. Going from being a TV personality everyone wanted to talk to, to searching for investors, partners, customers, and employees, was a real struggle.

When did you realize the process itself should be the product?

It started with my first e-commerce company. We had traffic that would not convert, and CRO felt like voodoo. Every test needed designers, developers, analysts, and product people, and two months later I'd forgotten what I wanted to test. So we built a tool that automates the whole CRO process.

Where does a human still step in?

Technically no one needs to, but brands want ownership, especially enterprises with style guides. For the first few months, a human approves everything. You get a message that a new test is ready, optionally prompt tweaks, click launch, and check results in about two weeks.

Do merchants believe a number like $600,000 from one button change?

Because it's an A/B test, you see the original and the variant and how much more money the variant brought in, so no extra proof is needed. For a luxury footwear brand, Laudet, our AI found an icon wasn't getting clicks, and a black quick add button drove about $600,000 more.

What stops the software from shipping a loser to all your traffic?

I'd say there are no losers, because you always learn something about your site and audience. Even a test that isn't a statistically significant winner tells you to abandon an idea or double down on the concept. Then the AI comes up with a different variation.

What happens when a winning test helps sales but hurts customer trust?

I'm not sure that's a real case, because when trust is high, sales are usually high. We have buckets of tests, including trust and risk tests like free shipping, free returns, badges, and social proof. More trust most of the time means more sales.

Can a smaller store get useful results without huge traffic?

You need enough traffic to support experimentation, but not Nike scale. Tens of thousands of visitors per month can work, depending on how many transactions you have, since that affects statistical significance. We work with young, growing brands that don't have the traffic of the biggest names.