Introducing Orca: AI-Powered Experiment Ideas for Web

The problem it aims to solve

I've seen repeatedly:

• experiment hypotheses are often too broad to generate meaningful insights

• valuable learnings rarely build into a lasting knowledge base.

  
  

So, testing becomes a series of isolated wins and losses rather than a structured learning process.

How it solves it

Provide Orca AI a website URL, the site's primary objective, and a little audience context, and it'll generate 3 hypothesis-driven experiment ideas to optimise the page in my preferred 'IF', 'THEN', 'BECAUSE' structure.

This means that when a result is determined, you have audience insight.

It's a proof-of-concept MVP at this point, but the idea behind Orca is that you would tell it the results of experiments, and the AI would remember how your audience responded to previous results when proposing ideas for future optimisations.

Built with AI

This project has also been an opportunity to explore AI across the entire product development lifecycle—not just within the product itself. From early concept development and UX flows to design, coding, analytics, testing and refinement, AI played a role at every stage.

The main tools I used were v0, Framer, ChatGPT, Claude, and Cursor, each proving useful in different parts of the process.

I'll be sharing more about what I learned building Orca, including the AI workflows that worked well, the tools I'd recommend, and the lessons from plenty of trial and error along the way.

If you'd like to try Orca, I'd love to hear what you think.

Try now