The Reality of AI Startups: Founder Experiences

AI startups often promise to turn advanced models into useful products, but the process is harder than founders expect. Julie Bornstein, CEO of Daydream, thought her experience in digital commerce would make creating an AI fashion startup easy.

“I had breakfast recently with Bornstein and her CTO, Maria Belousova, to learn about her startup… The conversation took an unexpected turn as the women schooled me on the surprising difficulty of translating the magic of AI systems into something people actually find useful,” the article reports.

Even though AI apps have captured public attention since ChatGPT launched in 2022, studies show that many AI enterprise projects deliver no measurable value, except in areas like coding.

Daydream’s Journey: Turning AI into Fashion Solutions

Bornstein’s initial plan was to use AI to help customers find perfect garments. Signing up over 265 partners with 2 million products was simple, but understanding customer requests proved complex. Factors like who the customer is, event type, season, and style preference all mattered.

“What we found was, because of the lack of consistency and reliability of the model, and the hallucinations, sometimes the model would drop one or two elements of the queries,” said Bornstein.

To improve results, Daydream postponed its app launch and expanded its technical team. CTO Maria Belousova and her engineers developed a system using multiple AI models for color, fabric, season, and location. OpenAI models proved good for understanding clothing, while Google’s Gemini was fast and precise.

Combining Human Insight with AI Models

Daydream also relies on human input. Popular user requests, like seeing Hailey Bieber’s style, are curated by humans to guide AI. Trend collections are updated manually when new styles, such as cottagecore, emerge.

“We have this notion at Daydream of shopper vocabulary and a merchant vocabulary… How do you actually merge these two vocabularies into something at run time?” Bornstein explained. Visual models also help AI interpret products more accurately.

Shared Challenges Across AI Startups

Other AI startups face similar issues. Duckbill CEO Meghan Joyce said, “It has been so much more challenging on the AI front… It took us 10 million real-world interactions to get to the point to even be relevant or knowledgeable about real-world actions.”

In one test, Duckbill’s AI claimed to have called a doctor and scheduled an appointment with a receptionist named Nancy, but no call was made. Mindtrip CEO Andy Moss noted that AI travel assistants can respond irrelevantly when asked unexpected questions, requiring careful design to avoid mistakes.

Timelines and Patience for AI Success

All three founders agree that creating useful AI tools takes time and effort. Initial development took years, but persistence is producing results. Bornstein and her peers suggest that AI may bring productivity breakthroughs around 2026 or 2027.

PHOTO: FREEPIK

This article was created with AI assistance.