Replacing expensive A/B tests with AI models

This article is a part of Startup Spotlight, a series that features young, up-and-coming startups.

Jingshen Jiang (left) and Yilin Du (right). Photo credit: Jingshen Jiang and Yilin Du

Jingshen Jiang ran a Reddit campaign targeting different audiences with the exact same promotion. The wildly different results made him wonder if he could predict consumer reactions beforehand. Recent AI advances made this possible, leading him to launch Voygic.

😟 Problem

Market research is expensive and slow. A single pre-launch panel test costs around US$2,000. Live A/B tests usually cost several thousand dollars.

Brands often want to test multiple ad variations but lack the budget. Cross-cultural sellers face the highest stakes. Asian brands selling to US consumers struggle to predict American reactions due to cultural gaps.

💡 Solution

Voygic builds reusable hybrid panels by turning real consumer reactions into digital twins, including: 

• Collects answers from real people viewing video ads.
• Builds predictive models that replicate each person’s specific tastes.
• Tests thousands of ad variations against these AI personas.
• Delivers feedback with 80% of human panel accuracy.

Image credit: Voygic

📊 Market size

The global insights industry reached roughly US$153 billion in 2024. Research software outpaced traditional market research, about 11.5% to 11.8%.

The startup estimates a primary market of 5,000 cross-border companies. Capturing just 1% of these targets would generate US$5 million in annual revenue.

🤝 Team

• Jingshen Jiang. CEO. He brings a strong network of Asian market researchers from his previous career.
• Yilin Du. CTO. He previously worked on AI cloud infrastructure and published research at a top language AI conference.

🚀 Traction

• Building proof-of-concept projects with US panel companies to expand data pipelines.
• Accepted into the Techstars Tokyo accelerator program.

🏆 Competition

The closest comparable startup is Simile, which builds expensive AI personas from in-depth human interviews. Voygic requires significantly less data per person.

💰 Financials

🚩 Risks