It’s hard to imagine now, but the cofounders of Etsy — Rob Kalin, Chris Maguire, and Haim Schoppik — operated out of a tiny Brooklyn apartment in the craft ecommerce shop’s formative months. By 2007 — just two short years later — Etsy had grown to 450,000 registered sellers. Today, tens of millions of people hawk their wares on the platform, which has a market cap close to $5 billion.

In the months following Etsy’s initial public offering in 2015, artificial intelligence (AI) played a key role in the retailer’s meteoric growth. Nikhil Raghavan, vice president of product for search and machine learning at Etsy, spoke onstage at VB Summit 2018 about the value automation brought to the table.

“We made two really big bets,” explained Raghavan, who cofounded Etsy acquisition target Blackbird Technologies. “One was bringing machine learning at scale to everything from search recommendations on down. The other big one was actually investing in behavioral economics … the hypothesis was that you could figure out proxies for brands that effectively hinge on psychology triggers that people might respond to.”

Those bets paid off. Etsy’s multiyear effort to rebuild its marketplace search framework around AI bore fruit in Q3 2017, with the launch of context-specific ranking and personalization tools. Thanks to a combination of computer vision and natural language processing, Etsy’s search engine can surface results “most relevant” to buyers based on signals like the time of day and buying histories.

That’s not to say implementation has been flawless. Etsy’s algorithms learn to rank search results from hundreds of millions of historical buyer transactions, which unavoidably prejudices search results, as Raghavan pointed out.

“It might learn to keep prioritizing … something that’s hot and converting really well but possibly cheap over something that’s super high quality and one of a kind,” he said.

Etsy’s dev teams work to keep the algorithms in check, and the firm’s investing in research that promises to mitigate AI bias. (In May, it opened a machine learning center in Toronto, following on the heels of locations in Brooklyn and San Francisco.) But at the end of the day, Raghavan said, transparency is a core part of puzzle, and it helps manage expectations while making good on the promise of “bringing in and educating people about the role AI can play in unlocking value.”