Malachyte Turns Spotify’s Skip Signal Into Merchants’ Agentic Edge

Spotify solved a problem that most digital merchants still haven’t: how to understand who shows up to shop but prefers to do that without registering or checking in.

For Spotify, an anonymous listener arrives with no profile, no playlist history and no declared preference. Spotify must decide what to play next and get smarter and smarter before that listener leaves so they’ll come back and maybe turn into a paying subscriber or continue to show up and be an attractive target for advertisers.

Now swap the song for a SKU and the playlist for a product page. That’s the application of the Spotify recommendations engine Malachyte Co-Founder and CEO Sidd Motwani is applying to commerce. He told PYMNTS CEO Karen Webster that’s a no-sale, no-conversion problem for a merchant without any digital footprints to action. In agentic commerce, it becomes a lot more problematic than that.

The Skip Is the Signal

Motwani and his co-founders, Ian Anderson and Shivaditya Sinha, were part of the Spotify team that built the behavioral intelligence infrastructure behind its recommendation engine. That infrastructure now powers more than 90% of Spotify’s recommendations across 800 million users and 1 billion items, he said. In more practical terms, it’s the operating logic behind the company’s newly funded, $10 million bet on agentic commerce.

One of Spotify’s challenges was to assemble a customized playlist for an unregistered listener, or the person the platform couldn’t recognize through a subscriber profile. The answer wasn’t to wait for a registration. It was to read intent from behavior, or what the listener skipped, saved, replayed, reordered or removed, and in what sequence.

At Spotify, the skip mattered because it wasn’t always a “I hate that song” verdict, Motwani said.

“It’s just about the context in which you’re skipping,” he said.

A song rejected on a morning run may be exactly right for the drive home. The signal has meaning only inside the moment that produced it.

Retail throws off the same kind of digital behavioral exhaust, Motwani said. A shopper scrolls past one item, hovers over another, checks a size, changes a price filter, clicks, backs out and adds something else to a cart. Malachyte reads those actions as a sequence, not as isolated declarations of taste.

Motwani compared the session to a sentence. Each action is a word, and order supplies the meaning.

It’s also how he said Malachyte can address retail’s cold-start problem.

“I don’t need any history about you,” Motwani said. “I can understand this current situation in under 200 milliseconds.”

The first prediction comes from patterns across prior sessions, and each new action sharpens the current one.

In other words, the storefront is built, rebuilt and tweaked as the shopper moves, Motwani said. Search results reorder. Product listing pages reshuffle. Recommendations, banners and other content change.

“When Malachyte’s powering that store, the shelves are personalized as they’re coming up to you,” Motwani said. “As you’re engaging with the products, it’s getting more refined.”

Personalization Is Nothing Without Orchestration

It’s easy for people to conclude that Malachyte is just another widget in the portfolio of widgets retailers already have and maybe don’t need more of, Motwani said. That’s a risky assumption since Malachyte is an operating layer. The shopper’s inferred intent is one input. Inventory, margin, promotions, likely size, the ad or search term that delivered the visit, and the merchant’s own priorities are others. The result is a product model that makes sense for this shopper, in this session, from this merchant, right now.

That doesn’t mean that merchants must give up control to a black box of inexplicable assumptions.

“[Merchants] literally have the ‘sliders,’” Motwani said.

A retailer can elevate a new collection or a bestseller, suppress low-inventory merchandise, change the weight of a promotion or steer a category differently as conditions change.

That distinction matters because a merchant isn’t Spotify. The merchant is optimizing for the next click at the same time it is managing inventory, margin, brand and customer experience. Behavioral intelligence becomes valuable when it can serve those commercial constraints without making the storefront feel manipulated, mechanical or not in the best interest of the shopper.

Meanwhile, a system that takes every signal too literally can become efficient at showing shoppers more of what they already saw, and bad at helping them discover something they didn’t know they wanted. Conversion goes up on paper while the store loses its magic.

Webster put the tension to Motwani directly.

“How do you program serendipity into the experience?” she asked, adding that one of her own recent shopping trips started with a search for a blue blazer and ended with a pink skirt. The pink skirt was not a failure of intent. It was the point of shopping.

The system must distinguish directed shopping from discovery, Motwani said. A shopper broadcasting a clear objective can see a tighter assortment, complementary products or a faster path to the right item. Someone still exploring should see a wider aperture. Malacyhte’s job is to understand what that click means during this visit and leave enough room for the unexpected find.

The Agentic Edge Is the Consideration Set

Agentic commerce raises the stakes because the digital shelf may disappear entirely. An AI agent can arrive with a consumer’s instructions, search across multiple merchants and return a short list. The shopper may never see the products the agent rejected, or even know which stores were considered. A merchant can have the right product at the right price and still lose the sale if the agent never lets it onto the screen.

That makes the consideration set the new front door. Merchants still need to orchestrate the experience for shoppers who browse directly, but they also need inventory, product attributes, pricing and availability to be intelligible and relevant to the agents acting for those shoppers. The agentic edge is making sure any agent can understand why a merchant’s product belongs in the answer.

Motwani said he expects human browsing and agent-led shopping to coexist.

“It doesn’t matter whether there’s an agent or a human coming in,” he said. “We would be serving that at the infrastructure layer.”

That layer is where shopper intent and merchant intent meet before a human sees the shelf or an agent builds the shortlist.

The hard part isn’t a shortage of data, Motwani said. Retailers have accumulated mountains of it, often in silos that make real-time use harder, not easier. The hard part is turning those data points into a decision quickly enough to shape the next product, page or agent response while the shopping session is still alive.

That’s why Malachyte’s Spotify lineage matters, Motwani said. The company’s wager isn’t that shopping is just another playlist, but it shares the common foundation that it’s possible to infer intent from behavior, update continuously, preserve context and help merchants win two customers at once. Those customers include the human with the need and the agent deciding which merchant gets the chance to meet it.

Put more precisely, personalization tries to predict what a known customer might want someday. Behavioral intelligence tries to decide what an unknown shopper and that shopper’s agent need next. That’s a bigger job, and potentially a more valuable edge.

Watch the full interview with Malachyte’s Sidd Motwani to learn more about:

• How Malachyte infers price, size and product preferences from behavior without requiring a login, cookie or stated preference.
• Why merchant-controlled “sliders” matter when inventory, promotion and merchandising priorities change.
• How retailers can preserve serendipity for human shoppers while making products legible to the AI agents assembling their consideration sets.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.