Ex-Spotify Team Raises $10M to Rebuild Personalization for Online Shopping

Updated August 6, 2026 5:54 pm

Three former Spotify engineers have raised $10 million to apply the recommendation technology they helped build at the music-streaming giant to online retail. Their new startup, Malachyte, is betting that e-commerce can become far more effective if stores respond to what shoppers are doing in the moment, not just what they bought last month or how they were segmented overnight.

The seed round, announced Thursday, was led by Bessemer Venture Partners and Gradient, with Harpoon Ventures also participating. Malachyte plans to use the capital to expand distribution and add product and commercial talent as it moves from early deployments into broader adoption among merchants.

The company was founded by Sidd Motwani, Ian Anderson and Shivaditya Sinha, who previously worked on Spotify’s behavioral intelligence stack. At Spotify, they helped develop a system called Vector AI, which is designed to infer what a listener is likely to do next rather than relying only on their historical behavior. According to the company, that system informs roughly 90% of Spotify’s recommendations for its hundreds of millions of users.

Malachyte is built on the same core idea: people reveal intent through real-time behavior, and those signals can be more useful than static customer profiles. In retail, the startup argues, that means a shopper browsing with a specific goal should see a very different storefront than someone casually exploring categories.

Why the founders think e-commerce needs a new personalization model

Malachyte’s pitch begins with a familiar problem in digital retail. Most commerce platforms still depend on broad demographic buckets, purchase history, or login-based customer records to decide what products to show. That approach can work for returning buyers, but it leaves first-time visitors with bland, generic experiences and often misses shifts in intent from one session to the next.

The founders say that traditional systems are slow to react because they are built around historical data. If a shopper bought hiking boots six months ago, they may keep seeing outdoor gear even when they now need workwear, a gift, or something completely different. In Malachyte’s view, that gap is one of the biggest missed opportunities in retail personalization.

Motwani, who is now Malachyte’s chief executive, said the company’s system begins forming a picture of the shopper almost immediately after a page loads. He described it as a model that can infer both general preference and current intent within a single browsing session, even if the visitor has never logged in before.

“[Our] system starts forming before the first click, using the context available the moment the page loads,” Motwani said. “Within a single session, we build a real read on both preferences and what someone is trying to accomplish right now.”

He pointed to simple examples of how the product is meant to work in practice. A shopper who searches for heavy-duty boots and then clicks on steel-toed styles can be shown work pants and gloves sooner, while dress shoes are pushed lower in the ranking. As the shopper continues browsing, the model keeps updating, improving recommendations not only for that session but also for future visits.

How does Malachyte’s “two-headed Vector AI” work?

Malachyte says its platform uses what it calls “two-headed Vector AI” to handle two separate but related jobs at once: predicting what a customer is trying to buy next and learning the shopper’s broader taste profile. The company says this lets merchants tailor experiences more precisely than systems that only match products to past purchases.

One head of the model is aimed at immediate intent. The other is meant to capture longer-term preference patterns. Together, the system is supposed to respond to each action a shopper takes and keep refining recommendations in real time.

According to Motwani, the platform is designed to treat every interaction as a signal.

“Every hover, click, scroll, search refinement and add-to-cart is a signal,” he said, adding that many current tools either ignore those actions in the moment or group them into coarse audience segments after the fact. “We read it continuously, so each action makes the user’s vector more confident about both preference and current intent.”

The company also argues that context is a major blind spot in retail systems. A person shopping on a phone late at night after clicking through from an email is often in a very different mindset from the same individual browsing on a laptop during the workday, yet many personalization engines treat those visits the same way.

Motwani said a customer browsing at 11 p.m. from an email link should not be understood the same way as the same person shopping on a desktop in the middle of the morning, because the circumstances surrounding the visit shape intent as much as the account history does.

What has Malachyte already built?

Malachyte says it has been developing and testing the underlying technology since 2024. Before narrowing its focus to e-commerce, the company worked with more than 20 enterprise customers across travel, grocery and retail, suggesting that the founders were initially exploring a broader set of use cases for their behavioral modeling approach.

The startup first went live in the fall of 2025 with Fun.com, giving it an early proof point in online retail. Since June 2026, the platform has also been generally available to Shopify merchants through a native integration, while larger retailers can connect through an API.

That distribution strategy matters. Shopify access gives Malachyte a route into thousands of merchants who want better personalization without building the technology in-house. The API option gives larger brands and enterprise retailers more flexibility to integrate the system into existing commerce stacks.

Why investors are backing intent-aware retail tech

The funding round suggests that investors see room for a new generation of commerce software focused less on static segmentation and more on live behavioral signals. Bessemer Venture Partners and Gradient led the round, while Harpoon Ventures joined in support.

Retail personalization has long been a crowded category, but much of the market still relies on recommendations based on prior transactions, product affinity, or broad audience labels. Malachyte’s founders are positioning their company as a more dynamic alternative, one that can adjust as soon as a shopper’s behavior changes.

That positioning may appeal to merchants facing pressure to improve conversion rates, increase basket size and reduce the friction that comes from irrelevant product suggestions. In a retail environment where acquisition costs are high, showing the right product at the right moment can have an outsized effect on revenue.

At the same time, the company’s roots at Spotify give it a strong technical story. The founders are not simply applying generic AI language to commerce; they are translating a recommendation infrastructure that already powers a major consumer platform. That background may help the company stand out in a crowded field of AI startups promising personalization.

What could this mean for online stores?

If Malachyte works as advertised, stores could begin serving different product mixes depending on what shoppers do second by second, rather than relying on yesterday’s data. That could change homepage merchandising, search ranking, email-to-site experiences and product discovery throughout the session.

For merchants, the practical upside is obvious: better relevance, potentially higher conversion and fewer missed opportunities when customer intent changes. A visitor who arrives looking for one item may leave with a broader, more useful basket if the store recognizes the shift quickly enough.

For shoppers, the experience could feel more intuitive and less repetitive. Instead of seeing the same recommendations on every visit, the storefront would adapt to context, device, browsing path and real-time signals inside the session.

Still, the success of the model will depend on execution. Real-time personalization can easily become noisy if the system misreads intent, overreacts to brief browsing patterns or surfaces products that feel intrusive. As with any recommendation engine, accuracy and restraint will matter just as much as speed.

Where the bigger opportunity may lie

Motwani says the long-term opportunity is not just better recommendations, but a closer link between merchandising and marketing. In his view, both functions should rely on the same behavioral understanding of the customer instead of operating from different data sets and different assumptions.

That could have wider implications for how retailers structure their digital storefronts. If the same live intent model can influence homepage curation, category ranking, promotional banners and email follow-up, it may reduce the gap between what shoppers are trying to do and what stores are trying to sell.

How Malachyte fits into the broader AI commerce wave

Malachyte is arriving at a moment when retailers are increasingly willing to adopt AI tools that promise measurable gains in conversion and customer engagement. Many of those tools focus on content generation, support automation or internal productivity. Malachyte’s approach is different because it targets the core shopping experience itself.

That makes the startup part of a wider shift in AI: from generalized assistants to specialized systems that act on highly specific business problems. In commerce, the problem is straightforward but difficult to solve well — understanding what a shopper wants right now, not just who they were yesterday.

There is also a clear strategic advantage in starting with e-commerce. Online stores generate rich behavioral data at scale, from search terms and clicks to cart additions and page depth. That gives a model like Malachyte’s plenty of signals to learn from, especially in sessions where the customer never signs in.

The startup’s challenge will be proving that its approach consistently outperforms more established recommendation engines. Retailers already have many options, and switching systems can be expensive and operationally risky. Malachyte will need to show that its real-time model can deliver enough improvement to justify the integration effort.

Key facts at a glance

What happens next for Malachyte?

The immediate next step is likely execution: hiring, expanding sales and proving that the platform can improve merchant performance across more stores and categories. The new funding gives the startup the resources to move beyond early adopters and build a more substantial commercial footprint.

It will also need to convince merchants that intent-aware personalization is worth the investment. The strongest argument may be that shoppers rarely behave like neat historical profiles. They arrive with shifting motivations, different devices and changing needs, and a system that adapts to those realities may produce better commerce outcomes than one that only remembers the past.

For now, Malachyte is making a broader bet on the future of retail AI: that the best product recommendation engine is not the one that knows your history best, but the one that understands what you are trying to do this minute.