Spotify AI Team Raises $10M for E-Commerce Makeover

A team that built the AI responsible for 90% of Spotify's recommendations has raised $10 million to bring the same intense, real-time personalization to online shopping. The move signals a fundamental shift in e-commerce AI, away from historical data and toward predicting a shopper's current intent according to TechCrunch.

Founded by ex-Spotify engineers Sidd Motwani, Ian Anderson, and Shivaditya Sinha, the startup Malachyte argues that the current state of online shopping feels primitive. Most personalization hinges on what’s in your past purchase history or your logged-in profile. The result is a system that treats a late-night phone browser and a daytime laptop researcher identically, failing to adapt within a single session. Vector AI was responsible for shaping music discovery for 800 million Spotify users. Now, it's being retooled to transform product discovery.

Why an ex-Spotify Team Thinks Your Shopping Should Feel Like Your Perfect Playlist

Music discovery today is nearly magical. A service like Spotify doesn't just play a song you liked six months ago. It learns from your skips, your replays, your explore sessions, building a model of your "taste" that can recommend a new artist you'll love before you've even heard of them. The founding team of Malachyte believes the same should happen with products.

“Every hover, click, scroll, search refinement and add-to-cart is a signal, and most systems either never act on it in the moment or aggregate it into a segment overnight,” CEO Sidd Motwani told TechCrunch. “We read it continuously, so each action makes the user's vector more confident about both preference and current intent.”

The core problem they're tackling is one of state of mind versus static data. Retailers have terabytes of transaction data, but that data is backward-looking. It doesn't capture whether a visitor today is casually browsing for inspiration or urgently trying to solve a specific problem before a deadline. The $10 million seed round, co-led by Bessemer Venture Partners and Gradient, is a bet that applying media-grade behavioral intelligence to commerce can unlock massive value by addressing that gap. It’s a move that, similar to innovations in fintech, requires redesigning core logic from the ground up rather than incremental improvements.

The Secret Ingredient Is 'Taste', Not Just a Shopping Cart

In the Malachyte framework, "taste" is the cohesive model of a user's long-term preferences, aesthetic choices, and even latent desires. It's what allows a music service to recommend a punk-rock song to a classical listener because it detects a consistent preference for complex, rebellious compositions.

“A search for 'heavy-duty boot' followed by two clicks on steel-toed boots is enough to move work pants and gloves up the page and push dress shoes down, with no account or history required. Every additional action sharpens the profile, so the experience gets more relevant the longer someone stays, and again on their next visit,” explains Motwani.

This stands in sharp contrast to standard e-commerce AI, which operates transactionally. If you buy a certain pair of boots, the system will recommend similar boots or a product frequently co-purchased with those boots. It's a model of similarity, not a model of you. A "taste" model, built in real-time, aims to understand your style so it can accurately recommend goods in categories you've never bought from before. This shift mirrors how financial AI is being used to replace legacy systems with intent-focused models, moving beyond rigid rules.

How An AI Learns Your Style From Clicks, Pauses, And Exits

The system depends on a continuous data diet far richer than purchase events. The technology, dubbed “two-headed Vector AI,” treats behaviors as sequential signals:

• Dwell Time: Lingering on a product detail page versus bouncing off instantly.
• Visual Flow: The order and speed of scrolls through a page.
• Pattern Recognition: If a user consistently browses hemp, linen, and organic cotton items, the AI infers a preference for sustainable materials, even if the user never searches for "eco-friendly."

This continuous learning loop means the model updates with each new interaction, refining its understanding of both long-term preference ("this person gravitates toward minimalist design") and immediate intent ("they seem to be assembling a professional workwear wardrobe tonight"). The goal is to achieve what Spotify does: a user’s actions in the first 30 seconds of a session make the next 30 seconds more relevant. It’s a dynamic process that makes static demographic or purchase-history segments look instantly obsolete.

The Challenge: Making Shopping 'Sticky' Without Feeling Creepy

Deep personalization walks a tightrope. There’s a thin line between "this is perfect for me" and "this AI is stalking me." As with any system that learns from user behavior, the risks of privacy intrusion and algorithmic entrenchment are real.

User Control: For this to be adopted at scale, platforms must give users visibility and control. How explicit is the "taste" profile? Can a user reset or tune it? Malachyte has not detailed its user-facing controls, but this will be a critical adoption factor.

Filter Bubbles: In music, a recommendation that’s too perfect can prevent discovery of new genres. In commerce, it could trap a user in a repetitive loop of similar items, missing spontaneous or novel products they might love. The Business Case: For retailers, the value proposition is compelling: stickiness. A site that feels intuitively curated keeps visitors engaged longer, increases conversion rates, and builds customer lifetime value far beyond a single transaction. It transforms browsing from a chore into a discovery experience, which is the entire goal of modern e-commerce.

What Real Retailer Adoption Will Look Like And What Could Go Wrong

Malachyte’s platform is already live. It launched with Fun.com in Fall 2025 and became generally available to Shopify merchants in June 2026 via a native integration. Larger brands can use its API. The practical result should be a browsing experience where product grids feel curated, search results are deeply contextual, and categories adapt to the user’s perceived mission.

Adoption hurdles for retailers remain significant:

• Integration Burden: Tapping into real-time user behavior requires deep, often messy, integration with existing site architecture and data silos. It’s not a plug-and-play widget but a core logic layer.
• Algorithmic Bias: If the training data or the interaction signals are biased, the system could inadvertently steer users toward homogeneous recommendations, reinforcing stereotypes (e.g., pushing only certain toy types based on perceived gender signals).

The ultimate test of this experiment isn't just the AI’s learning speed, but its understanding of human unpredictability. People shop on whim, for gifts outside their own taste, and click on things they later dismiss. The brilliance of a great recommendation engine is its ability to distinguish true preference from momentary curiosity. Malachyte's success depends on harnessing the relentless learning of systems like those that boost productivity while preserving the serendipity that makes discovery human.

If it works, online shopping could transform from a search-and-filter marathon into a guided, evolving conversation between shopper and store. If it fails, it will be another overhyped AI tool that crumbles under the complexity of real human behavior.

Why This Changes Everything

• This technology could dramatically increase e-commerce conversion rates by predicting purchase intent in real-time rather than relying on historical data.
• Consumers will experience shopping recommendations as personalized and dynamic as Spotify playlists, making discovery more intuitive.
• The $10M investment signals strong market belief that AI personalization can fundamentally transform the $6T global e-commerce industry.