E-commerce brands now spend roughly 40% more to acquire each new customer than they did in 2023, yet the website experience that greets those hard-won visitors has barely changed: the same static pages, the same overnight batch recommendations, and no way to recognize a first-time shopper from anyone else. The cost of that gap compounds fast: the average e-commerce brand loses $29 per new customer after accounting for marketing and returns, meaning every unconverted session deepens the loss. The problem is structural: legacy personalization platforms were built around logins and cookies, so roughly 90% of any site’s traffic gets a generic storefront regardless of what their behavior signals in the moment. Malachyte addresses this gap at the infrastructure level, using proprietary two-headed vector AI to build a live behavioral profile for every shopper from the first click, without requiring a sign-in, a cookie, or any prior purchase history. That profile updates continuously across search, recommendations, and product pages during a single session, applying the same architectural approach the team developed while building Spotify’s personalization infrastructure across 800 million users and a billion-item catalog to the specific economics and cadence of retail.
AlleyWatch sat down with Malachyte Cofounder and CEO Siddharth Motwani to learn more about the business, its future plans, recent funding round, and much, much more…
Who were your investors and how much did you raise?We raised $10M in seed funding. The round was co-led by Bessemer Venture Partners and Gradient Ventures, with participation from Harpoon Ventures.
Tell us about the product or service that Malachyte offers.Malachyte offers Behavior Intelligence Infrastructure for e-commerce.We build a live profile of every shopper — including the roughly 90% who never log in — from what they do in the moment: what they search for, click on, compare, and skip past. That profile updates continuously and powers search, recommendations, and product pages, so the experience adapts while someone is still on the site rather than after they come back. For brands, it shows up as revenue per visitor. Their merchandising team also gets no-code control over the live model, so they can promote inventory or change what it optimizes for and see the effect in minutes instead of after tonight’s batch job. It installs on Shopify in about 30 days with no replatforming.
What inspired the start of Malachyte?Ian Anderson (cofunder) and I spent years at Spotify focused on one problem: proactively predicting what a user actually wants to listen to right now based on their intent and likely next action, rather than purely on their listening history. With 800 million users and a billion-song catalog, solving that meant building a different kind of recommendation technology built on two-headed user vectors.The two-headed model indicates two things about a listener at once:
Long‑term taste (slow head): what they generally gravitate to over weeks and months, tuned toward keeping them subscribed and discovering new music for years (the metric that matters for lifetime value).
In‑the-moment intent (fast head): what they’re actually doing right now through their behavior on the app, which can shift by the minute.
The hard part was reading both, simultaneously, and being able to adapt fast enough to matter.Solving this led to an interesting insight. The same way the two-headed vector reads intent for known listeners, it also accurately predicts the intent for a brand new listener with no history. In e-commerce, this is known as the “cold start” problem.Armed with this new approach to cold start, Ian and I started talking about our own personal, digital experiences and where they felt frustrating or broken. Retail stood out immediately. It’s the one large category where most of your traffic is a stranger and nobody had built anything to handle cold start in real time. Instead, what retailers had was batch technology: a catalog and a set of rules refreshed overnight, not a system reacting to what’s happening on the site right now. The idea for Malachyte was born, which was to apply a two-headed user vector in a novel way.
How is Malachyte different?While we weren’t trying to solve for “cold start” at Spotify, it turns out that our novel application of user vector technology not only reads long-term taste, but also predicts in-the-moment intent. This is exactly what is needed for e-commerce use cases, preference and intent prediction based on real-time behavior. We call it behavior intelligence and it beats demographic and other forms of profiling for anonymous users every single time.The challenge is that commerce signals are not captured; most e-commerce platforms are built to track identity logins, cookies, and purchase history, not real-time behavior. The stakes are also higher for e-commerce than for music. In this industry, a wrong recommendation costs a sale, not just a skipped track. Inventory and pricing shift daily, and shopping has a purchase goal, so optimizing for a different level of engagement is essential. In e-commerce, visits that are weeks or months apart instead of back-to-back days or hours apart require a robust system that learns.Unlike most AI solutions, we are not using a large language model. We’re not predicting the next word in a sentence. We’re using vector AI to predict the next product a shopper wants, from real-time behavior. We leverage our own proprietary two-headed transformer; one head is a slower base model that learns a shopper’s general taste over time. The other fine-tunes continuously on what they’re doing in this exact session. Most personalization systems have a version of the first, but nothing like the second, which is why they require a login to know who you are, but lack insight into what you’re doing right now. This is the hard part: training a model that learns in real time and serving it to every shopper fast enough that nobody notices.Point-solution vendors (e.g. Bloomreach, Algolia) are solving real, adjacent problems like search and merchandising. We don’t think we’re in the same category as them, and a feature-by-feature comparison undersells what we’re actually building. Most require user login or cookies to personalize results and, frankly, consumers are just tired of the barrage of pop-ups and spending half their shopping experience x-ing out of them or wondering what personal info is being tracked without their knowledge.We built Malachyte so any brand can get access to cutting-edge technology that is trained on their user traffic without having to invest tens of millions of dollars in computing and building deep expertise in behavior intelligence.
What market does Malachyte target and how big is it?We sell to consumer e-commerce brands and retailers, from mid-market DTC through large established retail. The immediate market is what brands already spend on search, recommendations, and personalization software, which is a multibillion-dollar category — but that understates it, because those are line items on a website budget. What we’ve built sits underneath merchandising and, over time, the marketing spend as well, which is a far larger pool. The shift toward AI agents handling routine buying makes real-time behavioral understanding an infrastructure rather than a feature; McKinsey has put agent-orchestrated retail spend at $3–5 trillion by 2030.Beyond e-commerce, we see other use cases in travel, elearning, streaming, and others, where this behavior intelligence infrastructure will be crucial in the future.
What’s your business model?Malachyte has a subscription software model making user vector technology available for e-commerce. To accelerate time-to-value, we offer native integration on Shopify with new integration partners coming soon, including Salesforce Commerce Cloud and others. Most of our customers are live within a matter of days. Brands can also use our API for custom integrations.Because Malachyte replaces several disparate point solutions — search, recommendations, and new use cases we’re building now — the effective cost tends to be lower than the stack it displaces.
How are you preparing for a potential economic slowdown?Candidly, our category tends to do better in one. When acquisition gets expensive, brands stop trying to buy more traffic and start trying to get more out of the traffic they already have, which is exactly what we do. When consumer spending starts to drop, it becomes even more essential that brands leverage technology to ensure a strong revenue per visitor. We believe Malachyte’s tech will soften the blow of a downturn and help some brands survive the storm.
What was the funding process like?Faster and more focused than I expected, mostly because the investors who moved had already formed a view on this shift. Bessemer had published on agentic commerce being the next step and function change in consumer behavior, and Gradient is Google’s AI fund, so we weren’t spending meetings explaining why real-time behavioral infrastructure matters. The conversations went straight to the two questions that actually mattered: whether this specific team could build it, and whether the early production results were real. That kept the process tight rather than sprawling across dozens of firms. We also had so much internal momentum from existing investors that generated buzz and great initial metrics from pilots with our first few customers.
What are the biggest challenges that you faced while raising capital?Explaining a category that doesn’t have a name yet. The instinct is to describe us as a personalization or recommendations company, because those are the words people already have — but that drops us into a bucket of point solutions and undersells what we’ve built. The technology we’re commercializing runs in production at only a handful of companies in the world presently; Spotify and TikTok are two of them. Getting someone from “so it’s a recommendations tool” to “this is infrastructure” inside a single meeting took real work. What consistently moved people wasn’t the pitch, it was the results on first-time, anonymous visitors — because everyone in the room knew that’s the part nobody else solves.
What factors about your business led your investors to write the check?Three factors.First, the team had built this exact system before: Ian led the creation of the user vector to unlock Discover Weekly as a prominent personalization experience, and together we pioneered the technology into Spotify’s user representation platform (the advanced user vector), so this wasn’t a research bet. Along with our co-founder Shivaditya Sinha (COO), who brought the operational knowledge we needed to execute, we made an undeniable, experienced team.Second, the approach is documented in published research rather than resting only on our own claims. We worked with Stanford University early on to ensure we could bring this technology to e-commerce effectively.Third, and most important, we had production results before the raise — including a head-to-head against a customer’s existing stack where our largest gain came from brand new visitors the system had never seen before. That last one carried the most weight, because it was evidence that the hardest technical claim we make is actually true.
What are the milestones you plan to achieve in the next six months?Believe it or not, brands aren’t using their most valuable dataset: what customers actually do inside their own shopping experience. 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. We read it continuously, so each action makes the user’s vector more confident about both preference and current intent. The second underused layer is context: time of day, device, and where someone arrived from. A phone visitor at 11pm from an email link is in a different state of mind than the same person on a laptop mid-morning, and most systems treat them identically. None of this requires identifying anyone. It’s how someone behaves, not who they are.
Believe it or not, brands aren’t using their most valuable dataset: what customers actually do inside their own shopping experience. 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. We read it continuously, so each action makes the user’s vector more confident about both preference and current intent. The second underused layer is context: time of day, device, and where someone arrived from. A phone visitor at 11pm from an email link is in a different state of mind than the same person on a laptop mid-morning, and most systems treat them identically. None of this requires identifying anyone. It’s how someone behaves, not who they are.
First, we will be incorporating signals that no brand is capable of personalizing yet. So much of the most valuable context comes from what the user was seeing and doing before they arrived, like the search or ad that brought them to the site. Personalizing that first impression based on where someone came from is next.
What advice can you offer companies in New York that do not have a fresh injection of capital in the bank?Sell before you raise. We spent 2024 testing the technology with more than twenty enterprises across travel, grocery, and retail before we settled on e-commerce, and that did two things: it told us where the problem was most acute, and it meant that, when we did raise, we were talking about customers instead of hypotheses. The other piece of advice is to be honest with yourself about which of your claims is genuinely the hard one, and go prove that one first — investors and customers both quietly discount everything else until the central claim is demonstrated. New York City helps here, too. The customers are physically close, and you can get in a room with them in a way you can’t from most places.
Where do you see the company going now over the near term?Longer term, the same real-time understanding of a shopper that decides what they see should also inform what a brand spends to reach them. Those run as separate systems today, on a much coarser signal, and they shouldn’t. Bringing merchandising and marketing onto one behavioral profile is where this goes next. It’s gold for brands and retailers.While we focus on retail today, we’ll also expand across digital categories, including travel, finance, gaming, and more.
What’s your favorite summer destination in and around the city?It’s hard to pick just one! Weekdays it’s Madison Square Park when I need twenty minutes of not looking at a laptop, and Central Park when I need considerably more than twenty. Weekends are usually Greenpoint, which has quietly become my favorite corner of the city. Then there’s the escape tier: Jones Beach when I want the ocean without much effort, hiking around Storm King when I want to feel like I actually left New York, and I have never once said no to a weekend in the Hamptons.
You are seconds away from signing up for the hottest list in NYC Tech!
Sign up today
