The Next Wave of E-Commerce Startups are Rethinking Real-Time Personalization with AI

It's e-commerce's greatest irony: Brands invest heavily to pinpoint their ideal customers, only to greet them with the same generic storefront as everyone else. The chasm between intelligent acquisition and cookie-cutter experiences sabotages conversion funnels and deflates marketing ROI.

Maju Kuruvilla, Founder and CEO of Spangle AI, ex-Amazon VP, and former CEO of Bolt, makes the case that generative AI isn’t just the next phase of e-commerce; it’s the long-overdue fix for a system that’s been broken for decades.

• A long time coming: "Personalization has been a buzzword for two decades, but frankly, it's been a massive disappointment—no one truly delivered," says Kuruvilla. "Generative AI is the breakthrough that changes everything. We finally have the power to reimagine the entire e-commerce experience from the ground up," he says. His new venture, Spangle AI, which recently emerged from stealth mode with backing from investors like Madrona and Streamlined, is built to tackle some of e-commerce’s hardest problems using AI.
• Where clicks go to die: "Around 40% of the traffic coming to merchants is from paid or marketing channels," Kuruvilla notes. "Those channels are very targeted and very specific—but they all get funneled into the same spot, the same experience." Spangle’s fix? A seamless handoff from click to conversion. "We create a continuous experience based on where they’re coming from to drive more conversion, more engagement, more revenue." That means real-time, self-optimizing landing pages tailored to each user’s entry point and intent.
• For all intents and purposes: At the heart of Spangle’s tech stack is Product GPT, a proprietary model built specifically for retail, starting with apparel. "We built this model using the latest technologies to solve a simple but critical problem: helping customers find the right product," Kuruvilla explains. Unlike past personalization efforts—which were often "backward looking" and limited by rigid site structures—Product GPT is designed to anticipate intent in real time.

Kuruvilla notes a familiar misstep: "Imagine you buy a TV. You come back, and they show you another TV," he says. "That's not helpful when I'm shopping for my summer vacation in Europe. In fact, it's kind of offensive." That kind of misfire is exactly what GenAI is built to eliminate. "Now, with a fundamental model like this, you have an opportunity to really reimagine the space," says Kuruvilla.