Ex-Amazon VP Raised $20M to Rebuild E-Commerce with AI

What if your e-commerce site could rebuild itself in real time for every single visitor? Maju Kuruvilla, co-founder of Spangle, argues that AI makes this not only possible — but inevitable. After scaling Prime fulfillment globally at Amazon, he walked away to start a company in a basement, convinced that commerce's biggest transformation is just beginning.

Key Facts

• Traffic context gap: ~40% of e-commerce traffic originates from external sources (Instagram, Google, ChatGPT) and arrives stripped of intent context (Maju Kuruvilla)
• Reported conversion lift: Up to 50% increase in conversion and 30% increase in revenue per visit for Spangle customers (Maju Kuruvilla)
• Enterprise traction: 11 enterprise brands signed within ~10 months of coming out of stealth (Maju Kuruvilla)
• Pricing model: Success-based revenue share — Spangle only earns when it drives measurable results for brands (Maju Kuruvilla)
• Series A raised: $50 million Series A to accelerate growth and expand AI capabilities (Maju Kuruvilla)

Why 40% of E-Commerce Traffic Is Being Wasted Right Now

Roughly 40% of traffic arriving at a brand's e-commerce site originates from an external source — an Instagram ad, a Google result, a ChatGPT recommendation. Each visitor carries intent and context that brands almost universally ignore, forcing the experience to restart from zero and crushing conversion rates.

When a consumer clicks an Instagram ad for a soccer shoe, Instagram already knows a great deal about that person — their interests, the creative that caught their attention, whether they're a new or returning customer. The brand's site, however, knows almost none of this. The moment the visitor arrives, that accumulated context evaporates.

Maju Kuruvilla identified this as the core problem Spangle would solve. Rather than building a grand AI platform from day one, the team zeroed in on a specific, measurable pain point: the gap between marketing intent and on-site experience. The result is a wedge product that dynamically rebuilds the storefront to match the context each visitor brought with them — without requiring brands to overhaul their existing infrastructure.

The commercial case is straightforward. Brands already spend heavily to drive traffic through paid channels. If nearly half that traffic is underperforming because the site ignores the reason the visitor showed up, recovering even a fraction of that conversion gap is, as Kuruvilla puts it, 'practically free money.'

"When almost half of your traffic starting somewhere else has a story, has a context, and if they are coming here and if you are just completely ignoring all of that context, and trying to restart — guess what? It is going to impact consumer experience. It is going to impact the trust, it's going to impact the conversion." — Maju Kuruvilla

• External traffic sources include Instagram, Google, ChatGPT, and browser agents.
• Each source carries intent signals that standard e-commerce sites discard on arrival.
• Fixing the gap requires no changes to existing brand infrastructure.
• Results are measurable, fast, and publishable as case studies.

How Spangle's AI Merchandising Works in Real Time

Spangle's AI — internally called Product GPT — ingests real-time behavioral signals alongside enriched product data to dynamically curate the storefront for each visitor. It operates with a dual mandate: maximize relevance for the consumer while honoring the brand's commercial goals, all without manual A/B testing.

Traditional landing pages are manually created, visually optimized, and served identically to every visitor in a campaign. Kuruvilla sees this as fundamentally misaligned with where commerce is going. 'It's less about the visual elements, it's more about merchandising,' he says. The AI asks: given everything I know about this visitor, what are the right products, in what order, framed in what context?

On the signal side, Spangle reads ad campaign metadata, creative copy, targeting type (retargeting vs. prospecting), and real-time behavioral data as visitors interact with the site. On the product side, it enriches catalog items with contextual associations — occasions, cultural moments, athlete endorsements, seasonal relevance — so the AI can match products to intent rather than just keywords.

The result is a system Kuruvilla compares to 'an extremely smart merchandiser' at Nordstrom who can instantly surface five perfectly matched products for each unique shopper. Critically, the AI also carries a brand mandate: it can be tuned to prioritize conversion, push specific product lines, or support acquisition goals — giving brands control without requiring manual intervention.

"An AI is delivering and rendering this experience. The AI can have a dual mandate, where on one side it's delivering the great things for the consumer, and on the other side it's honoring what the brand or the retailer is trying to achieve." — Maju Kuruvilla

"You can enrich that soccer shoe with all the different ways people are talking about it, all the different occasions those probably are happening. So you can also enrich your products in ways that the context can be built out from the product as well." — Maju Kuruvilla

The Future of Commerce: Seller Agents vs. Buyer Agents

Kuruvilla believes commerce will ultimately operate as a negotiation between AI buyer agents — personal assistants, browsers, or platforms that shop on a consumer's behalf — and AI seller agents that represent brands. Spangle is building the seller agent layer: an AI that knows who is coming, what they want, and how to win the transaction.

The framing of buyer agents versus seller agents is one of the most clarifying mental models in this episode. A buyer agent might be a personal AI assistant, an Instagram algorithm acting on your behalf, or Amazon's recommendation engine. Each is optimizing to find you the best product. On the other side, every brand needs an equally intelligent counterpart — a seller agent that can respond intelligently to any channel, any visitor type, and any intent signal.

Kuruvilla is careful not to claim that agentic commerce will replace all other channels. 'E-commerce is only between fifteen and twenty percent of the overall commerce,' he notes. Agents will take a share of that — but emotional purchases, high-consideration decisions, and discovery-driven shopping will likely remain human-driven for the foreseeable future. The brands that win will master all modalities simultaneously.

This is why Spangle's long-term vision extends well beyond landing pages. The goal is an AI model that can 'run an entire store' — responding to a human browsing on Instagram, an agent querying via schema, or a direct visitor, with equal intelligence and brand consistency. That requires building from the ground up, not incrementally patching existing platforms.

"My larger view is e-commerce will operate most likely with a seller agent and a buyer agent. Whatever, there is going to be an agent who is responsible for shopping around and buying. And then, on the other side, you need an equally smart seller agent." — Maju Kuruvilla

"It is not going to be an incremental improvement of Shopify's template or e-commerce architecture. It is something you have to go back and build from the ground up, where you have a brain, it understands everything, it takes all the signals, it can always dynamically provide something." — Maju Kuruvilla

How Spangle Signed 11 Enterprise Brands in Under a Year

Spangle's go-to-market playbook relied on Kuruvilla's network, a free proof-of-concept offer, and a revenue-share pricing model that aligned Spangle's incentives with brand outcomes. The result: 11 enterprise brands signed within roughly 10 months of coming out of stealth, with case studies demonstrating measurable results.

Kuruvilla's first call was to people he already trusted. He reached out to former Amazon colleagues like Chris Rapp at Victoria's Secret and others across the commerce world, using those conversations to stress-test the idea and identify the sharpest version of the problem. 'You're selling even before you're building, and you're selling even while you're building,' he says.

The first customer, Uncline (part of the WHP Group), was brought in with a free POC. Spangle worked closely with their team — weekly check-ins, agreed-upon metrics upfront, rapid deployment — to generate a publishable case study. The pricing model that followed was revenue share: Spangle earns only when it drives measurable results. This structure eliminates the risk objection for enterprise buyers and keeps the startup honest.

Kuruvilla also emphasizes selectivity. Spangle deliberately limits customer volume to ensure deep partnership quality. 'We handpick our customers,' he says. Early enterprise customers are treated as co-development partners, and that relationship depth is what drives expansion into additional use cases over time.

"We give a free POC, being a great partner as an innovation partner with us. But right after, we made some of the metrics. Initially, it was a rev share model. We wanted to make sure that our pricing reflects a joint success." — Maju Kuruvilla

• Start with warm network connections to validate and close first customers.
• Offer a free POC to remove risk and generate a publishable case study.
• Use revenue-share pricing to align incentives and reduce buyer hesitation.
• Limit early customer count to maintain partnership quality and learn deeply.

Why Product Market Fit Is Now a Daily Moving Target

In the AI era, product market fit is no longer a milestone you reach and hold — it's a spectrum you must actively maintain. As Kuruvilla puts it, you're 'one deployment of OpenAI or one Anthropic release away from dying.' The key variable for founders today is speed, not just product strength.

The concept of product market fit originated in a SaaS world where building something durable enough to sell repeatedly for five to ten years was the goal. Kuruvilla questions whether that framing still applies. In an environment where foundational AI capabilities shift on a weekly basis, the window between achieving fit and losing it has compressed dramatically.

Host Pablo Srugo frames it well: the level of adaptation required today is categorically higher than it was even three years ago. A product that was frontier-capable in January may be table stakes by April. The founders who survive are those who treat relevance as an ongoing operation, not a destination.

For Kuruvilla, this doesn't mean abandoning long-term vision — it means holding that vision tightly while staying paranoid about near-term relevance. Spangle's wedge product is repeatable and proven, but he openly admits they won't consider themselves truly PMF until the AI can run an entire store autonomously. That goal keeps the team moving.

"I don't know whether the term product market fit is fully appropriate now. Because now you have to find your relevancy every day. You know, you're one deployment of OpenAI or one Anthropic release away from dying." — Maju Kuruvilla

"You start something small, and then you get distracted, and then your customers will start pulling you into directions where it is not aligned with where you want to go long term. And that's where you need to have high conviction of your vision." — Maju Kuruvilla

Traditional E-Commerce vs. AI-Powered Dynamic Commerce