Kevin Lee speaks with Aniket Deosthali, of Envive
AI, about how AI agents are transforming e‑commerce, merchandising, and
discoverability.
Aniket explains that Envive AI builds self‑improving
storefronts: e‑commerce sites that learn from every interaction to sell better
each day. Unlike traditional approaches focused purely on human shoppers,
Envive is designed for a world where the “shopper” is increasingly an AI
agent (ChatGPT, Gemini, agentic browsers, etc.), not just a person
clicking around a site. Their system is platform‑agnostic, working across
Shopify, Magento, Commerce Cloud, and others, with 70% of current customers on
Shopify.
A key theme is moving beyond “people who bought this
also bought that.” Traditional wisdom‑of‑the‑crowds recommendations
require huge data sets and struggle with new products. Envive’s “AI brain” adds
a reasoning layer: it tries to understand why someone is
buying something and uses that context to recommend complementary items. For
example, a shopper choosing a non‑toxic pan is likely health‑conscious, so
suggesting non‑toxic storage containers or related products makes sense. This
reasoning‑driven personalization consistently boosts AOV (average order
value), basket size, and revenue per visitor (RPV) in a way that
feels thoughtful rather than intrusive.
Kevin and Aniket discuss how most storefronts are still
designed for human eyeballs (hero images, video, branding) while agents only
“see” the underlying data: structured attributes, copy, content architecture,
and UGC. That creates a dual challenge: brands must now optimize for both
humans and agents, similar to how the mobile shift forced everyone to rethink
their sites. This leads into the idea of GEO (generative engine
optimization): instead of short keyword queries, AI answer engines deal with
long, conversational prompts and refinement queries, making long‑tail,
robot‑friendly content and strong first‑party data more important than
ever.
They also explore vectorization and data schemas: how
all product descriptors and content become mathematical representations that
allow answer engines and agents to understand adjacencies and relevance.
Marketers need to think about personas, use cases, and the full range of ways
people might describe their needs, then express that in content that LLMs can
ingest and interpret.
Another major theme is AI‑driven meritocracy. Kevin argues that as AI systems ingest more UGC, reviews, and real‑world feedback, it becomes increasingly difficult for weak products to hide behind great marketing. Negative experiences feed back into the models just as much as positive ones. Aniket agrees, framing AI as accelerating the internet’s convergence toward “truth” and rewarding provable performance. Good products, strong experiences, and authentic UGC become long‑term competitive moats, while bad products are penalized over time regardless of ad spend.