How Koddi’s Semantic Search redefines relevancy for commerce media

At Koddi, artificial intelligence is embedded in every step of how every ad decision is made. From forecasting and bidding to quality scoring and pacing, AI operates in real time across the entire delivery pipeline, ensuring that every impression is optimized for both performance and user experience.

Koddi has long ensured sponsored placements align with each retailer’s native shopping experience. Semantic Search takes this a step further by delivering relevance even without commerce media networks needing to provide a proprietary ranking model. This promotes a scalable, minimal lift path to relevancy without trading off speed or control.

That’s why today we’re introducing Semantic Search, powered by our in-memory vector retrieval service at Koddi. This was built for ultra-low-latency retrieval and provides a scalable AI-based method for filtering eligible ad bids based on semantic relevance, not just keywords. After deploying Semantic Search, networks have seen a +21% clickthrough rate (CTR), +32% return on ad spend (ROAS), and +19% conversion rate (CVR), demonstrating that when relevance improves, so does performance.

“Retailers shouldn’t have to choose between speed and relevance. With Koddi’s Semantic Search, our platform matches ads to what shoppers mean, not just what they type, driving stronger outcomes with less lift. Since launching, we’ve seen double digit gains in engagement and return, and we’re excited to bring those results to more commerce media networks.- Nanda Lella, CTO at Koddi

What Semantic Search does

Semantic Search allows Koddi’s platform to determine bid eligibility using product attributes already present in a retailer’s catalog, such as category, price, brand, or style, without requiring a preexisting publisher ranking system.

When a commerce media network sends Koddi an ad decision request, our model retrieves relevant bidders using these attributes. For instance, if a shopper searches for “women’s clothing under $50,” Koddi fetches only those products meeting that criteria in the retailer’s catalog, automatically, and in real time.

While Semantic Search led to strong increases in performance across all categories, it can be especially powerful in precise searches based on user behavior, catalog, and ad availability. It has been exciting to see both broad network performance increases, with individual categories experiencing 2-3x lifts at scale.

Under the hood, a transformer encoder maps catalog attributes into dense embeddings in a shared semantic space, enabling nearest-neighbor matching by meaning not just exact keywords. It’s built on the same transformer architecture used in generative AI, enabling semantic matching beyond exact terms. For example, “peanut butter” can surface for “jelly,” “sandwich,” or “snack” searches reflecting how people actually browse and buy.

Because Semantic Search runs entirely within Koddi’s architecture, retailer data stays in-ecosystem, catalogs remain secure with no reliance on third-party services or external indexing. Even with the added semantic search layer, ad decisions return in under 15 milliseconds, preserving a seamless shopping experience.

Impact for commerce media networks

Semantic Search marks a shift from rigid keyword matching to semantic understanding, unlocking new reach and revenue potential across commerce media networks.

With Semantic Search, you can:

• Drive incremental relevancy. Relevant ads deliver higher engagement and conversion. Retailers see happier shoppers; advertisers see stronger ROAS; and networks see increased investment and retention.
• Reduce onboarding friction. Koddi can operate directly off your catalog attributes, no ranking model required, accelerating time to launch and minimizing integration lift.
• Eliminate dependence on keywords. Traditional keyword-based targeting limits exposure and creates manual overhead. Semantic Search removes that barrier, broadening the eligible ad pool while maintaining relevance.
• Increase yield. Every irrelevant ad is a missed opportunity to potentially drive incremental spend from customers. With more relevant ads, clicks and conversions increase, driving up commerce media networks’ revenue and ROAS.
• Maximize fill rates. With Semantic Search, because the right ads appear in the right auctions, CMNs can build relevant bid density for each placement. Koddi has seen fill rates double across networks.

This innovation extends naturally to Koddi’s existing Quality Scoring and audience targeting capabilities. Commerce media networks can layer behavioral, demographic, or geographic segments on top of Semantic Search to refine delivery even further.

Built for commerce media performance

Koddi’s dedicated team of commerce media data scientists continues to refine and expand the capabilities of Semantic Search. As new product attributes are incorporated, such as color or price, results continue to improve. Early adopters have seen CTR lifts of 21% or more simply by enriching product data inputs, all while maintaining stable pacing and spend utilization.

The outcome is an intelligent, scalable ad decisioning process that bridges the precision of AI with the control and security commerce media networks demand.