Podcast Summary:

Voices of Search – How Do You Roll Out Embeddings Across Your Product Catalog?
Host: Tyson Stockton
Guest: Ryland Bacorn, Bokeaday
Date: November 21, 2025

Episode Overview

In this episode, host Tyson Stockton discusses with Ryland Bacorn (Bokeaday) the practical steps and strategic thinking behind implementing embeddings (semantic vector representations) across a large product catalog, specifically for enterprise ecommerce sites. The conversation explores how embeddings extend value beyond traditional SEO into areas like site search, product recommendations, and broader business operations—highlighting actionable strategies and common pitfalls.

Key Discussion Points & Insights

1. Initial Planning & Problem Definition

• Clarity is Essential (01:06):
  Ryland emphasizes starting with a precise understanding of the business problem embeddings can solve—not just for SEO, but also site search, recommendations, etc.

  "The first steps anybody should take in figuring out their problem is really get some clarity around it." (C, 01:06)

• Identify Use Cases:

  • SEO improvement
  • Site search relevance
  • Product recommendations
  • Enrichment of product data

2. Technical Foundation & Infrastructure

• Database Choice Matters (01:40):
  Not all databases handle embeddings; using specialized solutions (like Pinecone or knowledge graphs) is necessary.

  "Not every database can support embeddings. You need a specialized database. So like Pinecone or setting things up on like a knowledge graph…" (C, 01:24)

• Data Preparation:
  Ensure all product details (titles, SKUs, categories, sales data) are present in the database to facilitate downstream objectives.

3. Proof of Concept & Prioritization

• Start Small — 80/20 Rule (02:20):
  Focus on the catalog portion driving the most revenue or having the most robust data for testing.

  "...think of the 80/20 rule. So what's driving real revenue? Or what has the most robust amount of content..." (C, 02:20)

• Model Selection:
  Begin with accessible open models like MiniLM-Sentence Transformers, but consider ecommerce-specialized models (e.g., Marco) as you scale.

4. Building the Stack & Gaining Buy-In

• Cross-Team Buy-In is Critical (03:15):
  Building the embedding infrastructure requires collaboration and support from multiple teams.

  "...you really should be seeking the buy in and this is where that buy in is essential." (C, 03:15)

• Keep Database Indexing Up-To-Date:
  Set up nightly indexing to ensure evolving product data is reflected in embeddings.

5. Demonstrating Impact

• Site Search Enhancement:
  Embeddings improve semantic relevance in search, outperforming strict text matching even in systems like Google Site Search.

• Product Recommendations:
  Showcases the value of linking semantically related or complementary products.

• SEO Taxonomy & Internal Linking:
  Improved category structures and better internal links drive user engagement and search signals.

• Content Generation:
  Content teams can leverage embeddings and vector databases for more targeted, scalable content ideas.

  "...team can produce significantly more content based on recommendations that this vector database can provide..." (C, 05:12)

6. Broader Business Value—Think Beyond SEO

• Universal Applications Across Org (08:06):
  Embedding-powered tools benefit more than just marketing or tech teams—they can help HR, business intelligence, and beyond.

  "The applications here across the company are huge and the appetite is only growing." (C, 08:06)

• Advice on Positioning the Project:
  Tyson emphasizes framing embeddings as a business-wide initiative rather than just another SEO project—critical for securing resources and momentum.

  "...not necessarily trying to pin it just on a search value. But...think of the broader business objective and impact." (B, 07:35)

Memorable Quotes & Moments

• On Proving Value Quickly:

  "You don't want to test on things that aren't getting any attention anyway." (C, 02:35)

• On Internal Linking & User Experience:

  "Internal linking is going to help users get to more semantically relevant experiences, delivering a better experience." (C, 04:51)

• On Business-Wide Applicability:

  "Doesn't matter what department you're in, it's applicable all over." (C, 08:46)

Notable Segments & Timestamps

• [00:43] – Key implementation question: How to deploy embeddings at scale for ecommerce?
• [01:06–03:00] – Planning, selecting a database, data prep, starting with a proof of concept.
• [03:15–05:12] – Building buy-in, iterating, business impact examples (site search, recommendations, taxonomy).
• [07:10–08:06] – Tyson’s advice on positioning: framing for broader business impact.
• [08:06–08:57] – Ryland on cross-departmental opportunities and appetite for embedding technologies.

Takeaways for Listeners

• Start with a clear business problem; don’t treat embeddings as just an SEO tool.
• Use the 80/20 rule to prioritize segments of your catalog and build a proof of concept.
• Select the right infrastructure and keep your database up to date.
• Elicit buy-in by demonstrating cross-team value—think beyond marketing.
• Success with embeddings strengthens SEO, user engagement, content creation, and overall business intelligence.

For more insights, check out Ryland Bacorn’s work at bokeaday.com.