Skai Deploys AI-Powered Product Categorization Tool in Two Days with Snowflake Cortex AI

Skai offers a technology platform that helps organizations manage their ads and connect with customers on Google, social media platforms, and even ecommerce sites like Amazon, Target and Walmart — all through a single pane of glass. Skai Labs is a small, agile team within Skai that works with specific customers to deliver the custom platform features they need to get even greater performance from Skai.

To deploy new solutions at speed, Skai Labs uses low- and no-code tools to build custom configurations. But when a leading brand performance marketing agency managing over $4B in digital media needed to rapidly categorize thousands of products across multiple ecommerce platforms, Gal Zohar, Skai’s VP of solutions development, recognized a potential challenge for the Skai Labs team: “In the past, we’d need to rope in our data science team to solve this challenge. They have the tools, but their capacity is limited — and it would take a long time to get this on their roadmap.” But Skai Labs didn’t have time to waste, as product categorization was a significant pain point for this customer.

To move quickly, Skai Labs wanted to deploy a large language model (LLM) to automate product tagging and categorization. As a Snowflake customer, Skai Labs deployed an LLM using Snowflake Cortex AI, bringing a generative-AI-powered categorization tool to production in just two days.

Correctly categorizing products is key to ecommerce success. “Walmart, Amazon, Target and every store has its own taxonomy,” explains Zohar. “But for analysis purposes, you can’t look at these platforms in isolation. To get real results, our customers need a single taxonomy to get meaningful insights across different platforms.”

Skai already has ways to help its customers build categories that make sense across multiple platforms. But the situation was different when working with this particular customer — an already mature performance marketing firm with established product categories. “We needed to find a way to take existing products and build them around our customer’s unique and well-established list of categories while still matching category tags on major stores,” says Zohar. “But this isn’t always straightforward. Some products might fit into multiple categories, like a vacuum cleaner you can use at home or in your car. Or what about face cream? Does that fit in ‘beauty’ or ‘skincare’?”

Manually categorizing products would take too long. But using basic automation would run the risk of miscategorizing complex products. To deliver accurate product tagging at scale, Skai Labs turned to an LLM.

As an existing Snowflake customer, Skai deployed this LLM workflow quickly and easily through Snowflake Cortex AI. “We used an LLM through Cortex to take each product in our database and work through a series of prompts we developed,” explains Zohar. “It read the product name, figured out the appropriate category, and automatically filled this in for every product.”

In a single day, Zohar and his team created a prompt for the Mistral LLM and got it running on the customer’s product database. By day two, the team had made further progress and released the feature into production. “We’ve done a bit of tweaking since to add subcategories and increase accuracy,” says Zohar. “But to get to the point where we could deploy in production in just two days was a huge win for us, as I imagine this would have taken at least a month to deploy if we’d taken a different approach. That sort of fast pace is exactly how Skai Labs delivers such great results for our customers. And we could do this so easily. I didn’t need to become a prompt engineer or an expert in the latest LLMs.”

This rapid time to market also didn’t need to come at the expense of coverage or accuracy. Out of almost 100,000 products, all but 18 were automatically categorized. To ensure accuracy, Skai Labs used a filter to remove any miscategorized products. “We used Snowpark to apply existing machine learning code in Python to measure how relevant a term is to each product,” explains Zohar. “In all, 99.98% of products passed through that filter with no issues. And the customer has been very happy with the level of accuracy we’ve achieved.”

While the mixture of Snowflake Cortex AI and Snowpark has helped validate the accuracy of the AI-powered categorization tool, it has a wider impact on the teams at Skai. “Four to five months ago, it was often difficult to collaborate with our data scientists as neither of us had a common language,” says Zohar. “Snowpark solved this. We have three other projects that we’ve been able to work on with our data science teams thanks to Snowpark.” 

With product categories now optimized across all major online storefronts, everyone gets the insights they need to boost revenues. Skai’s client, the performance marketing agency, can get the data it needs to deliver even more informed guidance on where its clients can focus their marketing spend. And with this tailored guidance, these brands working with the performance marketing agency get far greater returns on their marketing budgets.

And as the performance marketing agency sees more value from Skai, it’s likely to stay with the platform — and that’s the core of what Skai Labs is about. “A big part of our remit is to keep customers happy,” says Zohar. “When they can see important sales trends across key product categories — and get insights into which categories are performing above expectations — it delivers so much impact for their business.”

While this specific product categorization solution was a bespoke build, the experience of working with LLMs in Snowflake Cortex AI has inspired Zohar and his team to think differently about future customer engagements. 

“We’ve played with LLMs before,” recalls Zohar. “But it was hard to scale them within a team of our size. With Snowflake Cortex AI, we can explore these features. It’s opened the way for my team to use LLMs. Having Cortex AI in my toolbox will make a big difference.”