Unlock next-level insights: Introducing Segment Reverse ETL with Snowflake Cortex

Today, we’re excited to share a new way for data teams to derive intelligent insights and activate those insights across an organization in minutes: Reverse ETL with Snowflake Cortex. This powerful integration abstracts away thousand-line SQL queries, enabling companies to quickly derive insights to enhance customer journeys.

Since the dawn of SQL in the 1970s, analysts and engineers have struggled to distill their bold analytical ambitions into the SELECT and HAVING clauses that can unlock insight from raw data. 

But there was usually something lost in translation between the idea in the head and the SQL query in the terminal. The best analytics & data science teams have found ways to get it done with thousand line SQL queries that find signal in the noise, followed by the python scripts to get those insights into the business systems – the CRM, the support tool, the email platform – to enhance the customer journey. 

With the advancement of Large Language Models (LLMs), data platforms, and Reverse ETL, what would have taken teams of PhDs and data engineers months to accomplish is now within reach for all businesses. 

In this post, we’ll show you how you can use Reverse ETL with Snowflake Cortex to derive new insights from your data – like sentiment analysis, advanced summarization, classification – and activate it everywhere you need, with just a few lines of SQL.

Generating Intelligent Insights with AI Traits

New LLM models open new insights for your data. But being able to apply these models to your data has been complex – requiring additional ETL jobs or, worse, copying & pasting into a chatGPT window. 😱 

Snowflake Cortex, currently in Public Preview, enables organizations to analyze data directly on top of data that already resides within their data platform. The capabilities are really powerful: with a single SQL function, you can summarize unstructured data, score sentiment for a review, or ask questions directly of your data. 

But generating insights is not enough to drive business impact. For analytics & data science teams to truly drive business outcomes, they need to equip teams across their org with these insights. That’s where Reverse ETL can help, enabling teams to sync these generative insights into hundreds of applications to influence, augment, and accelerate the customer journey. 

Reverse ETL with AI in Action

Let’s see how easy it is to generate a new insight from our data & use it to send a personalized message in a multi-channel marketing tool like Braze. For this example, let’s say we want to send a personalized email recommending the next book in a series, based on the sentiment analysis of customer reviews. We’re using Braze, but you can pick from hundreds of tools to connect. You’ll need a Snowflake account with customer data if you want to follow along. 

1. Selecting a Dataset

First, let’s get a feel for the dataset. We’ll be querying a Reviews table, which comes with a user_id and unstructured review field.  You can see a sample below.

2. Using AI Functions

Next, we’ll use two LLM Functions to extract insights from this messy data. The first LLM Function is sentiment, which is an out-of-the-box sentiment analysis model from Cortex. This will return a value from -1 to 1 based on the sentiment of the review. 

The second LLM Function is more configurable & flexible: complete, which takes a model, a prompt, and the text field to operate on. We’ll use the following prompt: “would the user like the next book or game in the series?” 

Here’s the full query: 

And we can see the result set here: 

You can see examples where the recommendation is more nuanced than the sentiment, based on the context of the review! A review that says “I am just so frustrated I don’t know how the story ends!” may be lower sentiment, but qualify for recommendation. 

3. Sync Traits to Profile & Destinations with Reverse ETL

Now, with a few clicks, we can set up a Reverse ETL job to sync these insights into the Segment user profile and into our marketing automation tool, Braze. 

We can see these traits on the user profile & synced downstream into Braze. Now our marketing team can build advanced audiences and send a personalized email to reviewers, recommending the next product in the series or suggesting a new series to jump into based on their feedback!

You can see the AI generated sentiment score reflected on both the Segment profile (left) as well as the Braze profile (right).  This score can now be used as a criteria to build cohorts to send personalized messages to depending on their last review sentiment. 

We’re Excited to See What you Build

Combining your data with the natural language understanding (NLU) from LLMs unlocks use cases for virtually any business. We demonstrated how this technology can be applied to sentiment analysis for reviews, but the same steps can be applied to unlock a better understanding of your customers through analyzing support ticket data, call transcripts, profile data, or other datasets.

Our goal at Segment is to provide world class data infrastructure for advanced data & marketing teams. We’re excited about the potential for AI Traits & Reverse ETL to unleash new insights across your organization and easily activate that data across your business applications. 

Reverse ETL with Snowflake Cortex is currently in Open Preview. Feel free to reach out to us to share your use case, questions, and ideas for Reverse ETL with Snowflake Cortex.