Understanding customer feedback is critical for businesses, but analyzing large volumes of unstructured text can be challenging. In this solution, you will use Cortex AISQL to systematically gain insights from unstructured customer feedback.

This solution is one of the templates inside Snowflake, that allows you to leverage multiple AISQL functions to answer different use case questions upon customer reviews.   You can access the template from within Snowflake by clicking the “Open in Snowflake” button above, or you can follow the instructions below to execute this code on your own.

Tasty Bytes is a global e-commerce company selling different merchandise. They collect customer reviews to gain insights into how their products are performing.

The following code can be copied into a notebook to execute.  It leverages multiple AISQL functions to answer different use case questions upon customer reviews.

Let's begin by running the query below. It sets the proper context for this session. It also creates and populates two tables, PRODUCT_REVIEWS and PRODUCT_CATALOG, with sample data for our analysis.

Create a Project > Notebook and add the following SQL code and add the cell name: IMPORT_DATA_SQL

Now add this code to another SQL cell and name it: CHECK_DATA_SQL

As a first step, let's perform a quick sanity check. We'll use the SNOWFLAKE.CORTEX.SENTIMENT function to score the sentiment of each review. We can then check if it correlates with the user-provided star rating to see if they align.

Add the following code to another SQL cell and name it: SENTIMENT_CHECK_SQL

Now, let's dig deeper. Suppose you want to know what the biggest complaints are for 'Electronics'. You can focus on the ones with negative sentiments, and use AI_AGG to analyze all relevant reviews and aggregate the common themes into a single summary.

Add the following code to another SQL cell and name it: AGG_TOP_ISSUES_SQL

-- The text may not display fully in the SQL cell. Please hover or double-click on the SQL cell to view the full text.  Print the result to a dataframe for easier reading using the following code in a Python cell, name it: DISPLAY_TOP_ISSUES_PY

To answer this question, we start with filtering to Clothing category. Another way to identify comments that mentioned product issue is to leverage our latest AI_FILTER to conduct filtering using natural language.

The next step we use the AI_AGG function to get a list of all product issues mentioned.
Create a SQL cell to add the following code and name it: COMMON_ISSUE_SQL

Now, we'll print the result to a dataframe for easier reading.  Use the following Python code and name the cell:

With the issues suggested through the AI_AGG function pipeline above, we can now leverage AI_CLASSIFY to turn into continuous data pipeline to keep classify the reviews.

Paste the following code in a SQL cell and name it CLASSIFY_SQL

Finally, let's close the loop. You can use AI_COMPLETE to help your support team draft empathetic and relevant responses to negative reviews, improving customer satisfaction at scale.

Create another SQL Cell and name it: GENERATE_SQL.  Add the following code to it: