Supercharge Marketing Analytics with Data Clean Rooms and Snowpark Container Services

Data engineers in the fast-paced world of marketing and advertising are expected to deliver granular insights, build sophisticated customer segmentations and measure campaign effectiveness with high accuracy. At the same time, they face a slew of challenges, including a growing thicket of privacy regulations and the ever-present need to protect sensitive customer data.

To unlock the full potential of an organization’s data while navigating this complex landscape,  data engineers can use a combination of Data Clean Rooms and Snowpark Container Services from Snowflake.

The challenge of data collaboration in a privacy-first world

The most valuable insights often come from combining their organization’s first-party data with data from their partners. For example, you might want to:

• Enrich customer profiles with data from a second-party partner to better understand their interests and behaviors

• Build lookalike models to find new audiences that resemble their most valuable customers

• Measure the effectiveness of their advertising campaigns by joining conversion data with a publisher's impression data

However, sharing raw, personally identifiable information (PII) directly with partners is a nonstarter. 

Secure collaboration and advanced analytics

Snowflake Data Clean Rooms enable a secure environment where multiple parties can collaborate on data without exposing the underlying PII. Think of it as a neutral ground where partners can bring data together for analysis, but with strict controls on what can be done and what data can leave the clean room.

For those who want to do more than just simple joins and aggregations or run complex machine learning models to power marketing analytics, Snowpark Container Services takes data clean rooms to the next level.

With Snowpark Container Services, teams can now run custom code and even entire Docker containers directly within an organization’s Snowflake Data Clean Room. This means they can:

• Bring their own machine learning models (e.g., built with Scikit-learn, PyTorch or XGBoost) and run them on the combined data in the clean room

• Use their preferred programming languages and libraries (like Python) to perform advanced data transformations and feature engineering

• Deploy and manage their code as a service, making it easy to integrate their models into their production workflows

Benefits for marketing and advertising

So, what does this mean for data engineers in a marketing or advertising organization? Here are just a few of the benefits:

• Enhanced audience insights: Securely join an organization’s first-party data with data from partners to get a more complete view of customers.

• More powerful lookalike modeling: Build more accurate lookalike models by training them on a richer, more diverse dataset.

• Improved campaign measurement: Get a clearer picture of an organization’s return on ad spend (ROAS) by more accurately attributing conversions to marketing campaigns.

• Greater flexibility and control: Use familiar tools and libraries to build and deploy models within a secure and governed environment.

• Reduced time to insight: Organizations can stop spending time on complex data integration projects and start delivering faster business insights.

How to build a lookalike model

Imagine a data engineer for a retail brand wants to find new customers who are similar to its most loyal shoppers. They have a list of their best customers, and they’ve partnered with a publisher that has a large audience of online readers.

Here's how the data engineer could use Snowflake Data Clean Rooms and Snowpark Container Services to build a powerful lookalike model:

1. Set up a Data Clean Room: The brand and the publisher both make their customer data available in a Snowflake Data Clean Room with appropriate policies, so no PII is exposed.

2. Train a lookalike model: The brand uses a Snowflake Notebook to build a lookalike model using XGBoost. The publisher can train the model on the combined data in the clean room, using their list of loyal customers as the target variable.

3. Deploy the model with Snowpark Container Services: The brand packages their model and its dependencies into a container, and the publisher deploys it as a service using Snowpark Container Services.

4. Score the publisher's audience: The publisher can now run their audience data against the brand’s model to get a "lookalike score" for each of their audience. The brand can then use these scores to create a custom audience for their next advertising campaign.

The combination of Snowflake Data Clean Rooms and Snowpark Container Services provides a secure, flexible and powerful platform for collaborating on data and running advanced machine learning models, all while respecting user privacy and adhering to compliance requirements. 

Additionally, with ML Jobs within DCR which is now in private preview, you can do the same SPCS based deployment but with a simplified ML development workflow and optimized ML runtime. To try this out, please reach out to your account team.

Embracing this new paradigm of data collaboration unlocks a wealth of new insights, drives better business outcomes, and becomes an even more valuable partner to marketing and advertising teams. 

Follow the step-by-step example to setup Data Clean Rooms and Snowflake Container Services here. It also provides sample notebooks which you can download and try out in your environment.

Learn more about Snowflake’s latest Clean Room innovations, including new AI/ML capabilities and the expansion to multi-party data collaboration.