Using Machine Learning to Optimize Ad Spend and Sales Lift through causal inference and Secure Data Collaboration
BlueCloud, in collaboration with Snowflake, has co-developed an advanced ML model accelerator that provides a means to rapidly analyze advertisement spend and attribute this to revenue/sales lift in a concise way while significantly reducing potential data interpretation biases and ensuring causation is considered in a way that clearly shows the factors that should be considered. This solution is focused on achieving business outcomes in a faster, more scalable way and allows customers to go beyond aggregated views by leveraging Data Clean Rooms for additional targeting capabilities. By incorporating causal inference algorithms for advertising, and offering support for multiple variations of these algorithms, the Solution Accelerator delivers a means to rapidly deploy a customizable solution tailored to the specific use case. For example, more accurate targeting based on external factors can provide more accurate predictions related to revenue, which can support the optimization of direct spend, but also areas such as inventory and overall forecasting.
Using the Snowflake Data Cloud, Marketplace, Snowpark, Data Clean Rooms (DCR) and Machine Learning, we have developed an accelerated model to achieve causal inference and maximize the return on investment (ROI) in ad spend effectiveness, operational efficiencies, and increased revenue. Given the upcoming NBA Finals, this blog will be focused on how this model can be applied to the NBA, individual teams and their Sponsors. Identifying causal factors can not only provide incremental operational improvements, but this model also enhances fan experience and opens doors to entirely new markets, while considering variables that may confuse the situation (confounding variables). Furthermore, we explore the potential of Generative AI and Azure Cognitive Services to enhance ROI and democratize insights, allowing for more accurate targeting, testing, and model accuracy.
Executive Summary
With the BlueCloud Causal Inference Optimization Accelerator, marketing executives gain access to an advanced data platform that combines the power of Snowflake and diverse datasets while addressing potential biases in data interpretation. By implementing state-of-the-art algorithms, the platform can analyze data in a more objective and accurate manner, helping marketing teams to better understand their market and avoid biases that can be created with predictive analytics that do not consider causation.
Data Clean Rooms provide an additional layer of data accessibility, privacy, and security, ensuring that sensitive customer data is protected while still allowing marketing teams to gain valuable insights at a deeper granularity to allow organizations to avoid common traps associated with assumptions made at the individual level based on group data. The combination of data privacy, advanced analytics, and support for multiple causal inference algorithms empowers NBA teams to create targeted ad campaigns, optimize their ad spend, and maximize revenue.
The co-development of this Causal Inference Solution for Ad Optimization with Snowflake presents a comprehensive and versatile path to accelerating the return of Marketing and Advertising dollars, enabling customers to leverage the full potential of diverse data sources and advanced analytics to optimize their marketing strategies, ad targeting, and ultimately, their return on investment.
Professional Sports is used as an example below as a means of demonstrating the concept with a focus on the return to the Team, Organization and Advertisers/Sponsors.
Overview
We have created an accelerated delivery model through a customizable template that presents an innovative approach to drive better ad purchasing, ad spend, and ad sales decisions within NBA Sports teams by leveraging public external datasets from sources like Ibotta, Experian, Axciom and Comscore for Identity Management, Targeting and Measurement, amongst others; alongside internal data and subscriber data, joined within a secure environment.
- Introduction
NBA teams are constantly looking for ways to optimize their marketing strategies and ad spend to reach broader audiences and drive revenue growth. This overview proposes an advanced, yet scalable method to achieve this by integrating external datasets related to cultural, social, financial, and behavioral data, as well as internal datasets from the teams themselves and subscriber data without requiring data movement or exposing PII data.
- The Snowflake Marketplace and Snowpark Advantage
Snowflake Marketplace and Snowpark offer an excellent platform to combine and analyze various datasets, allowing for seamless data integration, secure data sharing, and a scalable environment for machine learning (ML) model deployment. Marketplace provides access to a tremendous amount of external data to enhance the overall data picture, and the integration of Snowpark creates a framework for scalable, performant data engineering and data science delivery capabilities. All needed to deliver this solution in a well architected, cost-effective manner.
- Leveraging Public External Data and Internal Data
By integrating external datasets from sources like Ibotta, Experian, and Comscore via Snowflake Marketplace, we can gain valuable insights into consumer behavior, preferences, and trends. Combining this information with the internal datasets of a team enables the creation of a comprehensive ML model to better understand the causal relationships between ad purchasing, ad spend, and their impact on overall revenue. As an example to clarify this relationship, in a churn model, we can say we can identify and target 50% of customers who are likely to churn by just targeting 20% of total customers. Business-wise, it means with fewer resources, we could potentially avoid a 50% churn event. This model can also help calculate ‘lift’ which measures how much better we can expect to do with the predictive model as compared to doing without the model.
- Combining this data with External Sponsor/Advertiser and Subscriber data in a secure form
Through the use of Snowflake Data Sharing and Data Clean Rooms, companies can enable the personalization of segment insights while preserving privacy and complying with various regulations, such as the California Consumer Privacy Act (CCPA) and General Data Protection Regulation (GDPR). By allowing advertising partners to securely join first-party audience data without moving, copying or exposing any underlying PII.1
- Machine Learning for Causal Inference
Using Causal ML algorithms, we can analyze these datasets to identify the causal relationships between ads, revenue, and consumer behavior while taking into consideration areas that may otherwise distort the actual cause/relationship between ad and revenue, such as Marital Status, Age, etc. (example in Figure 1 below). This will enable marketing executives to make informed decisions about their ad campaign strategies, leading to increased ad spend effectiveness, operational efficiencies, and revenue growth.
Predictive Analytics through Machine Learning would allow for expanded segmentation and estimation based on data patterns, coupling this with Causal Inference algorithms allows for improved insight into the effects of specific variables and courses of action to take this to the next level of decision support.
“...using non-confounding effects can also help to reduce noise in the data and improve the accuracy of the causal effect estimate.”2
Figure 1: Sample Variables contained with Causal Inference Model
- Enhancing Ad Targeting Accuracy through Causal Machine Learning and Diverse Data
A key advantage of using a machine learning model with a wide array of data sources is the ability to improve ad targeting by identifying patterns and correlations that may not be apparent. For instance, consider a marketing executive who assumes that young male basketball fans in urban areas would be the ideal target audience for a new line of NBA merchandise. While this assumption may seem logical, it could potentially exclude other valuable consumer segments.
By leveraging a Causal inference ML model with relevant factors, marketing executives can uncover hidden trends and relationships that may not be obvious through human experience alone. Beyond this, without the ability to assess the results coupled with the corresponding potential variables, for example; spend, impressions, revenue attribution/sales lift, assumptions must be made that may exclude relevant causal variables.
For example, the model might reveal that female fans in suburban areas are also highly interested in the merchandise, or that older fans in specific income bands demonstrate strong purchasing intent for certain items. Adding the sales lift information may also support that this demographic also has a higher propensity to spend in specific areas than previously thought. With this information, the marketing team can make more informed decisions about their ad targeting strategies, ultimately leading to higher conversion rates and increased revenue.
- Expansion Opportunities with OpenAI and a Financial Example
The application of OpenAI technologies, such as GPT-4, can further enhance the marketing strategies of NBA Sports teams. By incorporating natural language processing capabilities, marketing executives can gain valuable insights from social media conversations, reviews, and other unstructured data to better understand consumer sentiment and preferences. This can then be fed into the causal ML model to analyze ‘what if’ scenarios and help create targeted campaigns that achieve the best results.
For instance, consider an NBA team that spends $1 million on an advertising campaign. Using traditional methods based on human intuition and standard operational data, the team's marketing department achieves a conversion rate of 2% and generates $2 million in revenue. However, by utilizing a causal inference machine learning model such as those described above, the team could potentially increase the conversion rate to 4%, while uncovering an unknown segment that may further drive the conversion rate. With this improvement, the same $1 million investment could result in $4 million in revenue, doubling the return.
If that $1m advertising campaign is able to be targeted to a new demographic that is typically underserved by the company or its sponsors/advertisers, the value of data sharing via Data Clean Rooms creates a synergistic effect as the NBA team is able to target the right audience on both the fan and advertiser side.
In the case of the NBA, Globalnewswire.com reports that 244 sponsorships come from the restaurant sector. In 2022, the NBA signed a one-year, $8.25m sponsorship with Taco Bell3. By leveraging Data Clean Rooms and the above solution, the revenue previously attributable to the sponsorship could be increased via more targeted ad campaigns to the demographics identified via the model development and learning process. Conversely, the NBA may determine that the marketing efforts dedicated to one group, perhaps by income, would be better spent on a different income level or focused on an entirely new market segment, bringing in additional sponsors.
In summary, by securely integrating diverse public and private external datasets, with internal data and leveraging advanced technologies and processes such as causal machine learning and Generative AI; marketing executives can make more informed decisions about ad targeting, leading to higher conversion rates, increased revenue, and more efficient use of their marketing budgets. The combination of targeting data, such as demographic, cultural, social, and behavioral data; measurement; and other data sources aligned to the powerful capabilities of Snowflake Marketplace, Snowpark, secure Data Clean Rooms, and Generative AI presents a promising opportunity for many types of organizations, especially the NBA as a whole and its teams to optimize their marketing strategies and maximize their return on investment.