Using Data Enrichment to Improve Experience Design and Loyalty
On July 30, 2021, DELVE hosted the second of two episodes in our webinar series focused on gaming.
During the webinar, Mikalai Tsytsarau, DELVE’s head of engineering and DELVE sales executive, Adam Harackiewicz, demonstrated how to use Google Cloud Platform to create a predictive lifetime value (pLTV) model using in-app data and applied analytics. They also discussed the specific skills your team needs for success.
- An introduction to predictive analytics and machine learning
- An overview of feature engineering
- Features and use cases for using BigQuery
- An introduction to Google Cloud Platform machine learning tools (BQML, AutoML, AI Platform, DataLab)
- How to train a pLTV model (generating features dataset, training BQML model, evaluating trained model)
In episode 1 of Data Science in Gambling and FTP Games series, DELVE’s Adam Harackiewicz, Sales Executive CEE, shared that a key to growing revenue in games is simplifying and personalizing the user experience, as gamers have short attention spans. Increasing engagement and loyalty leads to higher profits. While it may sound straightforward, many roadblocks stand in the way of creating a compelling app user experience including technology complexity, lack of confidence in mobile and web data, lag times in user experience, and data silos (CRM, marketing automation, etc.).
Harackiewicz noted that the ongoing uncertainty associated with third-party data is another reason why companies should accelerate their efforts to collect first-party data as it is essential to customer experience and loyalty (determinants of revenue). According to Merkle’s 2021 Customer Engagement Report, 52% of respondents prioritized digital experiences and/or strategies with the goal of collecting more first-party data. What’s more, 88% of respondents identified the collection and storage of first-party data as a high priority in the next 6 to 12 months. First-party data enables a range of experience design and loyalty analytics that can improve profitability.
Beyond targeting and experience personalization, first-party data can also reduce the need for “point in time” research because smart companies can run real-time data analytics as a form of market research to inform decisions with greater precision. Harackiewicz reinforced a key message from the episode 1 webinar, in that data is valuable in developing customer segments for which experiences can be tailored to increase gameplay, spend, and lifetime value.
Marketing Analytics: Measurement
Harackiewicz reshared a framework from DELVE’s episode 1 webinar that aligned various focus areas that determine revenue (full-funnel personalization, marketing mix decisions, product-market fit) to the actions marketers should take for optimization.
He reminds us that everything begins with measurement.
Because data analytics is enabling online casinos to make games even more enjoyable to play for as long as possible. This is accomplished by giving users more control and increasing the stakes. Data analytics can show online casinos small changes they can make (such as adding symbols to a slot game) to make a very big difference in outcomes.
How a user interacts with an online or mobile app, reveals a lot about their habits. This means that online casinos are more able to target their marketing efforts (online/offline) to engage their most profitable segments. A key to gaining a holistic customer view is integrating customer data into a centralized database such as a Customer Data Platform (CDP).
Marketing Analytics: Attribution
Another key to tailoring customer experiences and making the right marketing investments is understanding attribution. While many online casino marketers might tend to focus on multi-touch attribution, there are actually three approaches to consider for understanding cause and effect: Media Mix Modeling (MMM), Multi-Touch Attribution (MTA) and Incrementality testing.
Media Mix Modeling is viewed as a “tops down” approach to allocating media investment at a macro level. Multi-touch attribution assigns the credit for conversion to individual media/digital touch points that influence the customer purchase. DELVE recommends that online casino marketers run both models to arrive at a unified measurement approach. MMM reveals the channels that drove conversions, while MTA reveals the interactions within the channels that drove conversions. By applying data analytics, online casinos can spot which channels and tactics are working best, and which are not.
Marketing Analytics: Activation
Some consumer studies suggest that consumers engage in 10 or more channels of communication before making a purchase. However, 50% of consumer brands use 8 or fewer marketing channels to reach their target audiences. This means that there is huge potential for brands to activate in various channels. Engaging in the preferred channels that consumers frequent can decrease cost of acquisition and increase loyalty and lifetime value.
Harackiewicz reminds webinar attendees that online casinos collect customer data just like any other consumer business: they monitor on-site behavior and add enriched data to deepen customer insights.
Bringing it All Together: First-Party Data Foundation + Analytics + Activation
Harackiewicz provided a powerful example of how DELVE built a data foundation (including data enrichment) and applied analytics to help online gaming provider Stride Gaming, achieve impressive ROI and conversion rate (CVR) results. By enriching first-party data with sources such as weather patterns (e.g. when it is cold outside, people play more games) and currency exchange rates (e.g. when the game player’s currency is stronger), correlations can be made to guide when and how to activate their marketing tactics to attract more game customers and increase spending.
The Role of a CDP in a First-Party Data Foundation
While it may sound daunting to integrate data for analytics use, it doesn’t have to be. DELVE’s Head of Engineering, Mikalai Tsytsarau, explains why.
Consolidating customer data into a centralized repository like a CDP actually makes it easier to import additional second-party data not just for analytics but also for modeling, training machine learning algorithms, and more. CDPs integrate data from multiple sources (POS, web, mobile, CRM, email and others) to deepen customer insights.
Tsytsarau contrasts a data warehouse from a Customer Data Platform based on how the data is organized and stored. Where warehouses store data by type, CDPs store different data types within a unified customer record to provide a 360 degree view of each customer’s behavior, attributes and transaction history. As a result, he notes that this is why “CDPs are all the rage with marketing and analytics use cases.”
Tsytsarau illustrates an example CDP reference architecture for the online gaming industry and how multiple data sources can be brought into a single CDP structure with data stored in affordable storage (via a data lake).
Tsytsarau goes even further by offering a detailed process schematic and explains how each tool works at each step of the process, including training ML models, GPUs, and processing terabytes of data in BigQuery.
One of the use cases that Tsytsarau offers for user acquisition is Stride Gaming that was featured in Data Science in Gamling and FTP Games (episode 1). The business challenge is how to identify high-value users and how to retain them to grow lifetime value.
Machine Learning for User Segmentation and LTV Prediction
Machine learning is a powerful tool for creating clustering models for segmenting users and building lookalike audiences based on demographics (including data from companies like Alteryx), user interests, and other attributes available. The resulting segments can be targeted with tailored campaigns and calls to action, as well stages in the buyer journey.
Machine learning can also be useful for prediction — especially, lifetime value of different user segments. Lifetime value is important because it helps marketers prioritize investments and focus their campaigns where they will receive the biggest return (return on ad spend return on investments, etc.). Lifetime value is also essential for deciding how much to spend in acquiring a new customer and how much to invest to retain them.
Another useful technique that was shared in Data Science in Gamling and FTP Games (episode 1), was feature engineering based on event data in Firebase. This was showcased in a Google summer camp use case so the data is publicly available. Data is accumulated from event sources to enable the assembly of a plot table of features. This is doneby aggregating event statistics such as event types (intervals between events, average wins or slots used, etc.). Once the table of user features is created, ML algorithms can be trained to predict user value.
Tsytsarau provides a detailed explanation of how to approach structuring ML models in the webinar and offers a checklist and quickstart guide to assist marketers:
Tsytsarau also reminds the webinar attendees that while in-app offers are popular, they can interfere with LTV prediction. Additionally, when building ML models, be sure to update the models when the game functionality is updated to preserve the integrity of the model(s).
The Bottom Line
Online gambling and gaming brands can collect and structure data to reveal powerful insights to improve customer acquisition, customer experience, and customer lifetime value. By applying machine learning and feature engineering, brands can optimize game features and improve segmentation for targeting.
For a deeper dive into data, analytics and machine learning in online gaming, talk to a DELVE expert.
Applying In-App Analytics to Improve User Experience and Profitability On July 23, 2021, DELVE hosted the first of two episodes in our webinar series focused on gaming. During the webinar, Mikalai Tsytsarau, DELVE’s head of engineering and DELVE sales executive, Adam Harackiewicz demonstrate how to use Google Cloud Platform to create a fully-integrated custom analytics system consisting of distributed data storage, scalable feature engineering engine, machine learning, and predictive modules to...