INTRODUCTION
Businesses across domains are heavily relying on data to make policy decisions and to optimize key levers to achieve business outcomes. For example, retail businesses are using price elasticity to optimize prices and digital marketing teams in various industries are using data to optimize marketing expenditure.
While leveraging data in the right way can help businesses achieve their business goals, there is a huge risk in drawing wrong inferences and poor decisions if not done correctly.
For example, let’s say an analyst is tasked with finding the optimal price for a product using historical prices and sales volume using day level data. The analyst observed that a 10% decline in price is leading to a 10% boost in the sales volume and makes the recommendation to cut prices to boost revenue on the basis of the insight that the product is elastic.
Price | Sales volume |
100 | 1,000 |
100 | 1,000 |
100 | 1,000 |
100 | 1,000 |
90 | 1,100 |
90 | 1,100 |
When the team saw that cutting prices is not improving revenue, they started a deep dive and noticed that the jump in sales is happening due to increased footfall in the store on weekends and the product is inelastic.
Price | Sales volume | Day of week |
100 | 1,000 | Monday |
100 | 1,000 | Tuesday |
100 | 1,000 | Thursday |
100 | 1,000 | Friday |
90 | 1,100 | Saturday |
90 | 1,100 | Sunday |
In our experience of working across industries, we noticed that businesses are heavily relying on correlations and predictive models to make policy decisions and using correlations as causations leading to suboptimal and often time wrong decisions.
In this white paper, we will walk you through our novel approach to make product design decisions in telecom using Causal ML.
What is Causal ML?
Before we go deeper into the problem statement, let’s quickly understand a few basics:
Correlation: A statistical measure that shows the size and direction of a relationship between two or more variables. Correlation doesn't necessarily mean that one variable causes the other to change.
Causation: Causation indicates that one event is the result of the occurrence of the other event.
Causal ML: Machine Learning(Causal ML) sits at the intersection of machine learning and causal inference to determine cause-and-effect relationships. While traditional machine learning models focus on predicting outcomes based on correlations in the data, Causal ML aims to answer “what-if” questions, such as “What will happen if we change this feature?” or “What is the effect of this action on the outcome?”
CHALLENGE
In the fast-paced telecommunications industry, companies need to constantly innovate to stay relevant and competitive in the field. Our client, a leading telecommunications and digital services provider in the Philippines, struggled with multiple issues related to addressing customer demand while optimizing revenue in the presence of constraints. The client relied heavily on manual and intuition-based product designs with minimal experimentation and limited use of data-driven optimization. This approach led to inefficient campaign execution, unsatisfactory revenue outcomes, and even unintended product cannibalisation.
In the dynamic world of product design, understanding the causal relationships between product features and sales is crucial for making informed decisions. Traditional data analytics often reveal correlations; but to truly innovate, businesses need to delve deeper into causality. This is where causal machine learning (ML) comes into play, offering a powerful approach to design products that not only meet customer needs but also drive sales. Thus, to address issues related to product design, the client needed to shift towards an advanced, Causal ML-powered framework that could help their planners in creating and designing new products.
SOLUTION
The OneByZero team built causal ML models for the client with the following objectives in mind:
- Quantify and measure the effects of the product features on its sales count.
- Predict the sales count reliably in response to changes to the product features.
- Utilize ML models to create new product designs, discounts, and freebies while taking into account various conditions such as prevention of product cannibalisation.
Implementation
Hypothesis formulation & data preparation
The first step in leveraging causal ML is identifying the key features of the product and formulating hypotheses about their potential impact on sales. For instance, in an electronic gadget, features like battery life, screen size, and price might be considered. Hypotheses could include statements like "A longer battery life leads to higher sales" or "Reducing the price increases sales volume." These hypotheses must be aligned with existing business assumptions, which allows us to validate the insights from the model. We then selected which products would be included in the same model. This is to ensure that outlier products would not affect the model estimates and to obtain more accurate coefficients due to lower variance in the values of the features.
Development of multiple causal ML models
Using double machine learning, OneByZero developed several ML models that measured the feature effects to the sales count. Each model corresponded to a specific set of products which were grouped according to similarity. The business objective of these models is to use an existing product as a reference or base, change some of its features, and see the predicted effect on the sales count of the “new” product. Our models enabled the planners and marketing team to identify which changes to the product should be implemented to generate more revenue while maintaining the sales count. Aside from product features, these models also used various customer-based features which were aggregated according to the products.
Model evaluation
To ensure that the models are correct and working as intended, we examined the models’ accuracy to predict the sales count. Specifically, product A was used as a baseline to predict the sales count of the product B by changing product A’s feature values to match product B’s features. Because these models are intended to be utilized for specific use cases and business objective, only specific products were chosen as baselines; that is, model performance was measured only on the products that will be used as reference in designing new products. Through extensive data preparation and careful feature selection, we were able to measure the feature effects on the sales count while also maintaining low error rates.
RESULTS AND BENEFITS
Implementing Causal ML in product design leads to more effective and efficient product innovation. By understanding the true causal impact of product features, our client prioritised design elements that have the most significant effect on sales, mitigating the risk of costly design errors. This data-driven approach ensures that product development is aligned with market demands, leading to higher customer satisfaction and increased revenue.
Key Outcomes:
- Increased Sales Through Optimized Features: By focusing on features with proven causal effects on sales, companies can enhance their products in ways that directly contribute to revenue growth. In this specific use case, were able to predict the sales count of a new product with only 17% mean error.
- Improved Product-Market Fit: Causal ML helps ensure that new products are tailored to meet the actual needs and preferences of customers, improving market acceptance and reducing the time to market.
- Enhanced Decision-Making: Product teams can make more informed decisions, backed by robust data, reducing the reliance on intuition or guesswork.
WE ARE ONEBYZERO
Headquartered in Singapore with local presence in Asean nations, we are a modern data & AI consulting firm. We focus on transforming enterprises with cutting edge solutions to generate value from data. We specialize in serving the Telecommunications, Banking & Financial Services, Retail and Ecommerce industries.
We focus exclusively on AI/ML, Data & Martech
AI & ML, including Generative AI
We help organizations define their AI/ML strategy, develop and operationalize AI/ML & generative AI models, and implement MLOps to streamline operations. We have experience doing a variety of work in classical data science and cutting edge GenAI for Telcos
Modern Data Platforms
Our team of experts build robust data pipelines, design data warehouses and lakes with strong focus on data quality and lineage, to build C360, marts, support reporting, dashboards to help organizations uncover insights in their data using a modern data stack.
Martech & Personalization
We enable customers to modernize their digital platforms to deliver omni channel personalization use-cases. We have deep experience with marketing use-cases & personalized content generation, automated campaign grid optimization and next-best offers & actions.
