In anticipation of his upcoming presentation at Deep Learning World, part of Machine Learning Week, June 18-22, 2023 in Las Vegas, we asked Sagar Kewalramani of Google, a few questions about his presentation, Forecasting: MLOps and Forecast Decomposition — see what’s in store at the Deep Learning World conference.
Q: In your work with predictive analytics, what behavior or outcome do your models predict?
A: Forecasting models enable businesses and enterprises to anticipate future outcomes, including consumer behavior, sales, and supply and demand. These models play a crucial role in helping businesses make informed decisions about their operations, such as inventory management, staffing, and marketing budgets. In the current fast-paced business environment, where competition is intense and consumer preferences rapidly evolve, forecasting models have become indispensable tools for businesses seeking to stay ahead of the competition. Through data and analytics, forecasting empowers businesses to make better decisions, boost their profitability, and achieve long-term growth.
Q: How does machine learning deliver value at your organization – what is one specific way in which it actively drives decisions or operations?
A: Forecasting provides significant value to our organization in various ways, including driving decisions and operations. One particular way in which it helps us is by optimizing our inventory levels. Accurate forecasting of sales ensures that we have an appropriate level of inventory to meet customer demand, which prevents stockouts that may lead to customer dissatisfaction and lost sales. Furthermore, it helps us avoid overstocking, which reduces costs and saves storage space. Besides inventory management, forecasting also informs better decisions about production, marketing, and pricing by providing insights into future demand. This helps us to be better prepared to meet demand and optimize our profits. In summary, forecasting plays a crucial role in empowering us to make data-driven decisions and enhancing the efficiency and profitability of our operations.
Q: Can you describe a quantitative result, such as the predictive lift of your model or the ROI of an analytics initiative?
A: After implementing a new DNN-based forecasting model, our company was able to achieve a 2.7% increase in sales, thanks to the model’s improved accuracy in predicting demand. By stocking the appropriate amount of inventory and avoiding stock outs, the company saw a 7-digit impact on sales. Additionally, the new model resulted in a 9% reduction in inventory costs. Overall, the forecasting effort has proven to be a valuable investment that has positively impacted our sales revenue and profitability.
Q: What surprising discovery or insight have you unearthed in your data?
A: Through data analysis & machine learning in forecasting, we have discovered a robust correlation between online product searches and actual purchases. This finding suggests that businesses can leverage online search data to forecast future demand for their products. For instance, if a company notices a significant surge in searches for a particular product, they can use this information to ramp up production or implement a marketing strategy to boost sales. This approach can help businesses avoid stock outs and improve their revenue. This is just one instance of how forecasting can empower businesses to make informed decisions and achieve their objectives.
Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Machine Learning Week.
A: The key takeaway is that DNN models with Transformer architectures are distinct in forecasting due to their ability to learn from sequences of data with different time lags and patterns. It is crucial to continuously improve forecasting models, which can be accomplished through MLOps principles like continuous integration and delivery, allowing teams to iterate on their models and integrate feedback from stakeholders to enhance their accuracy and relevance. Furthermore, it is vital to prioritize transparency and interpretability in forecasting models, enabling stakeholders to understand how the models make predictions and have confidence in the insights presented. This involves clear documentation and visualizations that clarify the key inputs, assumptions, and outputs of the model in a user-friendly way. By prioritizing collaboration, iteration, and transparency in MLOps and forecasting initiatives, businesses can create and implement more effective models that deliver real value to the organization.
Don’t miss Sagar’s presentation — Click here to register for attendance.