In anticipation of his presentation at Deep Learning World, part of Machine Learning Week June 18-22, 2023, we asked Google’s Sagar Kewalramani a few questions about the deployment of predictive analytics. Take a peek at his presentation, Forecasting: MLOps and Forecast Decomposition, to see what awaits you at the Deep Learning World conference.
Q: In predictive analytics work, what behaviors and outcomes do models predict?
A: Predictive models allow companies and businesses to predict future outcomes such as consumer behavior, sales, supply and demand. These models play an important role in helping companies make informed decisions about operations such as inventory management, staffing, and marketing budgets. In today’s fast-paced business environment, where competition is fierce and consumer tastes are rapidly evolving, predictive models have become an essential tool for companies looking to stay ahead of the competition. Forecasting helps businesses make better decisions, increase profitability, and achieve long-term growth through data and analytics.
Q: How does machine learning add value to an organization? What are some specific ways it can proactively drive decision-making and operations?
A: Predictions bring great value to organizations in many ways, including driving decision-making and operations. One particular way it helps us is by optimizing inventory levels. Accurate sales forecasts ensure that you have the right level of inventory to meet customer demand and prevent out-of-stocks that can lead to customer dissatisfaction and lost sales opportunities. In addition, excess inventory can be avoided, thus reducing costs and saving storage space. In addition to inventory management, forecasting also helps you make better production, marketing, and pricing decisions by providing insight into future demand. This puts you in a better position to meet demand and optimize profits. In summary, forecasting plays a key role in enabling data-driven decision-making and increasing operational efficiency and profitability.
Q: Can you discuss quantitative results, such as predictive lift for models and ROI for analytics initiatives?
A: After implementing a new DNN-based forecasting model, we were able to achieve a 2.7% sales increase thanks to the model’s improved demand forecasting accuracy. By having the right amount of inventory and avoiding stockouts, the company has had his seven-figure impact on sales. In addition, the new model reduced inventory costs by 9%. Overall, our forecasting efforts have proven to be a valuable investment that has had a positive impact on our sales and profitability.
Q: What are some surprising discoveries or insights you have uncovered from your data?
A: Through data analytics and machine learning in prediction, we discovered a strong correlation between online product searches and actual purchases. The findings suggest that companies can leverage online search data to predict future demand for their products. For example, if a company notices a significant spike in searches for a particular product, this information can be used to boost production or implement marketing strategies to drive sales. This approach helps businesses avoid out-of-stocks and improve their bottom line. This is just one example of how forecasting can help businesses make informed decisions and reach their goals.
Q: Sneak Preview: What points will you bring to the table at Machine Learning Week?
A: The important point is that DNN models with Transformer architecture stand out in prediction because they can learn from sets of data with different time lags and patterns. Continuous improvement of predictive models is very important. This can be achieved through MLOps principles such as continuous integration and delivery. This allows teams to iterate on models and integrate feedback from stakeholders to improve accuracy and relevance. Additionally, it is important to prioritize the transparency and interpretability of predictive models. This allows stakeholders to understand how the model makes predictions and have confidence in the insights presented. This includes clear documentation and visualization that articulate the model’s key inputs, assumptions, and outputs in a user-friendly manner. By prioritizing collaboration, iteration, and transparency in MLOps and predictive initiatives, companies can create and implement more effective models that deliver real value to their organizations.
Don’t miss Sagar’s presentation — click here to register.
