What happens when your forecasting models are delivering accurate predictions, but stakeholders want more detail?
In retail analytics, forecasting at scale is already a complex challenge. But when businesses want to understand how predictions break down by customer demographics, geographic regions or shopper segments, data scientists face another problem: how do you deliver more granular insights without building and maintaining an entirely new set of models?
Join us online on Thursday 26th November for our next session with dunnhumby, featuring Marco Brambilla, Research Data Science Specialist as he explores three different approaches to tackling this challenge, examining the trade-offs between accuracy, complexity and scalability.
Using a creative analogy inspired by photography and zoom technology, Marco will break down the advantages and limitations of three forecasting approaches:
- Building more granular models: Why creating separate forecasts for every segment can quickly become complex, difficult to maintain and challenging to reconcile.
- Using historical proportions: How existing forecasts can be broken down using historical data, and why this approach can fall short when patterns change.
- Applying machine learning: How dynamically modelling proportions can provide more flexible, granular predictions without rebuilding entire forecasting systems.
Drawing on his extensive experience in retail forecasting, Marco will explore when each approach makes sense, the challenges involved and how data scientists can make more informed modelling decisions.
Whether you’re an aspiring data scientist, working with predictive models or simply interested in how machine learning can solve real-world business challenges, this session will provide an accessible, practical perspective on getting more value from existing forecasts.
Supported by
