Forecasting is an age-old science. One would think that IT systems deliver reliable forecasts for most businesses.
We have used forecasting to solve many non-traditional business problems, that the IT systems do not cater to.
Some examples of our forecasting work include:
- Internet of Things (IoT): Forecasting the electricity usage of smart meters over multiple short horizons, for a smart city pilot
- Telecom: Daily forecasting of telecom services (data, voice, SMS, roaming, etc) usage of millions of customers for a global MVNO, with the objective of optimizing service purchase
- Auto Retail: Forecasting the non-linear car sales (by model, color, trim) for a car dealer network, to optimize its OEM incentives
- E-Commerce: Forecasting the returns for various items sold by an e-commerce retailer
- Retail: Developing thousands of SKU-level models to predict sales under the impact of promotions and dynamic pricing
Increasingly, we use Machine Learning techniques to develop forecasting models. For some of our clients, we use Long Short Term Memory (LSTM) techniques, a type of Recurrent Neural Network (RNN).
We are currently experimenting with using external variables in our LSTM models.