Utilized CAE's forecasting engine to deliver a forecasting program at scale
Centralized Analytics Environment (CAE) is Fractal’s proprietary analytics platform which provides an analytics development infrastructure along with codified standard workflows and algorithms to develop analytics at scale with speed, thus making more capacity available for innovation.
The Business Challenge
Clients wanted to get market size forecasts for all 150 markets and 70 product segments across the globe on a monthly basis for budget planning.
Challenges with building forecasts at scale
- A wide product portfolio with dynamic market conditions and changing macroeconomic trends -> Diversity of data requires customized models for higher accuracy.
- Finance teams, brand manager, planners across all geographies are involved in the planning process -> Multiple stakeholders have varied modeling assumptions.
Desired outcome from CAE
- A standard, flexible, scientific and stable solution.
- An end-to-end platform from data input to final business alignment.
CAE’s forecasting engine, prebuilt with 32 forecasting models, chooses the best-fitting model for each combination, optimizes the model parameters and generates forecasts. The engine helps in:
- Selection of an optimal technique per scenario.
- Deployment of results/ business alignment.
Other benefits include:
- Flexibility to extend the modeling range and include modeling assumptions.
- Access to results through an intuitive graphical interface; results can be downloaded for further applications.
- Reasonable accuracy of forecasts.
- Little to no manual input.
Fractal Analytics leveraged CAE’s forecasting engine to build scaled forecasts for 8,000 combinations in 75% less time. Getting the results with high accuracy in lesser time improved adoption of statistical forecasts by 90%.
- Automated forecasting approaches can be designed to be as effective as custom models in providing accurate forecast in the long run.
- Lean and scaled processes significantly reduce global forecast publication time from 3 months to 3 weeks.
- Reduced model output generation time by 75% by standardizing and automating the process, and through distributed processing of results.
- Improved business adoption of statistical forecasts by 90% by incorporating feedback on key business drivers in the model building process.
- For each different combination, the forecasting engine automatically chooses the most appropriate forecasting model, optimizes the model parameters, and produces the best-fitting forecasts.
- The engine, running 8,000+ time series at the same time on a big data platform, requires little manual input and delivers forecasts with reasonable accuracy.