Databricks Simplifies Forecasting

Modern retail demands rapid, accurate forecasting, a challenge that has outpaced traditional tools and available expertise. Databricks aims to bridge this gap with its new MMF Agent, an AI-powered workflow designed to make sophisticated demand planning accessible to more teams.

Enterprise forecasting has become exponentially complex, involving millions of time series, rapid SKU proliferation, and shrinking planning cycles. Legacy systems and even advanced statistical methods struggle to keep pace with this scale and variety.

The Talent and Technology Squeeze

Running enterprise-scale, multi-model forecasting requires deep expertise in statistical methods, machine learning, and distributed systems. This specialized talent is scarce, creating a bottleneck for many organizations. Even for teams with the necessary skills, the setup and evaluation process can consume days or weeks, a timeline often too slow for agile retail planning.

Databricks' own Many Model Forecasting (MMF) framework, an open-source solution integrating over 35 forecasting models, addressed some of these challenges. However, it remained largely the domain of forecasting experts due to the knowledge required for configuration and interpretation.

MMF Agent: AI Guidance for Demand Planners

MMF Agent, built on Databricks' Genie Code AI assistant, wraps the MMF framework in an interactive, guided experience. It automates data quality checks, time series classification, compute configuration, and model selection.

The agent interacts with users, leveraging its understanding of the Databricks data environment and Unity Catalog. This allows planning leaders, even without deep data science backgrounds, to engage using familiar demand planning terminology.

This guided approach significantly compresses setup and experimentation time, potentially reducing it from days to hours.

Furthermore, MMF Agent aims to improve forecast accuracy through better data preparation and targeted model selection. Crucially, it democratizes access to advanced forecasting techniques, previously limited to organizations with specialized data science teams.

For existing MMF users, the agent also offers a pathway to customization. It can guide engineers through modifying the framework, such as adding new model classes or adjusting evaluation logic, lowering the barrier to bespoke solutions.

Getting Started

MMF Agent skills are available in the Many Model Forecasting GitHub repository. Databricks suggests that demand planning leaders explore the offering to improve forecasting capabilities with existing teams and tools.