Platform Delivery
Cutting the cost of a sales forecast by about 60%
Brought in a national beverage distributor and scoped the work that followed: 15 won deals and $2.3M. One project moved its sales-forecasting model onto Snowpark, cutting compute cost by about 60% and running about 3× faster.
- Industry
- Wine & Spirits Distribution — United States
- Focus area
- Platform Delivery
- Headline result
- ~60%
- Headline result
- ~60%
- Compute cost down by about 60%
Context
A national wine and spirits distributor whose sales forecasts drive how it plans for seasonal demand and shifting consumer preferences.
The challenge
The forecasting pipeline ran pre-processing, training and scoring in Pandas on large warehouses. It was compute-heavy and slow, and it was getting more expensive as the data grew.
My role
Solutions engineer on the account: brought it in and scoped the extensions and expansions that followed. The forecasting work was delivered by the firm's data science team.
Approach
Moved the codebase from Pandas to Snowpark
Ran feature engineering on a standard warehouse rather than a Snowpark-optimised one
Rebuilt the training model to current best practice
Process
Outcome
- Compute cost down by about 60%
- Forecast runs about 3× faster
- The account: 15 won deals and $2.3M between 2022 and 2025