BLUME2000 – ML-based Demand Forecasting with Google Cloud
2023-10-09 | Case study | Insights
The ML-based demand forecasting brings us numerous advantages, including improved accuracy, faster and more efficient predictions, better inventory management, waste reduction, and boost in sales. We were genuinely impressed by the remarkable precision and high level of automation achieved through this approach.
Kristofer Klein | Head of Marketing Channels, Webshop & Social Media
On average only 12% of daily delta between actual vs forecasted
Very low cost of model and processing the workflow
The Challenge
The Challenge
BLUME2000 is a well-established company in the floral industry with a strong reputation for quality and innovation – online and offline. Fresh flowers are at the heart of the product lineup, yet their limited lifespan poses a distinctive challenge in terms of potential waste but also to effective steer an entire week. To minimize waste and be as efficient as possible, the client sought to align the supply and demand as seamlessly as possible. Moreover, budget planning for both short and long term posed a puzzle for the client, as the precise impact of previous activities on sales was unknown.
The Approach
The Approach
A time series forecast, taking into account the daily sales and further external factors including special days, marketing spending, and coupons, was built within Google Cloud. An explanatory analysis revealed insights to the past patterns of sales as well. Furthermore, the integration of various spending data was automated in Google Cloud's BigQuery through data connectors, streamlining the entire process for a smoother workflow.
The Results
The Results
A Looker Studio dashboard provides the forecasted revenue for the upcoming days and weeks, as well as a detailed breakdown of historical revenue based on external factors. This allowed for a comprehensive understanding of the impact of marketing activities on sales. BLUME2000 could make informed decisions regarding short and longer-term sales. If the revenue forecast for the upcoming weeks fell below the desired target, they utilize insights from historical revenue decomposition to allocate the budget appropriately among different marketing channels.
The Goals
- Fresh flowers are central to the product lineup
- Limited lifespan of fresh flowers poses a unique waste challenge
- Client aims to minimize waste by aligning supply and demand efficiently
- Budget planning is challenging due to uncertainty about the impact of previous activities on sales
The Approach
- Time series forecast created using daily sales data and external factors (special days, marketing spending, coupons) was built in Google Cloud
- Explanatory analysis conducted to gain insights into past sales patterns
- Automation of integration for various spending data using Google Cloud's BigQuery
- Streamlined workflow for improved efficiency
The Results
- Looker Studio dashboard offers forecasted revenue for upcoming days and weeks
- Detailed breakdown of historical revenue based on external factors available
- Provides insights into the impact of marketing activities on sales
- Enables informed decisions for short and long-term sales strategies
- Budget allocation adjustments based on historical revenue insights when forecast falls below target
- Average daily delta between actual and forecasted demand is ~ 12%