From guesswork to actionable insights

This global sporting goods company is constantly evolving its products to meet the needs of competitive athletes and fashion-conscious consumers alike. In the never-ending race to bring new products to market, understanding the interplay among a variety of marketing levers can make or break the success of a product launch.

The enterprise’s existing approach was to review sales of similar products then augment that number to match pre-existing sales targets across all campaigns. However, this would often result in a huge surplus of inventory, which in turn meant taking losses from selling items at deep discounts and distracting customers from newer, more profitable products.

Having delivered a variety of data solutions for the organization over the past several years, our team had a solid understanding of the business and saw an opportunity to be more rigorous in aligning demand with buy quantity — and ultimately improve the performance of product launches.

We proactively proposed a lightweight machine learning prediction pipeline to project revenue over the critical first eight weeks of a new product’s life cycle — the period in which an item is most popular. Our data scientists and data engineers undertook the following:

• Initial data exploration and analysis to evaluate disparate data sources and inform the development of an impactful data product.

• Feature selection to identify the data with the most predictive power and improve the signal-to-noise ratio across numerous categories and dimensions (marketing levers).

• Machine learning ensemble to deliver pre-launch what-if predictions under a range of assumptions and update forecasts throughout the initial launch window.

• Fully automated orchestration from daily data ingestion and forecasting to insights delivery via democratized dashboard and alerts for key stakeholders.

To drive adoption and prove the value of this solution, we ran an analysis phase, including presentation and insights about which business levers were most impactful in determining product launch success.

This helped the business align on 10 to 20 marketing and inventory levers and product category features. Our team began collecting this information to leverage in predictions and ease the burden of the metadata collection process.

The foundational AI/ML solution includes these technologies:

• Microsoft Power BI dashboard for interactive data visualization.

• Databricks Lakehouse as a unified platform for data, analytics and AI.

• Databricks ML Model Registry to manage the full lifecycle of XGBoost models.

• Python for task automation, data analysis and data visualization.

Our solution has resulted in significant impacts, including:

10 million+

1400+

at-risk supply of units identified so the company can take action and increase sell-through.

campaign products accurately forecasted through automated analysis of marketing levers.

15-30%

Average variance from predictions to actual campaigns.

The ecommerce team is now able to deliver high value to the buying department in terms of accurate, actionable information:

• Improved forecasting to inform buy quantities or to release pre-purchased inventory to other channels based on predicted demand.

• Whether the right levers are being activated to sell the desired number of units per campaign.

For customers, items are now far more likely to be in stock since inventory levels are accurately aligned with purchase predictions, and more relevant information is served along the buying journey.

Another round of data visualization optimization is on the agenda for this particular solution. We are continuing to provide the sporting goods company with new ideas on how technologies like AI and ML can make a measurable difference to its business.