Grocery delivery app saw 10x faster time to market
Company Size: Operates in 26 cities with thousands of products and over 2K employees
Solution: Forecast demand at SKU level across cities
Platform Use Case: Demand forecasting
Data Stack: BigQuery
partner stores’ data included in models
accuracy for forecasting top revenue categories
Over 12,000 suppliers and 26 cities of data
This grocery delivery app company needed to build and maintain a demand forecast model that could accurately predict sales volume at the SKU level. In addition to needing a granular forecast for each product, the company also needed to anticipate demand for each item across more than 26 cities with over 12,000 suppliers. This complex modeling challenge made it impossible to build and deploy a scalable solution in-house.
The forecasting challenge had resulted in overstock of certain products at some stores, and understock of other products elsewhere — leading to massive inventory costs and a poor customer experience.
From raw sales and SKU data to accurate predictions
Given the scope and complexity of this client’s supply chain, the desired predictive analytics project seemed insurmountable. Yet Pecan accomplished the task in just a few days, beating the existing models created over several months in both accuracy and actionability.
The predictions of this model were 80-90% accurate, partially due to Pecan’s data enrichment with regional weather, traffic, and economic data.
Unprecedented demand forecasting—and no more stockouts
With the Pecan platform, the customer deployed their models 10x faster without compromising model accuracy and usability. Since the business team could rely on Pecan to solve their long-awaited demand forecasting needs, the internal data science team could focus on custom business questions and use cases.
Pecan’s models consistently performed accurately, even with dozens of forecasts that spanned varying time horizons, configurations, business aspects, frequency, and other factors over several weeks.
Within the highest-value product categories, Pecan accurately forecast 50% of the SKUs with under 20% prediction error. Additionally, by leveraging Pecan’s proprietary data enrichment assets, the customer revealed unexpected business insights about the strongest predictors of demand for their top-priority SKUs.
Ultimately, the company saw overstock instances reduced by several orders of magnitude through using Pecan’s AI-powered predictive analytics to forecast demand. Pecan’s forecasts advanced the company’s efficacy and supply capability. Most importantly, the accurate, granular forecasts improved their customers’ experience by minimizing stockouts to historical low levels.