Optimizing replenishment via out-of-stock analytics

About the Project

Gopuff observed a significant latency in the replenishment targets for high-performing SKUs, leading to limited availability of high-demand items in their busiest warehouses. This "slow ramp" purchasing behavior resulted in a failure to meet the growing demand, particularly for on-demand grocery.

• Inadequate Item Availability: There was a scarcity of top 40% highest selling SKUs in high volume warehouses and it was unclear if this was a vendor related issue or a purchasing related issue.
• Purchasing Model Shortcomings: The purchasing model was failing to account for true demand, often underestimating the quantities needed. The need for an effective way to quantify "lost demand" to better inform purchasing decisions.
• Concerns from site-level staff: The store footprints meant that inventory couldn't be sent across all items due to cost and space considerations.

Actions Taken:

• Data Investigation: A deep dive was conducted to identify patterns in which product-location combinations were most frequently out of stock.
• Identifying Key Factors: Found that high national Out-of-Stock (OOS) level was driven by high velocity items and sites within a subset of 200 fast growing sku's.
• New Metric Development: A new way of measuring out of stock severity was developed--the "lost demand" metric projected the potential sales that could have been achieved if products were sufficiently stocked using a proportion of site level sales of an item and how out of stock that item was.
• Cross-Functional Meeting: A new weekly cross-functional meeting was established where the biggest opportunities were force-ranked and addressed by buyers and merchandisers to be reported back on the following week.
• Model Integration: The lost demand metrics were incorporated into the model used to set replenishment targets.

Results:

• Significant Sales Impact: Implementation of the new strategies resulted in an unexpected 4% increase in monthly sales across the US business.
• Model Improvement: Data science worked to adjust the replenishment model to incorporate the nuances of the lost demand metric.
• Enhanced Business Practices: The "lost demand" metrics and weekly meetings were ingrained into regular business practices, allowing anomalies to be quickly surfaced and opportunities to be incorporated into the replenishment model.

‍