100-Store Grocery Chain: 76% Write-Off Reduction in 30 Days

The Challenge

A 100-store regional grocery chain (operating under the Dobrininsky / Natali Plus banner) was experiencing write-off rates of 5.8% across fresh produce categories — nearly four times the industry benchmark of 1–2%. The chain's existing ERP-based replenishment relied on static reorder points and weekly manual review by purchasing managers, a process that consistently over-ordered slow-moving SKUs and under-ordered high-velocity items.

Shelf availability had fallen to 70%, directly causing stockouts that drove customers to competing chains. The operations team estimated $170,000 in annual write-off losses before accounting for lost sales from empty shelves.

Why Manual Replenishment Was Failing

The chain's purchasing team was skilled and experienced, but they were working against structural limitations of their tools:

• Batch data updates — inventory counts updated nightly, meaning orders were placed on 24-hour-old data in a fast-moving fresh environment.
• No SKU-level demand curves — every fresh SKU was governed by the same reorder logic, ignoring the radically different demand profiles of milk vs strawberries vs packaged salads.
• No feedback loop — when an order resulted in a write-off, that signal was never fed back to improve the next order. The same errors repeated every week.

The Pilot Setup

The pilot ran across all 100 stores simultaneously over 30 days. Bright Minds AI integrated directly with the chain's existing ERP via API, pulling real-time sales velocity, on-hand inventory, and supplier lead times. No data warehouse or data team was required.

The AI model was trained on 18 months of historical sales data, stratified by store format, location type, and product category. From day one, the model generated daily replenishment recommendations that the purchasing team reviewed and approved before submission.

Key configuration decisions:

• Fresh produce, dairy, and bakery were prioritised in the first two weeks
• Minimum order quantities and supplier constraints were integrated into the recommendation engine
• A dedicated dashboard showed write-off rates and shelf availability per store, updated daily

Results: Verified at 30 Days

All metrics were measured against a pre-pilot baseline taken from the 30 days immediately preceding the pilot start date.

Write-Off Rate: 5.8% → 1.4%

Fresh produce write-offs fell from 5.8% of category revenue to 1.4% — a 76% reduction. The largest gains came in highly perishable categories: berries, cut salads, and fresh bakery. These three subcategories accounted for 68% of the total write-off reduction.

Shelf Availability: 70% → 91.8%

Shelf availability improved from 70% to 91.8% — a 21.8 percentage point gain. This was driven by AI identifying systematic under-ordering patterns across 34 high-velocity SKUs that the manual process had consistently under-ordered on peak shopping days (Thursday–Sunday).

Sales Growth: +24% in 30 Days

Total fresh category sales grew 24% over the same period. Management attributed this primarily to the availability improvement — customers who had previously abandoned shopping trips due to empty shelves returned and found product in stock. No promotional activity or pricing changes occurred during the pilot period.

What the Operations Team Said

"In the first week, our purchasing manager said the recommendations looked too aggressive. By week three, she was asking us to increase the model's authority. The numbers were right and she knew it."

— Operations Director, Dobrininsky / Natali Plus

What Happened Next

Following the 30-day pilot, the chain moved to full deployment across all 100 stores and expanded the AI model to include ambient grocery, beverages, and household categories. The purchasing team now spends approximately 30 minutes per day reviewing AI recommendations rather than 3–4 hours building manual orders.