Aggregate forecasts looked accurate. Subclass forecasts were 80% MAPE, exactly where inventory is bought. Profitmind corrected it at 66x ROI.
The retailer's merchandise plan had a forecast error nearly twice that of AI at the subclass level (80% MAPE versus 35%) but aggregate-level errors cancelled out, creating a misleading picture of accuracy. This mattered operationally because inventory is purchased, allocated, and marked down at the subclass level, not the aggregate. Over-forecasted subclasses drove excess inventory and markdown exposure; under-forecasted subclasses created stockouts and lost sales. The plan was producing false confidence precisely where accurate forecasting had the most commercial consequence.
Profitmind's Planning Agent automated demand forecasting across approximately 3,000 subclass plans per week. Over a 38-week evaluation period, forecast accuracy was validated at every level: by department, by forecast horizon spanning 1 to 21 weeks, and by volume decile. The model trained continuously on this retailer's specific data, improving from approximately 70% MAPE in early calibration weeks to 25–35% MAPE at steady state. Statistical significance was confirmed at p < 10⁻¹⁵⁴ across the evaluation period.
Profitmind outperformed the merchandise plan in 75% of head-to-head comparisons. Annual working capital savings from reduced overbuying reached $8M. The ROI ranged from 66x to 97x depending on adoption rate, and even at just 11% adoption, the retailer achieved 22x ROI. Every 10 percentage points of additional adoption adds approximately $1M in annual value from overbuy reduction alone, establishing a clear and measurable path to scale.