Ridgeline Snacks: from 68% to 91% forecast accuracy on seasonal outdoor SKUs.

The problem: outdoor snacks with weather-driven demand and an ERP that couldn't see it

Ridgeline Snacks builds trail mix and natural snack products positioned for outdoor recreation occasions — hiking, camping, cycling, and general active lifestyle use. The demand pattern for their top 80 SKUs is directly correlated with regional outdoor activity, which is itself driven by weather conditions. A warm, dry weekend in the Upper Midwest drives measurably higher trail mix velocity in Minneapolis-area retailers. An unexpected cold snap in mid-May flattens velocity two weeks earlier than seasonal index would predict.

Their ERP demand model — Oracle Demantra, calibrated on 3 years of scan data — was using seasonal indices that averaged weather effects over multiple years. It couldn't respond to a specific weather event in real time. By the time the POS data showed a demand spike in week 1, their production run was already committed through week 4. Ridgeline's supply planning team described their forecasting problem exactly: "we're always chasing the season, never ahead of it."

Their 12-week MAPE over the prior year was 68% — meaning the average forecast deviation from actual was 32 percentage points. On their fastest-moving seasonal SKUs, the MAPE was worse: 58% (42-point average deviation). Their safety stock buffer for weather-sensitive SKUs was 22% of average weekly volume — essentially a hedge against a forecast they couldn't trust.

The signal fusion deployment

Heatvelo connected to Ridgeline's Oracle Demantra instance via API and pulled 24 months of SKU-level scan data for their top 200 SKUs by volume. The initial baseline calibration took 8 days — longer than average because Ridgeline's SKU mix included a number of regional variants (West Coast trail mix formulations vs. Midwest formulations) that required separate baseline models.

The signal fusion configuration for Ridgeline's outdoor snack SKUs prioritized three signal categories: (1) Regional temperature deviation from seasonal baseline, at zip-code level for their top 8 distribution markets; (2) Regional precipitation events, specifically tracking dry-weekend probability as an outdoor activity enabler; (3) Social trend velocity for outdoor activity content, tracking hiking/camping/cycling content creation velocity as a leading indicator for outdoor occasion demand.

The weather signal lead time for outdoor snack demand proved to be 7–10 days — weather-forecast data for 2 weeks out translated into demand forecast confidence for weeks 2–4. The social activity signal provided a longer lead time: outdoor activity social content typically builds 3–5 weeks before peak season activity dates, giving Ridgeline's supply team a 3-week window to adjust production commitments before POS showed any demand change.

The outcome: 23 percentage points of MAPE improvement

Over the 12-week pilot period, Ridgeline's Heatvelo forecast tracked actual demand at 91% MAPE accuracy — a 23-point improvement over their Oracle Demantra baseline measured over the same SKUs in the prior equivalent period. The improvement was concentrated in their weather-sensitive SKUs: trail mix and outdoor portion-pack products improved by 28–31 MAPE points, while their shelf-stable pantry staple SKUs improved by only 8–12 points (expected — these SKUs have lower weather correlation).

The supply planning team used the signal flags in the Heatvelo forecast output to make one specific production decision during the pilot: a weather event flag in week 3 predicted a +28% demand deviation for their Pacific Northwest market trail mix SKUs in weeks 5–6. They increased the production run commitment 3 weeks before the demand spike materialized in POS data. That single decision prevented a stock-out on a SKU group that represented approximately $1.2M in annualized revenue through their Pacific Northwest distribution.