Oakbend Beverages: caught a TikTok demand spike 4 weeks before it hit POS.

The problem: craft beverage SKUs driven by social content, not seasonal patterns

Oakbend Beverages produces a line of 28 sparkling water flavors, ranging from established classics (lemon-lime, plain) to trend-driven seasonal and limited-edition flavors (cucumber-mint, yuzu-ginger, black cherry-hibiscus). Their forecast challenge is asymmetric: the classic SKUs are highly predictable, while the trend-driven flavors are extremely sensitive to social content and can experience 200–400% demand spikes within a 6-week window when an ingredient combination trends on social media.

Their ERP system — SAP IBP — was calibrated on 2 years of scan data per SKU. For new and limited-edition flavors, the historical data volume was insufficient to build a reliable seasonal model. For established flavors, the model worked reasonably well. But Oakbend's revenue growth was concentrated in the trend-driven flavors — and those were the ones where forecast accuracy was worst: 58–65% MAPE on trend-sensitive SKUs.

The operational consequence was predictable: when a flavor trended, they'd stock out within 3 weeks. By the time the sales team escalated to supply planning, the production window for that specific SKU had passed. They'd miss 4–6 weeks of peak demand, and by the time they had inventory back, the trend had partially subsided. The revenue opportunity from trend-driven SKUs was being systematically underdelivered.

The signal fusion deployment

Heatvelo's deployment for Oakbend focused primarily on social trend velocity as the key signal category for their trend-sensitive SKUs. The configuration mapped each of Oakbend's 28 sparkling water SKUs to their core ingredient combinations, then tracked mention velocity for those ingredient combinations across short-form video and recipe content platforms.

The critical insight from the baseline calibration: cucumber-mint as an ingredient combination had an unusually high "trend correlation coefficient" — meaning when cucumber-mint beverage content velocity accelerated, Oakbend's cucumber-mint sparkling water SKU saw a demand response within 2–3 weeks with high consistency. This correlation was calibrated from 18 months of prior trend/demand data. The model assigned a 0.87 correlation coefficient to the cucumber-mint pairing — the highest in Oakbend's SKU set.

During week 1 of the Heatvelo pilot, the social trend velocity model flagged cucumber-mint mention velocity accelerating at 4.2x its baseline rate. The signal attribution in the forecast output showed: "Social velocity flag: cucumber-mint ingredient trending. Projected demand impact: +220–340% for SKU OB-CM-16OZ, weeks 3–7." Oakbend's supply planning team received this forecast at 6am before their morning standup.

By week 2, the supply team had increased the cucumber-mint production run by 280% of planned volume. By week 5, when POS data finally started showing the demand surge, Oakbend had inventory ready at their top 3 retail distribution partners. They met the demand wave instead of chasing it.

The outcome: a stock-out avoided on a $2.3M revenue window

The cucumber-mint demand spike peaked at +340% above baseline in week 6 and began mean-reverting in week 8. Oakbend met the demand at their primary grocery accounts through weeks 5–8. Based on prior trend events where they had stocked out (and comparing to their other trend-sensitive SKUs that still ran under ERP-only forecasting), the supply team estimated the cucumber-mint pre-production decision protected approximately $2.3M in revenue from the trend window — revenue that would have been lost to stock-outs if they'd waited for POS data to show the signal.

Overall pilot accuracy: 89% MAPE across Oakbend's full 180-SKU catalog (vs. 72% ERP baseline). The improvement was concentrated in the trend-sensitive flavor SKUs, where MAPE improved by 28–34 points. Classic SKUs improved by only 6–9 points — consistent with the expectation that social velocity signals add less value to stable, predictable demand patterns.