Inventory stockouts are the primary performance bottleneck for this account. When top-selling SKUs go out of stock, ROAS suffers directly and immediately — there is no amount of ad optimization that compensates for zero available units. This article documents the stockout situation observed in early March 2026, the root causes, and the AI-driven shipment model being developed to prevent recurrence.
See also: [1] | [2]
As of the March 4, 2026 weekly call, ROAS was 3.35 (up slightly from 3.3 the prior week), but performance remained below trend. The cause was confirmed as inventory, not ad quality.
| SKU | Units in Stock | 30-Day Sales Velocity | Units Inbound |
|---|---|---|---|
| Old World White Popcorn | 0 | 425 units/mo | 2,400 (pending pickup) |
| Kidney Beans (5 lb) | 0 | ~72 units/mo | 288 (in transit) |
| Various 25 lb SKUs | 0 | Low velocity | — |
The Old World White Popcorn stockout is particularly damaging given its sales velocity. With 425 units/month in normal conditions, every day out of stock represents meaningful lost revenue and suppressed organic rank.
Inbound shipments scheduled for March 9 were delayed due to unresolved pickup logistics on Carly's end. Units were in the system but not yet moving. This is a process gap — shipments need to be confirmed in motion, not just scheduled.
Even with inventory constraints, several metrics were trending in the right direction:
The interpretation: the underlying account health is improving. Resolving inventory issues should produce a step-change improvement in ROAS, not just incremental gains.
Gilbert is building an AI tool to automate optimal shipment quantity calculations and prevent future stockouts. The model uses three inputs:
The model outputs a recommended shipment quantity that accounts for the time gap between ordering and availability, ensuring buffer stock is maintained.
Demo planned for the following week's call.
Manual inventory management at scale is error-prone. The current process relies on someone noticing a stockout or low-stock condition and reacting. The AI model shifts this to proactive, data-driven reorder triggers — the same logic as a reorder point system, but automated and continuously recalibrated against actual velocity.