A creator posts a try-on haul at 8 p.m. By midnight, one shade of your bestseller is sold out, and two colors you barely stocked are the only thing anyone wants. I'm Hylke Reitsma, and I built Forthcast, an AI demand forecasting app for Shopify. I spend most of my time looking at what happens to a store's inventory in the hours after a video lands, and video commerce breaks almost every assumption a normal forecast is built on.
If you're running live drops, creator collabs, or shoppable video with a partner like Videowise, this post is about the inventory side of that playbook: how to stock for a spike you can't fully predict, how to read it early, and what to do with the surplus when the moment passes.
Standard demand forecasting assumes demand is reasonably smooth. It leans on your sales history, picks up weekly rhythms and seasonal trends, and projects the curve forward. That works for steady-state DTC. It falls apart the moment an outside event, a creator, a live drop, a clip that takes off, injects demand that was never in your history.
Three things break at once. The curve is front-loaded, so most of the demand arrives in the first hours instead of spread across the week. It concentrates on a handful of SKUs, often one specific color or size the creator happened to show, not your usual bestsellers. And it's driven by something your model has never seen, so a model trained on last quarter under-predicts the peak and then over-predicts the long tail after it.
The cost lands on both sides. The Corsten and Gruen worldwide study put the average retail out-of-stock rate around 8.3% on an ordinary day. During a spike, those stockouts concentrate on exactly the SKUs everyone is watching. And it isn't only lost sales. IHL Group estimated the global annual cost of stockouts and overstocks combined at about $1.7 trillion. Video commerce is unusual because it can hand you both problems in quick succession: an empty shelf during the peak, and a pile of surplus after it.
You rarely get days of warning. You get hours. The job is to spot the spike while you can still act on it.
Watch first-hour velocity against each SKU's normal baseline, not against yesterday's total. A hero SKU selling far above its usual rate in the first hour is a different event than a store-wide bump. Watch which variants are moving, because demand almost never spreads evenly across sizes and colors. And watch sell-through on the featured SKUs so you know how much runway you have before the shelf is empty.
This is where anomaly detection earns its place. Forthcast does demand-spike and anomaly detection at the SKU level, so the item pulling away from its own baseline gets surfaced while there's still time to expedite, cap orders, or swap the featured variant. Its reorder-point alerts are ranked by revenue priority, which matters when three SKUs go critical at once and you can only chase one first. You want to know which stockout actually costs money, not just which one happened first.
Planning a known campaign is a different job than reacting to a surprise. When you have a drop or a collab on the calendar, you can forecast deliberately and stock for it.
Forthcast forecasts per SKU with seasonality and trend detection, up to about twelve months out, so you can plan a whole season rather than a single week. For a specific campaign window, the useful move is to size safety stock to the window, not to the year: raise the service level on your hero SKUs for the drop, and keep the rest lean. Forthcast's safety-stock optimization is built for exactly that trade-off.
Lead time is where most drop plans come undone. A supplier's promised lead time and their actual delivered lead time are rarely the same number. Forthcast learns each supplier's real delivered lead time over time and sets reorder timing against reality, not the date on the PO. That's the difference between stock landing the week before your drop and landing the week after. When a SKU hits its reorder point, Forthcast drafts a purchase order straight from the SKU drilldown, and it handles multi-location inventory with bundles and kits, which is how a lot of creator collabs get sold in the first place.
The riskiest moment isn't the peak. It's the week after.
Two failure modes show up. In the first, you under-order the replenishment and stay out of stock through the tail, while demand is still elevated and buyers still care. Forthcast's lost-sales tracking shows you the revenue those stockouts actually cost, so the tail doesn't get written off as noise. In the second, you panic-order to a peak that has already passed. Weeks later, you're holding surplus of a trend SKU whose moment is over. That's the overstock hangover, and it's the part of video commerce nobody posts about.
This is why Forthclear is built into Forthcast. Forecast carefully so the hangover doesn't happen, and when it happens anyway, move the surplus straight from Forthcast without bolting on another tool. Over a few campaigns, Forthcast's forecast-accuracy reporting shows your per-SKU bias, so you can see whether you systematically over-order or under-order after a spike and correct the pattern instead of repeating it.
Black Friday and Cyber Monday amplify all of this, because you're stacking planned promos on top of creator moments that can fire at any time. We broke down the demand patterns we're seeing heading into this season in our 2026 Holiday Demand Outlook. Here's a month-by-month shape for BFCM 2026 that leaves room for both the planned drops and the surprises.
None of this needs enterprise tooling. Forthcast is $19.99 a month, flat, at any catalog size, with a free trial on the Shopify App Store. That's the point: the small and solo operators running the most interesting video campaigns shouldn't need an enterprise planning suite to survive their own success.
Video commerce rewards stores that can absorb a spike without going dark, and without drowning in leftovers afterward. That's a forecasting-and-replenishment problem: forecast the featured SKUs to the campaign window, watch first-hour velocity against each SKU's own baseline, and clear the tail before it becomes dead stock. If you want to go deeper on the mechanics, here's how we think about AI inventory forecasting for Shopify.
Work backward from the event date using your supplier's actual delivered lead times, not the ones promised on the PO. For hero SKUs, add safety stock sized to the drop window specifically, and keep everything else lean so you're not funding buffer you don't need. If the collab sells in bundles, plan the components together, not as separate SKUs.
Move it before it becomes dead inventory. First confirm the tail is genuinely over by checking whether demand is still elevated. If it isn't, clear the surplus quickly. Forthcast has Forthclear built in, so you can move overstock straight from the same place you forecast, instead of carrying trend stock for a year.
Standard forecasting handles steady demand well but under-reacts to spikes driven by outside events, because those events aren't in your sales history. The workable approach is to plan safety stock for known campaign windows in advance, and use SKU-level spike and anomaly detection to catch the surprises early, while there's still time to act.
.jpg)