You’re splitting inventory across Amazon FBA, your own warehouse, and Walmart WFS. Getting it wrong bleeds cash on both sides — aged storage fees or lost sales and tanked search rankings. Here’s how to forecast properly.
This post is part of our E-Commerce Multi-Channel Operations Guide.
If you sell on one channel, inventory forecasting is straightforward: look at velocity, account for lead time, reorder. When you sell on Amazon, Shopify DTC, and Walmart simultaneously, every assumption in that model falls apart.
The core problem is inventory fragmentation. Amazon FBA requires you to ship inventory to their warehouses weeks before you need it. Walmart WFS has its own inbound requirements. Your DTC channel ships from your 3PL or in-house warehouse. Each pool of inventory is physically separated, with different replenishment cycles, different storage cost structures, and different consequences for getting it wrong.
A single-channel forecast treats your total inventory as one pool. Multi-channel reality means you’re managing three or four separate pools, each with its own demand curve, each competing for the same supply from your manufacturer. Forecast at the aggregate level and you’ll have too much in one channel and too little in another — even when your total inventory position looks fine.
Overstocking and stockouts don’t hurt equally across channels. Overstock at Amazon FBA triggers aged inventory surcharges — after 181 days, you’re paying $6.90 per cubic foot on top of monthly storage. Overstock at your 3PL costs far less per unit. Meanwhile, a stockout on Amazon doesn’t just lose today’s sales — it drops your organic search ranking, which takes weeks to recover. A DTC stockout loses the immediate sale but doesn’t compound.
This asymmetry means you can’t apply the same safety stock logic to every channel. The cost of being wrong is different on each platform, and your forecasting model needs to reflect that.
Each channel has its own demand pattern, and treating them as interchangeable is where most forecasting models go sideways.
Amazon demand is heavily influenced by search rank, PPC spend, and Buy Box share. A product selling 20 units/day can jump to 50 units/day if you win a competitive keyword or a competitor goes out of stock. It can also drop to 8 units/day if your listing gets suppressed or a new competitor launches at a lower price. Amazon velocity is high but volatile.
Key signals to track: daily unit sessions percentage (conversion rate), PPC spend and ACoS trends, competitor pricing shifts, Buy Box win rate, and any upcoming deals or Lightning Deal commitments.
Your DTC channel is directly tied to your marketing spend and campaigns. Demand is more predictable day-to-day but spikes hard around email drops, influencer posts, and paid ad pushes. The signal here is your marketing calendar. If you’re planning a 20%-off sale next month, your forecast needs to reflect that 2–4x normal volume during the promo window.
Walmart Marketplace is typically lower volume than Amazon but more consistent. Velocity ramps slowly as your items gain traction in Walmart’s search algorithm. The demand signal is more linear — look at trailing 30-day and 60-day averages. But watch for Walmart’s own promotional events (Rollback pricing, Walmart+ deals) which can spike demand unexpectedly.
The most common mistake: forecasting total demand across all channels, then splitting inventory proportionally. This ignores that each channel’s demand is driven by different factors. Forecast each channel independently, then reconcile against your total available supply. Channel-level forecasting first, allocation second.
Safety stock is the buffer between your forecast and reality. Get it right and you absorb demand variability without tying up excess cash. Get it wrong and you’re either constantly scrambling or constantly paying storage fees on units that aren’t moving.
For each channel, calculate safety stock independently:
Safety Stock = Z × σd × √LT
Where Z is your service level factor (1.65 for 95% service level, 2.33 for 99%), σd is the standard deviation of daily demand, and LT is lead time in days.
Practical example: Your product sells an average of 15 units/day on Amazon with a standard deviation of 5 units. Your manufacturing lead time plus FBA inbound processing is 45 days. At a 95% service level:
Safety Stock = 1.65 × 5 × √45 = 1.65 × 5 × 6.71 = 55 units
For the same product on DTC selling 8 units/day with a standard deviation of 3 units and a 14-day lead time from your 3PL:
Safety Stock = 1.65 × 3 × √14 = 1.65 × 3 × 3.74 = 19 units
Same product. Wildly different safety stock requirements. This is why single-number forecasting doesn’t work.
Lead times vary dramatically by fulfillment channel, and underestimating them is the single biggest cause of stockouts:
Your reorder point for each channel = (Average Daily Sales × Lead Time in Days) + Safety Stock. For the Amazon example above: (15 × 45) + 55 = 730 units. When your FBA-sellable inventory hits 730 units, it’s time to send the next shipment. Build this as an alert in your inventory tool or spreadsheet — don’t rely on checking manually.
Forecast and set safety stock per channel, not across channels. The cost of being wrong, the lead time, and the demand volatility are all different. One aggregate number guarantees you’ll overstock somewhere and stockout somewhere else.
Steady-state forecasting is hard enough. Layer in seasonality and promotions and the complexity multiplies — because each channel has its own seasonal pattern.
Amazon Q4 demand can be 3–5x normal for many categories. Your DTC site might see a smaller lift (1.5–2x) unless you’re running aggressive Black Friday campaigns. Walmart’s peak is similar to Amazon’s timing but typically at lower volume. The mistake is applying a single seasonal multiplier across all channels.
Use the prior year’s channel-level data to build seasonal indices. If your Amazon sales in November were 3.2x your average monthly sales last year, start with a 3.2x index for this November. Adjust for year-over-year growth rate and any known changes (new competitor entries, price changes, expanded PPC budgets).
Promotions require a different forecast method entirely. You’re not predicting organic demand — you’re predicting response to a specific offer on a specific channel.
Amazon’s FBA storage fees triple from October through December ($2.40/cubic foot monthly for standard-size). And their receiving times slow to a crawl. If you’re sending Q4 inventory to FBA, it needs to arrive by late September. Miss that window and you’re either paying for express inbound shipping or losing the most profitable weeks of the year. Build your Q4 forecast and ship schedule by August.
Even with good forecasts, reality diverges. Your Amazon listing gets suppressed for a week and sales drop 60%. Your DTC site goes viral from a TikTok mention and you blow through three weeks of inventory in four days. Now you have too much inventory in one place and not enough in another.
Sometimes — but it’s expensive and slow. Creating a removal order from Amazon FBA takes 10–14 days and costs $0.97–$1.04 per unit. Then you need to ship those units to your 3PL or to Walmart. By the time the inventory arrives, the demand signal that triggered the rebalance may have changed.
The better approach is to rebalance at the replenishment stage, not the fulfillment stage. When it’s time to order from your manufacturer, adjust the allocation split based on current velocity and weeks-of-cover by channel. If Amazon is running hot and DTC is slow, shift more of the next production run to FBA. This is cheaper and faster than physically moving inventory between warehouses.
Calculate weeks-of-cover for each channel every week: current sellable inventory divided by average weekly sales. If Amazon has 4 weeks of cover and DTC has 12 weeks, your next allocation should skew toward Amazon. Set thresholds — anything below 4 weeks of cover triggers an urgent reorder; anything above 10 weeks means you’re overstocked and should throttle inbound.
Don’t try to move inventory after it’s in the wrong place — it’s too slow and too expensive. Instead, rebalance at the next replenishment cycle by shifting allocation percentages based on weeks-of-cover by channel.
You don’t necessarily need expensive software to forecast well. But you do need something beyond gut feel and a single spreadsheet tab.
No single tool covers all channels equally well. Most multi-channel operators end up with a primary tool (Inventory Planner or Carbon6) supplemented by spreadsheets for the reconciliation and allocation layer. That’s normal. The goal isn’t one perfect tool — it’s one reliable process.
Below $5M in revenue, demand planning is a founder or ops manager responsibility. The volume doesn’t justify a dedicated headcount. Use the tools and formulas in this post, review weekly, adjust monthly.
Between $5M and $15M, you need someone spending 10–15 hours per week on demand planning. That could be a fractional operator, a senior ops hire with demand planning as one responsibility, or a well-configured tool doing the heavy lifting with human review.
Above $15M with 100+ SKUs across three channels, consider a dedicated demand planner. At that complexity, forecasting errors compound fast and the cost of getting it wrong justifies the headcount. A good demand planner at this stage saves 3–5x their salary in reduced overstock and prevented stockouts.
Not sure where your demand planning stands? Our multi-channel operations guide covers the full operational stack, including how forecasting fits into the broader workflow.
30-minute discovery call. We’ll map your stack and show you where the margin is hiding.
Schedule Call