The 12-week rolling forecast serious Amazon operators run in 2026: reorder points, safety stock, lead-time math, and the Days of Inventory thresholds that stop stockouts before they bleed rank.
Free for life
Nova MCP connects your account to Claude, ChatGPT, Gemini or any MCP-ready assistant. Sign up and it stays free for life on your account.
Connect your data
Explore more expert insights to grow your Amazon business
DOI formula, category benchmarks, color-coded thresholds, and reorder point strategies. The single metric that controls both stockout risk and storage costs.
Compare SoStocked, RestockPro, InventoryLab, Forecastly, and more. Which inventory tool fits your business: forecasting, profit analytics, or restock automation?
Every KPI that matters for Amazon sellers in 2026, organized by category with real benchmarks. Revenue, profitability, advertising, inventory, and strategic metrics with healthy ranges.
Forecasting Amazon inventory is not a procurement problem. It is a profit problem. Every unit short of demand pushes your listing down the search ranks, triggers a low-inventory-level fee on standard-size SKUs, and burns paid traffic against an out-of-stock badge. Every unit over demand pays monthly storage, then aged-inventory surcharges starting at 181 days. The job is to thread the needle, week after week, across your entire catalog. Here is how serious operators do it in 2026.
Our take
Pick your top 20 SKUs by revenue. Compute a 12-week forecast using the basic reorder-point formula, with safety stock tied to lead-time variance not a flat percentage. Then check actual vs forecast every Friday for 8 weeks. The point is not the model. The point is the discipline of comparing forecast to reality every week.
Best fit if
Skip if
General retail forecasting boils down to one question: how many units will I sell, by when, and with what variance? Amazon adds three constraints that change the answer.
The first is the low-inventory-level fee, introduced in April 2024 for standard-size FBA SKUs with historically low days of supply. Under-forecast and you pay a per-unit penalty on top of every fulfilment fee. The second is restock limits set by Capacity Manager, which caps how much inbound volume you are allowed regardless of what your forecast says. Forecast 6,000 units, get approved for 4,000, and the remaining 2,000 either sits in a 3PL or pays an overage fee at auction. The third is the aged-inventory surcharge schedule that begins at 181 days and escalates sharply past 271 and 365 days.
Marketplace Pulse has been documenting the broader pattern for years: Amazon fees only go up, and the inventory-side fees are now where most of the variance hides. A forecast that ignores them is technically a forecast and operationally a guess.
Generic retail forecasting vs Amazon inventory forecasting
| Variable | Generic retail | Amazon FBA |
|---|---|---|
| Lead time | Supplier + freight | Supplier + freight + Amazon receive (7-21 days) |
| Penalty for stockout | Lost sale + customer churn | Lost sale + search-rank decay + low-inventory fee |
| Penalty for overstock | Carrying cost | Storage + aged-inventory surcharge (181+ days) |
| Capacity ceiling | Warehouse space | Capacity Manager restock limit |
| Demand signal | POS data | SP-API sales + BSR + session data |
| Refresh cadence needed | Weekly | Hourly during peak season |
Nova insight
Compare forecast units with actual units over the same window. Set safety stock from lead-time variability, and model Amazon receive time separately from supplier and transit lead time.
The math is older than e-commerce. What changed is the inputs. The base reorder-point formula is:
Reorder Point = (Daily Sales x Lead Time) + Safety Stock
Where safety stock = Z x sigma_demand x sqrt(lead time), and Z = 1.65 for a 95 percent service level
eComEngine's reorder-point playbook walks through the same structure for FBA-specific use, and supplychainmath.com's safety stock guide covers the underlying Z-score table if you want to tune for a 90, 95, or 99 percent service level. The honest answer for most brands: 95 percent is the right anchor. 99 percent doubles your safety stock for marginal stockout-rate gains.
The number that breaks most spreadsheets is lead time. On Amazon, total lead time has four components, and treating them as one number is the single biggest accuracy leak.
Add it up and a "60 day lead time" assumption is more like 75 to 110 days end-to-end. Forecast against the short number and you stock out four weeks before the reorder lands.
Deep dive
Amazon Days of Inventory (DOI): formula, benchmarks and color-coded thresholds
One forecast cannot answer every question. The brands that get this right run three in parallel, each with a different horizon, refresh cadence, and decision use.
Nova insight
The tell that a brand has not yet built the three-layer view: the same spreadsheet is used to decide both "should we reorder unit 12345 this week" and "should we keep this brand at all next year". Those are different questions answered by different math.
Nova surfaces sell-through, Days of Inventory, BSR trends and P&L impact at SKU level across 23 marketplaces. The forecast model is yours. The inputs are clean.
Forecasting tooling has exploded in the last three years. Most of it solves the wrong problem. Better algorithms cannot save a forecast running on dirty or stale inputs. These four data points, refreshed at the right cadence, account for the majority of accuracy gains.
The four inputs and where they come from
| Input | Why it matters | Refresh cadence |
|---|---|---|
| Sell-through velocity per SKU | Direct demand signal, last 30-90 days | Hourly |
| Days of Inventory (DOI) | Reorder trigger and overstock alarm | Daily |
| BSR trend (7D, 30D, 90D) | Leading indicator of demand shifts | Daily |
| Seasonality index per SKU | Year-over-year monthly multiplier | Monthly |
Nova surfaces all four natively. Sell-through and DOI sit in the FBA analytics view. BSR trends live in the BSR tracker. Seasonality indexes come out of the historical P&L data in the profit and loss module. None of that is forecasting per se. It is the input layer that any forecast model worth its name has to consume.
Related read
Amazon FBA restock limits: Capacity Manager, bidding and 6 ways to maximise allocation
Pre-Nova setup was a Google Sheet refreshed every Monday by the operations lead with last 30-day velocity and a flat 21-day safety stock. Three of six SKUs went out of stock at least twice per quarter, costing roughly $180K in lost revenue and a 12-position search-rank slide on the worst offender. The brand also paid the low-inventory-level fee on two SKUs in Q1 2026 for the first time.
Stockout incidents / qtr
Forecast accuracy (8wk)
Days of Inventory band
Low-inventory-level fee
How: The model itself did not change much. What changed was the input refresh cadence (weekly to daily) and decomposing lead time into its four components. Both shifts came directly from putting Nova between the spreadsheet and the raw Amazon data.
A forecast can be sound on paper while execution breaks at three points.
Nova insight
Check whether the TACoS target assumes full stock availability, and define when campaigns should pause as DOI approaches the reorder point. Connect the two systems and ad spend should automatically throttle when inventory falls below a defined floor.
The trap is wanting a perfect forecasting system before you have a workable one. The brands that get this right ship something rough in week one and iterate. BigCommerce's FBA overview covers the foundational mechanics if you are still bedding down the basics.
Inside three months that loop produces a forecast accuracy gain that compounds: tighter safety stock means lower DOI on average, lower DOI means fewer aged-inventory surcharges, fewer surcharges mean higher contribution margin per SKU. The same operating discipline that runs your weekly cash review cadence also runs your inventory loop.
Amazon inventory forecasting is not about picking the cleverest model. It is about feeding a workable model the right inputs at the right cadence, then comparing forecast to actual every single week until your accuracy band tightens. Most brands jump straight to the model and starve it of fresh data. The result is a sophisticated forecast running on month-old velocity, missing the BSR jump that signalled the demand shift two weeks ago.
Build the inputs first. The math will look after itself.
Join thousands of successful sellers who use Nova Analytics to make data-driven decisions and maximize their profits.
Claude
Gemini
ChatGPT