An Amazon inventory forecast, whether it runs on a regression, a seasonal decomposition, or a language model wrapped around either, needs four inputs from your books before it can say anything useful: sales velocity measured only on days you were in stock, lead time from purchase order to units available at FBA, inbound stock already on its way, and a seasonal pattern from prior years. It also needs one input from the accounting side that most forecasting discussions skip: the FIFO cost of the units you are about to reorder, because a forecast that says "order 1,200 units" without saying what those units cost in cash is half a forecast. Get the four inputs wrong and the smartest model returns a confident number that is off by a week of sales. The demand forecasting glossary covers the vocabulary; this piece is about the data.
Velocity is units per day. The naive version divides units sold over a window by days in the window. The problem is that a stockout day sells zero, and the naive calculation counts it as a day of demand rather than a day of no supply. Every stockout drags measured velocity down, which lowers the reorder point, which causes the next stockout. The spiral is self-reinforcing and it hides in a clean-looking average.
Illustrative numbers. Take a seller with a stainless tea infuser. Over the last 90 days, 780 units sold. The product was out of stock at FBA for 12 of those days.
Lead time is 55 days: 40 days from PO to the supplier's dock, 15 days inbound to FBA including receiving. Reorder point before safety stock is velocity times lead time.
The gap is 73 units, or 7.3 days of real demand. A seller who reorders at 477 runs out a week before the shipment lands, every cycle, and the next 90-day window contains another 7 stockout days that pull the naive average lower still. The lost margin from those days is its own calculation; here the point is narrower. The velocity figure must be computed from inventory records that know which days were zero because of supply, and that requires on-hand by warehouse by day, not a sales report.
Amazon adds a cost to running thin. Its Seller Central page on 2026 FBA fulfillment fees describes a low-inventory-level fee on standard and bulky products when inventory falls below 28 days of supply relative to demand. A reorder point built on corrupted velocity causes stockouts and drifts the product into that fee band on the way down.
Lead time has to be measured, not assumed, and measured from the PO date to the date the units were available to sell at FBA, which includes Amazon's receiving time. A supplier that quotes 30 days and delivers in 40, plus two weeks of inbound and check-in, is a 55-day lead time. Use the last four POs and take the longest, not the average, unless you are adding safety stock separately.
Inbound stock is the quantity already ordered and not yet received. A forecast that ignores it double-orders. A forecast that counts it as on hand ignores the three weeks it is still on the water. ConnectBooks tracks transfers as in-transit until received, so the inventory reports show on-hand, in-transit, and available as separate figures, which is the split a reorder calculation needs.
A Q4 product with a velocity of 10 a day in September may run at 25 a day in November. A model trained on the trailing 90 days does not know that. It needs last year's November, on the same SKU or a close proxy, and it needs to know whether last year's November was itself depressed by a stockout.
The ConnectBooks restock report is driven by sales history and velocity, with lead times and inbound stock factored in. It does not yet factor seasonality. A seller preparing for Q4 takes the report's output and adjusts it by hand for the seasonal multiple, using the method in how to set reorder points for Q4. That is a limitation worth stating plainly, because a restock number that looks authoritative in September and ignores November is dangerous in a way a blank cell is not.
The units you reorder will not cost what the units on the shelf cost. Suppose the infuser's current inventory layer is 550 units at $7.20 landed. The supplier's new quote, with a freight increase, is $8.10. A reorder of 1,200 units is:
The forecast that says "order 1,200" is saying "spend $9,720 in about 30 days," and the seller planning cash against the $7.20 figure is $1,080 short. The margin effect follows later. At 10 units a day, the 550-unit layer at $7.20 lasts 55 days; after that, every unit consumes the $8.10 layer and margin per unit drops $0.90. A profit forecast for the quarter after next has to switch cost layers on the day the old layer runs out, which is exactly what FIFO does and exactly what an average-cost system smears across both periods.
This is why inventory forecasting belongs next to the books rather than in a standalone tool. The velocity comes from inventory records, the lead time from PO history, the inbound from transfers in transit, and the cost from FIFO layers. Each lives in the accounting system if the accounting system tracks inventory at the SKU and warehouse level.
AI helps most in the cleaning, not the predicting. Identifying which of the 90 days were stockout days, spotting a lead time that has drifted from 40 to 55 without anyone updating the master record, flagging that last November's velocity was itself corrupted by a two-week outage: these are pattern-finding tasks on reconciled data, and a model does them faster than a planner with a spreadsheet. How AI changes inventory planning covers that shift.
Two honest boundaries. Crunch, the analytics AI inside ConnectBooks, does not forecast. It analyzes what happened: which products, periods, and cost lines moved, and why, including inventory and storage. Ask it whether a product's sales fell in August and it will find the weeks and the products responsible and explain what moved. Ask it what to order for November and it will not answer, because that is not what it does. The restock report is the reorder tool, with the seasonality caveat above.
And no model, in ConnectBooks or anywhere else, knows the things that are not in the data. The supplier's minimum order quantity. The factory's holiday closure. A competitor's launch that will halve your velocity in October. A tariff change that moves landed cost after the PO is placed. A forecast is a statement about the past extended forward, and the person extending it is accountable for the parts the past does not contain. The model produces a number. The seller signs the purchase order.
Running an e-commerce business comes with plenty of challenges, but ConnectBooks is here to make your life easier. With real-time insights, seamless integrations, and detailed tracking of your profitability and inventory, you can stay ahead of the game. Whether you’re selling on Amazon, Shopify, Walmart, TikTok or eBay, ConnectBooks helps you manage your finances with 100% accuracy and confidence, so you can focus on growing your business.
Ready to level up? Start making smarter, data-driven decisions every step of the way. Try ConnectBooks Free Today or Schedule a Demo
