On July 1, a tumbler brand has to decide how many units to order for the holidays, and the order it is weighing will take $522,000 out of the bank before the first tumbler sells. The brand is invented, and every figure is set out so each sum can be checked. Ecommerce demand forecasting is how that number gets chosen: an estimate of how many units of each product customers will buy, by channel and by month, over the period the order has to cover.
Many brands this size have no analyst, and the CEO or COO builds the forecast in a spreadsheet the week the deposit is due. That is enough if the spreadsheet ends in the right place, because the forecast does less than it seems. The size of the order sets how low cash falls; demand only decides how long it stays down. So the useful forecast is the one that tells you whether you can carry the order if demand comes in low.
What is demand forecasting for ecommerce?
Demand forecasting for ecommerce is estimating the units customers will buy, per product, per channel and per period. It produces units by month, a range around them, and the date the units have to be on the shelf.
Two distinctions keep it honest. Demand is what customers would have bought; sales are what they bought while stock lasted, so a month with a stockout understates demand. And a demand forecast counts units, while a revenue forecast counts dollars and a cash forecast counts the dates money moves. Units come first, because price changes, discounts and channel mix distort a forecast built in dollars. The forecast tells you how many units the next order has to pay for; the rest of this post stays on the money side, whether that order can be paid for and how fast it pays back.
The data and the methods, channel by channel
Most of the data is already in the systems you sell through. Pull at least twelve months, two years if the product has a season, and keep a notes column for each month (a stockout, a creator video, a price change, an ad budget cut), because the numbers cannot say why March was strong.
| Channel | What to pull | What to mark | Method |
|---|---|---|---|
| Shopify | Units sold by product and day, from the sales reports or order export | Sitewide sales and large discount codes | Run rate times a seasonal index |
| Amazon | Units ordered by product and day, from the business reports in Seller Central | Out-of-stock days, promotions, Prime Day and Lightning Deal weeks (high case, not base) | Run rate times a seasonal index |
| Wholesale | Each account's past purchase orders and every open order | Shops that order through a wholesale marketplace; count their cash on that platform's payout terms, not net 30 | Build-up by account |
The three methods, in the order a spreadsheet would use them:
- Run rate. Average units per month over a recent stretch with no stockouts and no unusual promotions. On its own it suits a steady product and a horizon of a month or two.
- Seasonal index. Divide each month of last year's sales by last year's normal month. Multiply this year's run rate by each month's index. Growth is built in, because the base is this year's rate and only the shape comes from last year.
- Build-up by account. For wholesale, count confirmed orders in full, and count accounts that have not ordered yet at their usual size, times the share you expect to come back. A few dozen stores ordering or not ordering moves wholesale more than any trend.
A worked example: the holiday order, step by step
The tumbler is a 20-ounce insulated model at $34 on Shopify and Amazon and $17 wholesale to independent gift and home shops. It comes from one factory, about eleven weeks from deposit to the third-party warehouse (3PL). The order placed on July 1 lands in mid-September and has to cover October, November and December. The duty rate below is a placeholder; the real rate depends on the product's classification and origin.
Step 1: clean last year's history
Last year, Shopify sold 18,000 tumblers from April to June, but one week in May, when a creator video took off, sold about 6,000 more than a normal week. Take that extra out and the normal month is 4,000 units.
December needs the opposite correction. The store sold out on December 12 after selling 7,200 units, about 600 a day. Six more days to the last holiday shipping day would have sold about 3,600 more, and the slower weeks after the holiday added about 1,200 the year before. Clean December demand was about 12,000 units, not 7,200. That 12,000 is a judgement: it assumes the daily pace would have held to the shipping cutoff. The Black Friday sale stays in, because the brand will run it again.
| Last year | Shopify units | Shopify index | Amazon units | Amazon index |
|---|---|---|---|---|
| Normal month (April to June, spikes and Prime Day removed) | 4,000 | 1.0 | 2,400 | 1.0 |
| October | 4,800 | 1.2 | 2,400 | 1.0 |
| November | 10,000 | 2.5 | 4,800 | 2.0 |
| December (Shopify cleaned for the stockout) | 12,000 | 3.0 | 6,000 | 2.5 |
Step 2: apply this year's run rate
From April to June this year, with the same spikes removed, Shopify averaged 5,000 units a month and Amazon 3,000, each about 25 percent above last year.
| This year's forecast | October | November | December | Total |
|---|---|---|---|---|
| Shopify (5,000 × 1.2, 2.5, 3.0) | 6,000 | 12,500 | 15,000 | 33,500 |
| Amazon (3,000 × 1.0, 2.0, 2.5) | 3,000 | 6,000 | 7,500 | 16,500 |
Step 3: build up wholesale by account
Last year, 220 shops placed holiday orders. By July 1, 140 of them have ordered 7,000 units, about four cases each, to ship in October. The other 80 usually order around 50 units, and the team expects 60 to come back: 3,000 more. Expected wholesale is 10,000 units. Those are the shops' purchases; whether the tumbler then sells off their shelves, its sell-through rate, decides next season's orders.
Step 4: put a range around it
Last year's holiday plan was a single number built from raw sales. The store ran out on December 12, and the peak came in about 18 percent above plan. The team changed three things afterwards: it cleans stockout months before computing an index, it plans three cases instead of one, and it writes the forecast down before the season to compare after. The 20 percent band below comes from that one miss, and one season is a thin record; the band may prove too narrow.
| Holiday demand, units | Low (−20%, confirmed wholesale only) | Expected | High (+20%, every account) |
|---|---|---|---|
| Shopify | 26,800 | 33,500 | 40,200 |
| Amazon | 13,200 | 16,500 | 19,800 |
| Wholesale | 7,000 | 10,000 | 11,000 |
| Total | 47,000 | 60,000 | 71,000 |
Current stock should run out about when the containers land, so the expected-case order is 60,000 units across several containers. Landing in mid-September leaves time to send Amazon's October units on to Amazon's warehouses, which have to receive them before they can sell.
Step 5: cost the order and date the payments
The factory price is only part of what each tumbler costs. The landed cost adds freight, duty and the charges to reach the 3PL, and every one of them is paid before a unit sells.
| Date | Payment | Amount | Per unit | Running total |
|---|---|---|---|---|
| July 1 | 30% deposit, factory price $6.20 FOB | $111,600 | $111,600 | |
| August 12 | 70% balance, before the goods ship | $260,400 | $6.20 (both) | $372,000 |
| August 17 | Ocean freight | $48,000 | $0.80 | $420,000 |
| September 8 | Duty and tariffs, 20% of factory value (illustrative) | $74,400 | $1.24 | $494,400 |
| September 15 | Customs fees, broker, port, trucking, receiving | $27,600 | $0.46 | $522,000 |
| Landed cost | $522,000 | $8.70 |
Tariffs on many imported goods have moved several times since 2025, and the stacked rate can be far above this placeholder. Price your own HTS code with your broker before July, because duty is a cash payment at entry.
At $8.70 landed, the tumbler keeps a gross margin of 74.4 percent at $34 and 48.8 percent at $17 wholesale; a blended margin would overstate what each wholesale case earns. None of the $522,000 reaches the P&L as cost of goods sold until the tumblers sell, which is why September can look profitable while the bank account is the thinnest it will be all year. The walk from factory deposit to payout follows the same payments for a single production run.
Step 6: date the cash coming back
Each sale returns what is left after the costs of selling it: here, about $21 per Shopify sale after payment fees, shipping and ads, about $16 per Amazon sale after Amazon's fees and ads, and about $16 per wholesale unit after freight to the shop. Your own contribution margin by channel replaces these. Shopify cash is counted in the month of the sale, Amazon cash the month after (a cautious simplification), and wholesale, shipped in October on net 30, in November. The guide to order to cash for ecommerce traces each path from checkout to the bank.
| Cash back from the order | October | November | December | January | Total |
|---|---|---|---|---|---|
| Low | $100,800 | $360,400 | $328,800 | $96,000 | $886,000 |
| Expected | $126,000 | $470,500 | $411,000 | $120,000 | $1,127,500 |
| High | $151,200 | $548,600 | $346,200 | $80,000 | $1,126,000 |
The high case returns its cost fastest but sells out around mid-December, leaving about 11,000 units of demand unmet, roughly $210,000 of cash never collected at these per-unit figures.
What the low, expected and high cases do to cash
Before this order, the cash forecast shows $725,000 in the bank on October 1. Fixed costs (salaries, the 3PL's minimum, software, insurance) run about $120,000 a month, and the floor is $250,000, about two months of them. The bank line counts only this order and the fixed costs.
| Bank balance, end of month | October 1 | October | November | December | January |
|---|---|---|---|---|---|
| Low | $203,000 | $183,800 | $424,200 | $633,000 | $609,000 |
| Expected | $203,000 | $209,000 | $559,500 | $850,500 | $850,500 |
| High | $203,000 | $234,200 | $662,800 | $889,000 | $849,000 |
The same result in one table, laid out to copy into a spreadsheet:
| Three cases, one 60,000-unit order | Low | Expected | High |
|---|---|---|---|
| Holiday demand, units | 47,000 | 60,000 | 71,000 |
| Bank on October 1, after the order | $203,000 | $203,000 | $203,000 |
| Lowest month-end balance | $183,800 (October) | $203,000 (October 1) | $203,000 (October 1) |
| Month the order has paid back its $522,000 | December | November | November |
| Units left in January | 13,000 | 0 | 0 |
| Units of demand unmet | 0 | 0 | 11,000 |
The order digs the hole; demand decides how long you stay in it. Every case starts October about $47,000 under the floor on these assumptions. In the expected and high cases, October's sales cover the month's costs. In the low case they do not, and the balance slips to about $66,000 under the floor before November's sales pull it back.
The low case is slow, not a loss. It leaves 13,000 tumblers, about $113,000 at landed cost, on the shelf in January. At a normal pace of about 8,000 units a month across Shopify and Amazon, that stock lasts two to three months, since January and February run below a normal month. Weeks of supply measures how long it will sit; the cash those units return arrives later, as net receipts on each channel's payment dates. Stock is written down only when it will sell for less than its cost after the costs of selling it, so a slow tumbler at $34 puts no write-down on the P&L, though the storage it uses is still an expense each month. Clearing it below cost is the dead stock decision.
Buying for the high case costs more than it looks. Ordering 71,000 units would take about $96,000 more cash, push the October 1 balance to about $107,000, and in the low case leave 24,000 units, about $209,000 at landed cost, unsold in January.
So the July decision is about finding roughly $70,000, enough to stay above the floor if demand comes in low. Start with the factory: ask for terms on the balance, 30 or 60 days after the goods ship. At 60 days the $260,400 balance moves into mid-October, after the season's first sales. A factory you have bought from for years is more likely to agree, it may still say no, and a little more per unit for terms can cost less than borrowing. After that come a credit line arranged before the deposit goes out, a summer purchase held until November, or a smaller order and a December sellout. The cash conversion cycle shows how long this money stays tied up across a year, and a 13-week cash flow forecast is where these dated payments belong week by week.
How to forecast demand for a new product
A new product has no history, so the forecast borrows one. Take the closest product you have launched before and its first 90 days by channel, adjusted for what has changed since: store traffic, the email list, the Amazon listing's reviews. A waitlist or preorder page adds a number, converted at your own rate from past waitlists, not someone else's.
Then size the first order to the low case you can carry. Suppose the 20-ounce tumbler sold 9,000 units in its first 90 days and the team expects 12,000 for a new 32-ounce, with a low case of 6,000 and a high of 18,000, since some 32-ounce buyers would have bought the 20-ounce anyway. The factory's minimum at its quoted price is 15,000 units at an illustrative landed cost of $10.20, so $153,000 goes out before launch. In the low case, 9,000 units, $91,800 at landed cost, are still on the shelf after 90 days. In the high case it sells out in about 75 days and, with eleven weeks to the next delivery, stays out of stock for weeks.
Common ways ecommerce demand forecasts go wrong
Forecasting from sales during a stockout. The December sellout above turned about 12,000 units of demand into 7,200 of sales. Forecast from the sales report alone and next year's order is too small, and the stockout repeats.
Treating a spike as the base. A creator video, a press mention or a marketplace feature can sell a month's units in a week. Plan it as an upside case. A viral moment on TikTok Shop is a cash event as much as a sales one.
Forecasting dollars instead of units. A price change, a heavier discount or a shift toward wholesale moves revenue without moving units, and the order is placed in units.
Forgetting when the money goes out. Wholesale holiday orders ship weeks before the holiday, and stock bound for Amazon has to arrive there before it sells, so the payments land months before the season. The forecast has to carry dates, not only units.
Never checking the forecast against what happened. Write it down before the season and compare it with actual sales after, alongside the other inventory KPIs you review each month. The size of the last miss is the best guide to the next range.
Whatever the method, the order is decided by the low case, not the expected one: the $522,000 goes out either way, and only the low case shows whether you can wait for it to come back.
FAQ
How far ahead should an ecommerce brand forecast demand?
Forecast far enough ahead to see every payment the order needs and when the cash comes back. For an imported product with an eleven-week lead time, that often means forecasting three to six months ahead, decided several months before the season starts. Update the forecast monthly as actual sales come in.
How accurate does a demand forecast need to be?
Accurate enough that the low case still fits the cash you have. A forecast that misses by 20 percent is workable if that range was planned for and the slower case can be carried. The size of past misses, written down each season, sets a better range than any rule of thumb.



