A first demand forecast needs four numbers you already have and about two hours: weekly units by SKU, the promotion calendar that produced them, supplier lead time, and the days each SKU sat at zero. Run those four through three adjustments and you have a number good enough to place a purchase order against.
It won’t be precise. Precision is rarely what the decision in front of you needs.
Brands new to DTC skip this because forecasting sounds like a software purchase. How much machinery the job deserves depends on how expensive it is to be wrong, which for a lot of brands means a spreadsheet and one hour a month.
Ecommerce Demand Forecasting Is Only Worth Doing If It Changes a Decision
A demand forecast is an estimate of how many units you will sell in a future period, built to support a specific buying or spending decision. It earns its hours when a decision moves because of it. If the number arrives and you would have ordered the same quantity anyway, you built a report.
That test kills a lot of work before it starts, which is the point. It also sorts your catalog: the SKUs where the answer changes get the effort and everything else runs on a reorder point you check quarterly.
The Three Decisions a DTC Forecast Actually Feeds: What to Reorder, What to Spend, What to Promise
Reorder quantity is the obvious one and it’s where the money sits. The forecast tells you how many units to buy between now and the delivery after next.
Spend is the second. Your paid budget is a demand input. The forecast and the media plan have to run off the same assumption or one of them is already wrong.
The third is what you can promise: delivery dates to a retail buyer, fill rates to a wholesale partner, an inventory commitment to a marketplace. All three run off the same operating plan. A forecast nobody trusts makes the next ecommerce planning cycle harder.
You Already Own the Four Numbers a First Forecast Needs
Every input a first forecast needs is already in your store admin, your inbox and your promotion calendar. For a first pass you don’t need to buy tooling or scope an integration project.
If you’ve put this off because the data didn’t look clean enough, it’s probably clean enough for a first pass. The gaps you find are worth knowing about anyway.
Weekly Units by SKU, for as Long as You Have Sold the Thing
Weekly beats monthly because promotions and paydays move inside a month and a monthly total hides both. Export units rather than revenue so price changes don’t contaminate the signal. Twelve months shows you a shape. Six months is enough to start if the seasonal adjustment stays a guess you’ll correct next year.
The Promotion and Ad Calendar That Produced Those Units
A sales history records what you sold and what you did to sell it. Without the second half you’ll read a promotion as underlying demand and buy against a spike that only happens when you discount. Write down the dates, the offer and roughly what you spent.
Supplier Lead Time and Reorder Minimums
Lead time sets the horizon. The forecast has to run at least as far out as the gap between placing an order and selling it, which makes a 60-day forecast useless on a 90-day lead time. Measure that gap to the day the units are sellable, which includes the time your 3PL needs to receive them and put them away.
Minimum order quantities matter as much as the forecast. When a supplier’s minimum is six months of cover, the question stops being how many and becomes whether this SKU deserves six months of your cash.
Stockout Days, Because a Zero in the History Is Missing Demand and Not Low Demand
Every day a SKU sat at zero is a day your history understates demand. Mark those weeks and estimate what they would have sold at the rate of the weeks around them.
Returns pull the other way. The National Retail Federation put the 2025 online return rate at 19.3% of sales, worth $849.9 billion across retail. If one in five online units comes back, a forecast built on gross orders buys a fifth more inventory than the year consumed.
The Base Forecast Is Last Year’s Units Adjusted Three Times
Take last year’s weekly units for the SKU, adjust for growth, adjust for your own seasonal shape, then adjust for promotions you can name. That’s the whole method that most demand forecasting examples dress up with vocabulary.
Do it on one SKU first, ideally one whose sales you already know by feel.
Adjust for Growth Using the Rate the Business Is Running, Not the Rate You Hope For
Use the trailing rate the SKU is actually running, measured over the last three months against the same three months last year. The number in your growth plan is a target. Putting that target into the forecast is how brands finish a season holding inventory they planned to sell.
Category growth is a useful sanity check. The U.S. Census Bureau put second-quarter 2026 ecommerce sales at $340.2 billion, up 12.2% year over year while total retail rose 6.7%. A SKU growing at 40% against a category growing at 12% is either taking share for a reason you can name or showing you a promotion.
If your growth rate changes depending on which three months you pick, an outside read on the history settles it faster than another month of internal debate.
Adjust for Seasonality Using Your Own Weekly Shape
Divide each week’s units by the yearly average to get an index, then apply your own index instead of a category curve. Your peak is set by your customers and your own promotion habits, which rarely match the category average.
Two years of history gives you a shape you can trust. One year gives you a shape to hold loosely, because one unusual season otherwise becomes your whole seasonal assumption.
Adjust for Promotions Only Where You Can Name the Promotion on the Calendar
Lift belongs in the forecast when you can name the offer that caused it and you intend to run it again. A Black Friday week you ran last year and will run again is exactly that kind of lift. An unexplained spike stays noise until somebody explains it.
Where you can’t name the cause, leave the week at its underlying rate and write the spike down somewhere you’ll see it next year.
Run the three adjustments in that order and what comes out is your base forecast. It’s one number, it’s defensible, and you can walk every step of it past whoever signs the purchase order.
How Much Forecasting Sophistication Your Decision Justifies
The right amount of forecasting machinery is set by the cost of being wrong. Demand forecasting for ecommerce runs across a wide range of practice. Two brands of the same size can sit at opposite ends of it depending on lead times, assortment and cash position.
Locate Yourself: Write Down the Cost of Being Wrong by One Month of Cover
Take your top SKU, multiply one month of units by unit cost, and write that number down. That’s roughly what a month of overstock ties up in cash. Run the same arithmetic against margin for what a month of stockout costs in gross profit.
When both numbers are small enough to absorb without a conversation, a spreadsheet is the right tool. When either would change how you budget the year around it, the decision has earned more structure.
The Two Edges and Where Most Brands Your Size Actually Sit
At one edge is a monthly spreadsheet covering the top twenty SKUs, rebuilt by whoever places the orders. At the other is a demand planning system with a named owner, weekly statistical forecasts and an exception queue.
Those weekly forecasts run on time series analysis. The common method is exponential smoothing, which weights recent weeks more heavily than old ones.
Most DTC brands between $5M and $500M sit closer to the spreadsheet edge than they expect. Many are right to be there. When we look at how brands this size actually place orders, the spreadsheet is usually carrying more of the job than anyone in the room admits.
Ecommerce demand planning is a habit before it becomes a system. Buying the system first is how brands waste money here.
What Moves You Along It: More SKUs, Longer Lead Times, a Bigger Buy
Three things move a brand toward the structured end:
- SKU count: forecasting twenty variants by hand is about an hour. Forecasting four hundred is a job. The threshold sits wherever your buyer stops finishing the A list before orders are due.
- Lead time: a 30-day lead time forgives a bad forecast because you can correct it next month. A 120-day lead time turns one wrong number into two seasons.
- Buy size against cash: when a single purchase order is a meaningful share of your working capital, the forecast has become a financing decision.
When one of those crosses a point you can feel in the business, that’s your signal. A revenue milestone on its own says nothing about where you belong.
The demand forecasting companies selling at the structured end are real and their tools work. They’re also priced for a business whose forecast error costs six figures a season. That price is the honest test for whether it’s time to look at them.
Forecast the SKUs That Can Hurt You and Leave the Tail Alone
A small number of SKUs carry most of the risk. Forecasting the rest at the same level of care is how a monthly hour turns into a monthly day.
Shopify’s ABC analysis grades every variant by revenue contribution, with A-grade products accounting for around 80% of revenue, B-grade 15% and C-grade 5%. That grading is the 80/20 rule in ecommerce applied to your own catalog. The A list is where forecasting effort pays and the tail can run on a quarterly reorder point.
Assortment size is the biggest single lever on how forecastable a business is. The clearest illustration is a company that never let its assortment grow.
In-N-Out was founded in 1948 by Harry and Esther Snyder in Baldwin Park, California, on the principle of doing one thing well. As CNBC reported in 2024, In-N-Out’s core menu has barely changed since. A fixed menu is what makes demand predictable, training simple and fresh-never-frozen supply workable.
Nobody is asking you to freeze your catalog. Every SKU you add makes the forecast harder, which makes the assortment decision and the forecasting decision one decision.
Safety Stock Is Where the Forecast Meets Your Cash
Safety stock is the buffer covering the difference between your forecast and reality across your lead time. When it’s too thin, an ordinary forecast miss turns into stockouts on your best sellers. It’s also the line where an optimistic forecast quietly becomes a cash problem.
Size it from how variable the SKU’s weekly demand has been and how reliable the supplier has been. A steady seller from a supplier who hits dates needs a fraction of the cover a lumpy seller from a slow supplier does.
The forecast gives you a number. Turning it into a stocking policy is the job of inventory management. Ecommerce inventory optimization is mostly the discipline of holding cover where it earns its cash and cutting it everywhere else.
Marketing Spend Is an Input to the Forecast, Not a Surprise to It
The forecasting failure we see most often in a DTC launch has nothing to do with the model. It’s a media plan the person building the forecast never saw.
If paid spend is going up 40% in Q4, the forecast has to carry that assumption at the SKU level where the spend will land. Otherwise you’ll hit the revenue plan and still be out of stock on the three products the ads featured.
Deciding how much to spend on marketing and deciding what to buy are usually two separate conversations. They need to be one, with both sets of numbers on the table.
What the Platform’s Built-In Forecasting Covers and Where It Stops
Your platform’s inventory reports are genuinely useful and most brands under-use them. They also stop short of a forecast in one specific way.
Shopify’s days-of-inventory-remaining report divides a variant’s current stock by its average daily sales. Both that report and the ABC grading read the last 28 days of sales to do it.
Twenty-eight days can’t see your season. In February the report tells you that you hold 90 days of cover, which is true about February and silent about November. Filling that silence is what third-party Shopify demand forecasting apps are for.
Most of those apps read your whole sales history and run some flavor of AI or machine learning over it. What you are buying is the seasonal shape the 28-day reports cannot see.
If you’re weighing a forecasting app against another season of spreadsheets, a working call on the buy decision settles it faster than a trial does.
You Will Know the Forecast Is Working Before It Is Accurate
The first sign a forecast is working is that the argument about what to buy gets shorter. Accuracy shows up later and it matters less than people expect.
A forecast wrong by 20% in a way you can explain beats one wrong by 8% for reasons nobody can name. The explainable one tells you what to change next month.
Track the Error in Plain Percentage Terms, on the SKUs That Matter
For each A-list SKU, record forecast units and actual units each month, then the difference as a percentage of actual. You don’t need a weighted error metric or a dashboard.
After three months you’ll see whether you run consistently high or consistently low, which is a bias one number corrects. Forecast error belongs on the short list of which numbers to track.
A Monthly Review Beats a Better Model
One hour a month, the same week each month, with the buyer and whoever owns paid spend in the room. Update actuals, re-run the three adjustments, and change the orders that need changing.
That cadence catches more money than a better model does. Most forecast damage comes from a number nobody revisited after the plan changed.
When Forecasting Stops Being a Spreadsheet Problem
The spreadsheet stops working when the number of decisions outgrows the number of hours. That usually shows up as a buyer who can’t finish the A list before orders are due. It’s a capacity problem that a forecasting app or a planning module in the ERP you already run can genuinely fix.
It stops working in a second way that no tool fixes. The forecast comes out fine and the business still can’t decide what to do with it: which SKUs to keep, which channels to serve, what the next twelve months should look like.
Those are strategy questions. Answering them is where an ecommerce strategy consultant earns more than a planning app does.
Start this week with one SKU and four numbers. The method scales up when your decisions do. The discipline you build running it by hand is the part you keep whatever tool you eventually buy.
