Demand Forecasting for eCommerce: How to Stop Guessing Stock Levels

If you've ever stared at a stock report and thought “well, we'll just have to see how it goes”… you're not alone. For a lot of growing eCommerce brands, buying stock still comes down to gut feel: order roughly what sold last time, add a bit extra just in case, and hope for the best. The problem is that “just in case” gets expensive fast, whether that's cash tied up in stock that won't move for months, or a best-seller going out of stock right when demand peaks.

Demand forecasting isn't about predicting the future with total precision, nobody can do that. It's about replacing guesswork with a repeatable process, so your stock levels are based on evidence rather than instinct. Get it right, and you'll order more accurately, free up cash flow, and avoid the two costliest mistakes in eCommerce: overstocking and stockouts.

What is demand forecasting, really?

At its simplest, demand forecasting is the process of estimating how much of a product you'll sell in a future period, using data you already have. That might be historical sales, seasonal patterns, marketing plans, market trends, or a combination of all four.

It sits right alongside the inventory management and peak-planning work most eCommerce brands already do, the difference is direction. Inventory management asks “what do we have and where is it?” Demand forecasting asks “what will we need, and when?” Get the forecast right, and the inventory management side becomes far easier to manage.

Why “we'll just reorder what we sold last time” doesn't work

Reordering based on last period's sales feels safe, but it quietly builds in every mistake you made last time too. If you were out of stock for two weeks last month, your sales data for that period is artificially low, and if you reorder based on it, you'll under-order again. If a product had a one-off spike because of a viral TikTok moment, reordering the same amount could leave you sitting on stock nobody wants once the moment passes.

Simple reordering also has no way to account for what's coming next: a planned promotion, a new sales channel launching, a supplier's lead time getting longer, or the fact that Black Friday is eight weeks away. Forecasting builds those variables in, rather than assuming next period will look exactly like the last one.

The building blocks of a decent forecast

You don't need a data science team to forecast well, you need a handful of inputs used consistently.

Historical sales data.

Your single best predictor, but only if you clean it first. Strip out stockout periods, one-off bulk orders, and anomalies so you're forecasting from genuine demand, not from what you happened to have available.

Seasonality.

Most eCommerce categories have a rhythm, Black Friday and Christmas are the obvious ones, but plenty of brands also see smaller peaks around payday weekends, back-to-school, or category-specific moments (think beauty around Valentine's Day, or fitness around January). Map at least 12–18 months of sales to spot your own pattern rather than assuming it matches the wider market.

Lead times.

A forecast is only useful if you know how long it takes to act on it. If your supplier lead time is eight weeks, you need to be forecasting eight-plus weeks ahead, with enough safety stock to cover the gap if a shipment runs late.

Growth and marketing plans.

If you're launching on a new marketplace, running a big campaign, or expecting a spike from an influencer partnership, your forecast needs to flex for that, historical data alone won't capture demand that hasn't happened yet.

External factors.

Currency shifts, competitor stock-outs, viral trends, even weather can move demand for certain categories.You won't be able to forecast all of it, but building in a review point lets you course-correct quickly when something unexpected shows up.

Simple forecasting methods worth knowing

You don't need to jump straight to complex statistical models. Most growing brands get real value from methods that take an afternoon in a spreadsheet:

Moving average.

Takes your average sales over a set number of past periods (say, the last 3 months) to smooth out noise and spot the underlying trend. Simple, and a good starting point if you're forecasting for the first time.

Weighted moving average.

Same idea, but gives more weight to recent periods, useful if your sales trend is genuinely shifting rather than staying flat.

Seasonal index method.

Compares each month's typical sales against your yearly average to build a seasonality multiplier, so your forecast automatically scales up for December and down for quieter months.

Exponential smoothing.

A step up in sophistication, this weights recent data more heavily still and adjusts faster to changes in trend, better suited to fast-moving or rapidly growing product lines.

As your product range and order volumes grow, most brands eventually move from spreadsheets into dedicated forecasting or inventory planning software that can handle this automatically across hundreds of SKUs, but the underlying logic is the same either way.

Turning a forecast into a stock decision

A forecast on its own doesn't order stock, it needs to feed into two practical numbers:

Safety stock: the buffer you hold to cover forecast errors, demand spikes, or supplier delays without running out.

Reorder point: the stock level at which you place a new order, calculated from your average demand during lead time, plus your safety stock.

Get these two numbers right foreach SKU (rather than applying one blanket rule across your whole catalogue) and you start making noticeably fewer emergency reorders and noticeably fewer markdowns on stock that isn't moving.

Where forecasting and fulfilment meet

Forecasting accuracy matters more the closer you get to peak periods, because the cost of getting it wrong multiplies, a stockout in a quiet month is a missed sale, but a stockout during Black Friday week can mean missing your biggest revenue moment of the year. It's the same principle behind the peak-readiness planning most brands already do; forecasting is simply what happens further upstream, informing what you buy and when, rather than how you handle the rush once it's already at your door.

It's also where working with a fulfilment partner pays off beyond just pick, pack and dispatch. A 3PL sitting across your order and inventory data has visibility a lot of brands don't get from sales reports alone, real-time stock levels, dispatch trends, and early warning when a SKU is moving faster or slower than expected. That operational data is exactly what feeds a better forecast next time round.

The takeaway

You'll never forecast demand with total accuracy, and that's fine, the goal isn't perfection, it's consistency.

A basic, repeatable forecasting process beats an accurate one-off guess every time, because it improves with every sales cycle you run it through. Start with clean historical data, layer in seasonality and lead times, and build in a regular review point.

Stock decisions stop being a guessing game, and start being something you can actually plan around.

‍