One of the hardest questions in inventory planning is also one of the simplest to ask:

How much of this product am I likely to sell next?

Order too little and you risk running out before your next shipment arrives. Order too much and cash becomes tied up in products that may sit on the shelf for months.

Demand forecasting helps make that decision less dependent on guesswork.

You don't need an advanced forecasting system to get started. For many ecommerce stores, your own historical sales data provides a useful foundation for estimating future demand.

In this guide, we'll look at how to turn past sales into a practical demand forecast, how to choose an appropriate time period, and when historical data needs to be adjusted before you use it for inventory decisions.

1. What is inventory demand forecasting?

Inventory demand forecasting is the process of estimating how much of a product customers are likely to buy during a future period.

For example, instead of simply knowing:

We sold 420 units of this product last month.

a forecast attempts to answer:

Based on what we know today, approximately how many units are we likely to need next month?

That estimate can then support decisions about purchasing, replenishment, safety stock, and inventory levels.

Historical sales are usually a good place to start because they show something concrete: what customers actually bought.

But forecasting isn't simply copying last month's sales into next month's plan.

A useful forecast considers whether past demand is representative of what is likely to happen next.

2. Start with clean historical sales data

Before choosing a forecasting method, make sure you're measuring the right thing.

For each product or SKU, you ideally want a history containing:

• Date

• Units sold

• Product or variant

• Location, when relevant

• Returns or cancellations

• Stock availability

• Promotions or discounts

The amount of history you need depends on the product.

For a relatively stable product, the previous 30, 60, or 90 days may provide enough information for a basic short-term forecast.

For a highly seasonal product, you may need a year or more of data to understand recurring patterns.

Watch out for stockouts

One of the easiest forecasting mistakes is treating zero sales as zero demand.

Suppose a product normally sells around 10 units per day but was out of stock for seven days.

Your data might show:

Week 1 — 70 units
Week 2 — 70 units
Week 3 — 0 units
Week 4 — 70 units

The monthly average would be:

210 ÷ 28 = 7.5 units per day

But customers didn't suddenly stop wanting the product during Week 3. You simply didn't have anything available to sell.

Using 7.5 units per day as your forecast could therefore cause you to underestimate future demand.

When possible, identify periods when products weren't available and avoid treating those days as normal demand.

Watch out for unusual sales

The opposite problem can happen too.

Imagine a product normally sells about 200 units per month:

January — 190
February — 205
March — 198
April — 410

Before assuming demand has doubled, ask what happened in April.

Perhaps you:

• Ran a major promotion

• Sent a large email campaign

• Were featured by an influencer

• Received an unusually large wholesale order

• Heavily discounted the product

Those 410 units are real sales, but they may not represent normal demand.

Good forecasting starts by understanding why the historical numbers look the way they do.

3. Establish a baseline with average sales

One of the simplest ways to forecast demand is to calculate average sales over a recent period.

Suppose a product sold:

360 units during the previous 30 days.

Its average daily sales would be:

Average Daily Sales = Units Sold ÷ Number of Selling Days

So:

360 ÷ 30 = 12 units per day

If you're planning inventory for the next 30 days, a very simple forecast would be:

12 × 30 = 360 units

This is a reasonable starting point for a product with relatively stable demand.

But choosing the right historical period matters.

How much history should you use?

There's no single correct answer.

Consider a product with these monthly sales:

If you use the entire six-month average, you'll get:

1,805 ÷ 6 = approximately 301 units/month

But the product sold 390 units in the most recent month.

Demand appears to be growing, so 301 units may now be too conservative.

A shorter window reacts more quickly to changing demand.

A longer window tends to be more stable but slower to recognize change.

This is why the sales velocity period—the historical window used to measure recent demand—is an important forecasting setting.

For a relatively stable product, you might use the previous 60 or 90 days. For a rapidly changing product, the previous 30 days may better represent current demand.

The key is consistency: choose a period appropriate to the product and review it when demand patterns change.

4. Look for trends, not just averages

An average tells you the general sales rate. A trend tells you whether that rate is changing.

Suppose a product's sales were:

April — 300 units
May — 330 units
June — 365 units
July — 400 units

A simple historical average gives:

349 units/month

But demand has been consistently increasing.

Forecasting another 349 units for August would ignore that trend.

One simple way to identify trend is to compare recent sales with an earlier equivalent period.

For example:

Previous 30 days: 400 units
Prior 30 days: 365 units

The percentage change is:

(400 − 365) ÷ 365 × 100 = 9.6%

You could describe the product as having an approximately 10% increasing recent trend.

That doesn't automatically mean next month's sales will increase another 10%. Trends can slow, accelerate, or reverse.

Instead, think of trend as an additional signal:

Historical average tells you what's typical.
Recent trend tells you where demand may be heading.

5. Use moving averages to smooth noisy demand

Daily sales can be messy.

Consider:

Monday — 8
Tuesday — 17
Wednesday — 6
Thursday — 14
Friday — 21
Saturday — 12
Sunday — 4

Trying to react to every daily change would produce an unstable forecast.

A moving average smooths these fluctuations.

For example, a 30-day moving average continuously calculates average demand from the most recent 30 days.

As each new day arrives, the oldest day drops out and the newest day is added.

This makes the forecast responsive to recent sales without allowing a single unusual day to dominate the calculation.

Weighted moving averages

Sometimes recent sales should matter more than older sales.

Suppose monthly demand was:

April — 280 units
May — 320 units
June — 380 units

Instead of giving each month equal importance, you might use:

April — 20%
May — 30%
June — 50%

The forecast becomes:

(280 × 20%) + (320 × 30%) + (380 × 50%)

56 + 96 + 190 = 342 units

This approach can be useful when demand is changing and you believe recent sales are more representative of what's coming next.

6. Account for seasonality

Historical averages become less useful when demand follows predictable seasonal patterns.

Imagine a store selling outdoor products.

Sales might look like:

January — 120 units
April — 260 units
July — 720 units
October — 290 units

A 12-month average might be mathematically correct but practically useless for forecasting July.

Instead, look for recurring patterns.

Ask:

• Did demand increase during the same period last year?

• Are certain products strongly associated with a season?

• Do Black Friday or holiday sales materially change demand?

• Does back-to-school affect particular products?

• Are there consistent differences between weekdays and weekends?

If you have multiple years of sales history, comparing the same period year over year can be particularly helpful.

For newer products without enough history, you may need to use similar products, category-level patterns, or conservative assumptions until more data becomes available.

7. Adjust for what you already know about the future

Historical sales tell you what happened.

Your forecast is trying to estimate what happens next.

That distinction matters.

Suppose historical data suggests you'll sell 500 units next month.

But you already know you're planning a major promotion.

Using 500 units without adjustment ignores information you already have.

Before finalizing a forecast, ask whether the upcoming period contains anything unusual:

Promotions

A discount, advertising campaign, or influencer partnership may increase demand.

Product launches

A related product launch might increase—or reduce—sales of an existing SKU.

Price changes

A significant price increase or decrease can change customer behavior.

Seasonal events

Christmas, Valentine's Day, Mother's Day, back-to-school, and other periods may materially affect particular products.

Supplier or product changes

A product being discontinued may justify deliberately reducing inventory even when historical demand remains strong.

This is where forecasting becomes more than mathematics.

The model provides a baseline. Business context improves the forecast.

8. Measure forecast accuracy

A forecast shouldn't disappear after you create it.

Once the forecast period is over, compare what you predicted with what actually happened.

Suppose you forecast:

500 units

Actual demand was:

450 units

Your forecast was 50 units too high.

The percentage error relative to actual demand is approximately:

50 ÷ 450 × 100 = 11.1%

Now imagine you repeatedly overforecast the same product by 10–20%.

That's useful information. Your forecasting assumptions may be systematically too optimistic.

Likewise, repeatedly underforecasting may indicate that your model isn't responding quickly enough to growth.

Don't judge your forecasting process based on whether one forecast was wrong. Forecasts will always contain error.

Instead, monitor accuracy over time and ask:

Is the forecast becoming more useful for making purchasing decisions?

That's a much better standard than expecting perfect predictions.

9. Turn the forecast into an inventory decision

A demand forecast by itself doesn't tell you what to purchase.

Suppose your forecast says:

Expected demand over the next 30 days: 600 units

But you already have:

450 units available
100 units incoming

You don't need to order another 600 units.

Likewise, if your supplier requires 21 days to replenish the product, waiting until you have only a few days of inventory remaining could be too late.

A purchasing decision needs to bring together several pieces of information:

Forecast demand
How much are customers likely to buy?

Current inventory
How much do you already have available?

Incoming inventory
What's already on order?

Supplier lead time
How long will replenishment take?

Safety stock
How much buffer do you want?

Target inventory coverage
How far into the future do you want the new order to support?

That's why forecasting should be viewed as the first part of an inventory-planning process, not the final answer.

A practical starting approach

If you're currently planning inventory manually, you don't need to implement every forecasting method at once.

Start with something you can understand and maintain.

For each important SKU:

1. Collect recent sales history and identify periods when the product was unavailable.

2. Calculate average daily sales over an appropriate period, such as the previous 30, 60, or 90 days.

3. Compare recent demand with the previous equivalent period to identify whether demand is increasing or decreasing.

4. Look for obvious seasonal patterns and unusual events.

5. Create a baseline forecast for your planning horizon.

6. Adjust it for promotions or other future events you already know about.

7. Compare the forecast with actual sales when the period ends.

Then refine the process as you collect more data.

A simple, understandable forecast that's reviewed regularly is often more useful than a sophisticated model nobody trusts.

Forecasting is about reducing uncertainty—not eliminating it

No inventory forecast will tell you exactly what customers will buy next month.

That's not the goal.

Forecasting gives you a structured way to use the information you already have instead of making every inventory decision from intuition alone.

Historical sales establish the baseline. Trends tell you how demand is changing. Seasonality provides context. Known future events let you adjust the estimate. And forecast accuracy helps you learn from what actually happens.

Put those pieces together and a question like:

“How much do I think we'll sell?”

becomes something you can answer with evidence rather than a guess.

And once you have that answer, you can tackle the next—and arguably more important—inventory question:

“Given that forecast, when should I reorder?”