What Is Demand Planning and How Do You Forecast Demand Accurately?

How demand planning works, which forecasting models fit which SKUs, and how to keep forecasts accurate through new launches, seasons, promotions and growth.

Last Updated:
November 9, 2026
Table of contents:

Restock on time, every time.

Get A Demo

Demand planning is deciding how much stock to buy or make, and when, based on a forecast of how many units each SKU (each product variant you sell) will sell. The forecast is the estimate. The plan is what you order because of it.

What Are Demand Planning Models?

Demand planning models are the methods used to estimate how much each SKU will sell. Some work from your sales history including statistical models and machine learning.

Others rely on judgement when there's little or no history, such as copying a similar product's launch or asking the people closest to the customer. Which one fits depends on how the product sells:

How the product sells Method that fits Where it goes wrong
Steady, similar sales most weeks Moving average or simple exponential smoothing Sales jump to a new level and the forecast lags behind
Growing or shrinking over time Holt's method (tracks level and trend) It misses seasonal peaks
Same peaks every year Holt-Winters (adds the season on top of the trend) You have less than two years of sales, so the peak hasn't repeated yet
Slow movers: sells some weeks, zero other weeks Croston's method, SBA or TSB (built for sporadic sales) Weeks when it was sold out look like weeks nobody wanted it
Brand new, no sales yet Judgement: copy a similar product's launch, ask your team, plan a few scenarios Launch optimism, and customers saying they'll buy more than they do

You can run Holt-Winters in a spreadsheet. Our forecasting in Excel models page has templates.

Once you have hundreds of SKUs selling in different ways, AI inventory forecasting software picks the method for each SKU.

How Do You Forecast a Product With No Sales History?

There are two common ways to forecast a product that hasn't sold yet:

  1. Take a similar product's launch sales and scale them up or down to what you expect from the new one.
  2. If it's a new colour or size of an existing product, give it a share of the parent product's sales, then adjust as its own sales come in.

Keep the launch forecast until a few weeks of sales arrive because first days distort the average.

How Accurate Can a Demand Forecast Be?

There's no single accuracy number to aim for. How close a forecast gets depends on how the product sells, how much history it has, and whether the future looks like the past.

In an anonymised analysis of 95,000+ in stock SKUs across 130+ Shopify brands (April to September 2026), the typical monthly forecast was off by (WAPE) 29% for steady sellers, 48% for products with up-and-down sales, 82% for new products and 94% for slow movers.

Set a target for each type of product, not one number for the whole store.

Before you judge accuracy, set aside the weeks a SKU was sold out (they hide real demand) and any one-off orders.

You don't need to check every SKU. Once a month, compare what you planned with what sold and look at the biggest misses by revenue; a few SKUs usually cause most of the error.

Which Demand Planning KPIs Matter?

Demand planning KPIs tell you two things:

  • How far off your forecasts were
  • Whether customers could buy what they wanted.
KPI What it tells you Formula
MAPE Average miss as a percentage. Breaks when a SKU sold zero Average of abs(Actual − Forecast) / Actual × 100
WAPE / WMAPE Same miss, but best sellers count more than slow movers Sum of abs(Actual − Forecast) / sum of Actual × 100
Bias Whether you keep forecasting too high or too low. Positive means too high Average of (Forecast − Actual)
MASE Whether your forecast beats simply repeating last period's sales. Below 1 means it does Forecast's average error / average error of "same as last period"
FVA (forecast value added) Whether a manual change or new tool made the forecast better. Positive means it helped Error before the change − error after
Stockout rate Share of your catalogue that's sold out SKUs out of stock / total SKUs × 100
On-time fulfilment rate Share of customer orders shipped on time Orders fulfilled on time / total orders × 100

See forecast accuracy formula.

How Granular Should a Demand Forecast Be?

Forecast at the level you buy at. If you order by SKU, forecast by SKU. If you also plan by product type or channel, make sure those totals add up to the same number as your SKU forecasts.

If the same SKU sells on your store, on Amazon and to wholesale buyers, add those together into one forecast before you order.

Fast sellers should be forecast weekly, slower ones should be done monthly.

Forecast bundles through their components, since each bundle sale uses up every item inside it.

How Do You Plan for Seasons and Promotions?

Seasons and promotions need different handling. A season repeats every year, so your sales history shows it. A promotion doesn't, so you have to tell the forecast about it: dates, the discount, and which other products it will pull sales from or add sales to.

Promotions are usually planned as a percentage lift ("+40% that week") rather than a unit count.

Promotions also make sales harder to predict, so you need more safety stock around them. Mark promotion weeks as one-off events so a spike doesn't become the new normal, and in your seasonal forecasting, place orders before supplier shutdowns such as Lunar New Year.

How Do You Adjust a Forecast for Growth and One Off Orders?

You have to tell the forecast how much you plan to grow, because past sales can't show growth that hasn't happened yet. Copying last year forward assumes zero growth.

Take one-off orders (a wholesale deal, an influencer post, a big promotion) out of your sales history and add them back as planned events, so the forecast doesn't expect them to happen again every year.

If last year's peak was capped by stockouts or missing sales channels, a growth target applied to it still lands below real demand.

Whatever ups and downs are left after this need safety stock.

See demand variability.

How Do Forecasts Become Purchase Orders?

A forecast becomes a purchase order once you subtract the stock you have and the stock already on order, then work out how much to buy and when to send it.

If you order on a fixed schedule, each order has to cover sales until the next order arrives, not just until this one does. Then round up to the supplier's minimum order quantity and case size.

Use the lead times your suppliers actually deliver in, not the ones they quote. Across 28,000+ received purchase orders from 400+ brands (January to September 2026), 36.5% arrived later than the expected date on the PO, and the median late PO was 6 days behind.

See automate purchase orders.

Can You Forecast Without a Data Science Team?

Yes. Most brands don't need a data scientist to forecast well. What matters is matching the method to how much history you have, and having someone who owns the plan and acts on it.

With only a few months of sales, use simpler methods (a short history favors simpler models). A few weeks of data can’t show a seasonal pattern.

See AI agents for demand forecasting.

How Do Teams Share One Demand Plan?

Teams share one demand plan by agreeing on one set of numbers that sales, marketing, finance and operations all plan from. The formal name is sales and operations planning (S&OP), which brings different business plans into one integrated set of plans.

The common failure is too many manual changes. Teams adjust much more often than they should, and many times for the wrong reasons.

See demand planning vs supply planning.

Frequently Asked Questions on Demand Forecasting & Planning

How Does a Dedicated Demand Planner Help Versus Just Using Software to Forecast Inventory?

A dedicated demand planner adds judgment that software can't. Reading a competitor's stockout, weighing a supplier's reliability, or overriding a forecast because of something happening on the ground the data hasn't caught up to yet.

Software can handle the repetitive calculation at scale. A planner decides when to trust it and when to step in.

Can I Integrate Demand Forecasting Into My Annual Operations Plan?

Yes, demand forecasting feeds directly into annual operations planning. Use it to set your baseline sales targets, then revisit the plan monthly or quarterly as new forecasts come in.

What Questions Should You Ask Vendors When Evaluating AI Reordering Features?

Ask how the model will handle a product with only one or two prior seasons of data, and whether reorder points update by themselves as the season moves on, or you have to reset them by hand each time.

How Do You Automatically Detect Purchase Orders That Don't Match Your Forecast?

Plan-vs-actual tracking in demand planning software is built for exactly this. It compares what you ordered against what the forecast said you’d need and flags the gap directly, instead of you noticing months later that a PO was sized off the wrong number.

If you need to plan purchase orders across locations rather than forecast demand, compare inventory planning software.

How Do You Forecast Ongoing Demand After an Influencer Spike?

Daily sales data shows the surge as it happens (the same way you'd spot a product going viral on TikTok before it sells out), and the days immediately after tell you whether demand is settling at a new, higher baseline or reverting to the old one, which is what should actually drive the next reorder quantity.

Authors
Youri Moskovic
CEO & Founder
Cyrus Mahler
COO
Nicolas Sabatier
Co-Founder & CTO