What is demand sensing and how does it work in fashion?

KEY TAKEAWAYS Summary by the editors

1. Demand sensing is short-term forecasting that uses recent, high-frequency data such as point-of-sale activity, shipments, promotions, weather and social trends to detect near-term demand shifts.
2. Traditional forecasts are typically updated weekly or monthly from historical sales; demand sensing tools refresh predictions daily or even hourly.
3. In fashion, demand sensing mainly improves in-season decisions such as replenishment, stock transfers and markdown timing, because most quantities are committed before the season starts.
4. Wholesale brands can only sense demand in partner stores if retailers share sell-out data, for example through standard sales report messages carrying sales by GTIN, outlet and period.
5. External signals such as weather and search trends add value only when tested against a baseline; many signals turn out to be noise for a given category.

Demand sensing is a short-term forecasting method that uses the most recent data, such as yesterday's sales, current stock, promotions, weather and online signals, to detect shifts in demand over the next days or weeks. In fashion it complements the seasonal forecast used for buying: it does not change what was bought, but it improves where stock goes and when prices change during the season.

What is demand sensing?

DHL defines supply chain demand sensing as "a short-term forecasting method that uses real-time data and AI to detect near-term demand shifts". Where conventional approaches typically update forecasts weekly or monthly from historical sales, demand sensing tools refresh predictions daily or even hourly as new information arrives, drawing on signals such as point-of-sale activity, shipment patterns, promotions, weather and social trends.

The idea originated in consumer goods with stable, repeat-purchase products. Fashion adapts it to a harder setting: short product life cycles, many new items and demand split across sizes and locations.

In practice, demand sensing is less a single algorithm than a discipline: shorter forecast cycles, more granular data and a clear rule for how much weight recent evidence should get against the original plan. Machine learning helps because it can combine many weak signals and learn, category by category, which of them actually lead sales.

How is demand sensing different from demand forecasting?

The two are connected. A good seasonal model provides the baseline, and demand sensing adjusts it as evidence arrives. Zalando's published forecasting model, for example, produces weekly forecasts for a 26-week horizon to support pricing decisions. It sits between the two categories and shows how forecast horizons are chosen to match the decision they support.

Read also

AI demand forecasting in fashion: how it works and where it fails

Which signals does fashion demand sensing use?

• Sales and stock by size and location, ideally daily: the core signal for everything else.
• E-commerce behaviour: product views, add-to-basket, wish lists and search queries, which often move before sales.
• Prices and promotions, so the model can separate a discount effect from a genuine change in demand.
• Weather: temperature and rainfall shifts affect seasonal categories such as outerwear and swimwear.
• External trend data such as search interest; research on new fashion products found that adding Google Trends signals improved forecasts.
• Partner sell-out: for wholesale brands, sales and stock reported by retail customers.

Wholesale is where many brands are blind. Without partner data, a brand sees only its own orders, which arrive weeks after consumer demand changes. Standard sales report messages exist for this purpose: GS1 Austria's guideline for the fashion sector describes the EDIFACT SLSRPT message carrying sales quantities by GTIN, outlet and period between trading partners.

How does demand sensing improve in-season decisions?

1. Replenishment: stores and the web shop are topped up based on this week's sales pace rather than the original plan.
2. Transfers: stock moves from slow to fast locations while there is still time to sell at full price.
3. Markdowns: slow sellers are identified early, allowing smaller discounts sooner instead of deep discounts at season end.
4. Re-orders: for carry-over and quick-response products, early signals trigger re-orders within supplier lead times.

H&M described this goal in 2020, when its head of advanced analytics and AI told Retail Dive the team was working to quantify how much of each item would sell, in order to get the right product to the right place at the right time.

What does it take to implement demand sensing?

The technical requirement is reliable, frequent data: daily sales and stock by variant and location, prices, and the external feeds selected. The organisational requirement is a process that can act on a changed forecast. If allocation runs weekly and markdowns are decided monthly, daily forecasts add little. Many businesses therefore start by increasing the frequency of one decision, often replenishment, and only then invest in additional signals.

Every external signal should earn its place. Test it against a baseline forecast on historical data and keep it only if it improves accuracy for the categories where it is used. Weather may matter a great deal for coats and very little for basics; web traffic may lead store sales in one market and not in another.

Measurement should match the decision. For replenishment, compare forecast error over the replenishment lead time and track availability of core sizes. For markdowns, compare full-price sell-through and final clearance stock between products managed with and without the sensing model. A short pilot with a control group of stores or categories gives far more credible results than a before-and-after comparison, because fashion seasons differ so much from year to year.

Read also

Why is sell-out data essential for AI in fashion wholesale?

What are the limits of demand sensing in fashion?

There is also a cost side. Daily data pipelines, external data subscriptions and the staff time needed to act on more frequent recommendations all add up, so the business case should be built on specific decisions that will change, not on forecast accuracy alone.

Demand sensing cannot fix a wrong buy: if the season's stock is committed and supplier lead times are long, faster forecasts mainly redistribute what exists. It depends on data that many brands do not yet receive, particularly sell-out from wholesale partners. And it can amplify noise when volumes are small. Used with these limits in mind, it is a practical way to make in-season merchandising decisions faster and better grounded.

Frequently asked questions

What is demand sensing in simple terms?

It is short-term forecasting that updates frequently with the newest data, such as yesterday's sales, current stock, weather and web activity, to spot changes in demand over the next days or weeks.

Is demand sensing the same as demand forecasting?

No. Demand forecasting usually refers to longer-term forecasts for planning and buying. Demand sensing focuses on the short term and on adjusting decisions such as replenishment and markdowns during the season.

What data is needed for demand sensing in fashion?

Daily sales and stock by size and location are essential. Prices, promotions, e-commerce behaviour, weather and, for wholesale brands, partner sell-out data add further signals.

Does demand sensing work for new fashion products?

Partly. Once a new product has been on sale for a short time, early sales and web signals can update its forecast quickly. Before launch, demand sensing has little to work with and seasonal forecasting methods apply.

GuideThe complete guide to AI in fashion merchandising and buyingRead the complete guide