Demand Forecasting

What is demand forecasting?

Demand forecasting is the process of predicting future consumer demand at the style, category, channel, or location level using historical sales data, trend analysis, seasonality patterns, and external market signals. In apparel merchandising, demand forecasting is the quantitative foundation upon which buy plans, inventory budgets, and receipt flows are built — it translates merchandising strategy into unit-level projections that drive purchasing decisions months before products reach the selling floor.

Demand forecasting in apparel operates across two distinct time horizons: pre-season forecasting (made far enough ahead of the season to cover design, development and production lead times, used for buy planning) and in-season forecasting (weekly adjustments based on actual sell-through).

Why demand forecasting matters in apparel

Apparel's long lead times and seasonal selling windows make forecast error expensive to correct: the buy is committed before the season's demand can be observed, and the window to sell it is fixed. A forecast that runs high results in excess inventory, markdowns, and margin erosion. A forecast that runs low means stockouts, lost sales, and disappointed customers during peak selling weeks.

The financial stakes are significant:

• →Overbuy cost: Excess inventory above plan has to be cleared through markdowns, and markdown depth compounds as the clearance window shrinks — the later an overbuy is discovered, the more margin it costs to move
• →Underbuy cost: A stockout on a seasonal style is lost revenue rather than deferred revenue whenever a reorder's production lead time outlasts the selling window that remains — a core size that runs out early in the season then stays out for the rest of it
• →Working capital efficiency: Accurate forecasts reduce safety stock requirements and improve inventory turns, freeing capital for growth investments

The structural challenge in apparel is that every new style in a seasonal line arrives with no sales history of its own, forcing forecasters to rely on analogous style matching, trend indicators, and category-level demand patterns.

Demand forecasting in practice: apparel example

This is an illustrative example with a hypothetical brand and figures. A men's activewear brand is forecasting demand for its Spring running shorts program. The team builds the forecast using multiple inputs:

• →Historical baseline: Last Spring's running shorts category sold 42,000 units across 8 styles, and the category has grown about 8% a year for three years.
• →Trend adjustment: The team plans +10% for this Spring, above the trailing 8%, on its judgement that demand in the category is still accelerating: 42,000 × 1.10 = 46,200 units
• →Assortment change: Two new performance-fabric styles replace two older styles that sold 14,000 of last Spring's units, and analogous style analysis suggests the replacements will sell 15% faster: 14,000 × 1.10 × 0.15 = 2,310 additional units
• →Channel shift: DTC e-commerce is planned at 35% of units against 28% last year. In this example the shift changes how the buy is split between channels, not the unit total

The consolidated forecast is 46,200 + 2,310 = 48,510, rounded to 48,500 units. The team stress-tests it with an upside scenario of 53,000 and a downside of 44,000, sets the initial buy at 46,000 units, and holds open-to-buy for up to 7,000 more as a chase reserve for in-season reorders on winners, enough to reach the upside if early sell-through supports it. How much of a buy to hold back also depends on how much of it reaches doors through replenishment rather than the initial allocation.

Demand signals: stable vs. noisy

Modern planning stacks expose more demand inputs than any team can act on — POS sell-through, e-commerce viewed availability, back-in-stock email signups, return rates, promotional response, weather correlations, social spike data. More inputs do not produce better forecasts. The working rule is to build forecasts against the smallest set of stable signals and treat noisy signals as context, not algorithmic input.

Stable signals — drive the forecast:

• →Full-price sell-through by size and channel
• →Size-level stockout timing (when core sizes went to zero)
• →E-commerce viewed availability on OOS size selectors
• →Back-in-stock email signups, segmented by size
• →Weeks of supply, recomputed against trailing velocity

Noisy signals — useful context, dangerous as primary inputs: climate and weather data, promotional response curves, CRM return reason codes (unless large and categorical), and social/influencer spike data. See How Modern Apparel Brands Approach Sizing & Replenishment for the full signal hygiene logic.

Common mistakes

• →Relying solely on last year's sales without adjusting for assortment changes, competitive dynamics, or macro trends — last year's results reflect last year's assortment, not this year's
• →Forecasting at too high an aggregation level — a category forecast of +8% means nothing if growth is concentrated in two styles while six others decline
• →Ignoring the difference between demand and sales — historical sales reflect constrained demand; if a style stocked out in week 6, actual demand was higher than recorded sales
• →Treating forecast accuracy as a planning-only problem — forecast quality depends on clean POS data, accurate inventory positions, and disciplined promotional calendars
• →Over-weighting noisy signals — climate, promotional response curves, and social spikes introduce variance that can override the stable sell-through signal at the worst moments; keep noisy inputs as human-in-the-loop context, not algorithmic inputs