Demand forecasting for beauty brands: a practical guide

By Elie Dufeu, CTO & Co-Founder, Metreecs. Published 1 October 2026.

Demand forecasting for beauty brands means predicting unit sales for every product and variant (shade, size, set) in every channel, then planning stock against shelf life as well as against stockouts. Brands that do it well forecast their own DTC sales and their retail partners’ sell-through as two separate signals, because the two rarely move together.

Lena runs operations for a skincare brand that sells on its own site and through one specialty retailer. Her new vitamin C serum launched with a plan of 10,000 units over 90 days, split 6,000 for the site and 4,000 for the retailer. In week two a creator posted a routine video, and the site sold its 6,000 units in 11 days. The lab quoted 120 days for a repeat run, so the restock landed in month five, after attention had moved on and daily sales had dropped to a fraction of the peak. (Lena is an illustrative example with round numbers, but most beauty planners have lived some version of it.)

Launch spikes are familiar to every beauty planner, but a forecast that expects the spike, treats DTC and wholesale separately, and respects an expiry date takes more work to build. This guide covers how beauty demand behaves, how to forecast a new product, how shelf life changes safety stock, and which accuracy metrics are worth tracking. If you would rather test the method on real data, you can see a launch forecast built on your own sales history.

Key Takeaways

• McKinsey’s State of Beauty 2026 report finds that TikTok beauty sales have risen about 260% a year since 2023, so one video can move demand faster than a 90 to 120 day production lead time can respond.
• E-commerce is the largest beauty channel at 28% of sales (McKinsey, 2026), which means most brands with a DTC site also sell through retail partners whose data behaves differently.
• Shelf life runs from about 3 months for mascara to 36 months or more for fragrance, so a safety stock rule that works in one category creates write-offs in another.
• Researchers at Monash University found that a gradient boosting model that clustered similar products beat human expert forecasts for new beauty items in a retail case study.
• An eightx analysis puts e.l.f. Beauty at about 168 days of inventory on hand, a wait that would use up nearly half the life of a 12-month product.

What is demand forecasting for beauty brands?

Demand forecasting for beauty brands is the process of estimating how many units of each product and variant customers will buy in each channel over a set period, using sales history, launch plans, promotions, and market signals. The forecast then drives production orders, reorder points, and the split of stock between a brand’s own site and its retail partners.

The unit of the forecast matters. A foundation sold in 30 shades is 30 forecasts, and a hand cream in three sizes is three more. Category totals hide that spread, which is why forecasting at product, variant, and channel level is the working standard for demand planning software built for retail.

The market context makes this harder to avoid. McKinsey’s State of Beauty 2026 report expects the global beauty market to grow about 5% a year through 2030, with e-commerce already the largest channel at 28% of sales. Growth at that pace brings more channels and more launches to plan, and each one adds a demand signal to track.

Why beauty demand is harder to forecast than it looks

Four features set beauty apart from most retail categories, and a forecast has to handle each of them.

• Launches and social spikes. McKinsey reports that TikTok beauty sales have risen about 260% a year since 2023. A product’s demand can sit flat for months and then multiply within days, which breaks any method that extends last year’s average.
• Two channels with different signals. Your own site shows orders by product and variant every day. A retail partner often shows you purchase orders (sell-in) and, at best, weekly sell-through.
• A long tail of variants. Shade and size ranges create many products with low, irregular sales, where a few units either way moves the weekly number by a large percentage.
• Shelf life, minimum order quantities, and lead times. An eightx analysis of beauty inventory planning reports usable life of about 3 months for mascara, about 12 months for liquid foundation, and 36 months or more for fragrance. The same source cites contract manufacturer minimums of 1,000 to 25,000 or more units per product, and overseas formulation and fill runs of 90 to 120 days from purchase order to landed stock.

Across Metreecs’ work with beauty and cosmetics brands, the forecast is usually fine for the top ten products and weakest for the shade and size variants behind them. Those variants carry the expiry risk, because a small run still has to clear a minimum order quantity.

One mistake we repeatedly see is reading a retailer’s purchase order as consumer demand. A purchase order shows what the retailer chose to hold, and that choice shifts when the retailer is rebuilding stock, cutting cover, or preparing a promotion.

Demand forecasting for beauty brands: DTC and wholesale signals

A DTC order and a wholesale purchase order answer different questions, so they belong in different forecasts. The table shows what each channel gives you and what it should be used to predict.

For wholesale, forecast in two layers. First estimate what shoppers will buy at the retailer, using sell-through data where the buyer shares it. Then convert that into expected purchase orders using the retailer’s target weeks of cover and its ordering history. Our article on harmonizing stock levels across channels covers the allocation side of the same problem.

Amir plans for a hair care brand that sells on its own site and through a drugstore chain. The chain’s orders for a best-selling shampoo had averaged 5,000 units a month for a year, then fell to 2,500 for two months. Reading that as falling demand, he cut the next production run.

The chain’s weekly sell-through data, which he had not been looking at, showed flat consumer sales: the chain was working down stock it had built for a promotion. When it reordered 9,000 units in month three, Amir paid for a rushed production slot and expedited freight on 6,000 of them. (Amir is another illustrative example with round numbers.)

How to forecast demand for a new beauty product launch

A launch has no sales history, so the forecast starts from analogs: products that looked like this one when they launched. The steps below work in a spreadsheet and carry over to software.

1. Pick analog products. Choose past launches with a similar price tier, format, channel, and level of launch support, starting with your own range.
2. Group them by demand shape. Some products spike and decay, some build slowly, and some settle at a steady rate. Assign the new product to the shape its analogs share.
3. Build a range of scenarios. Write low, base, and high weekly unit curves for each channel rather than one number.
4. Split the buy. Commit the first production run to the low-to-base scenario, and ask your manufacturer about a smaller second run or reserved capacity for the high case.
5. Watch sell-through weekly. For the first six weeks, compare actual sell-through rate by variant and channel with the scenario lines.
6. Set the reorder trigger early. Place the repeat order when sales cross the high-scenario line, early enough that the lead time lands before stock runs out.

The grouping step has research behind it. Namazian, Stuckey, and Betts at Monash University tested a gradient boosting model on beauty retail data that clusters existing products by demand similarity, places each new item in a cluster using its attributes, and adjusts for seasonality. In their case study it outperformed human expert forecasts. For a deeper treatment, see our AI guide to new product introduction.

Analogs reduce the guesswork without removing it. A creator-driven spike can still land outside the high scenario, which is why the reorder trigger and the manufacturer conversation matter as much as the first number.

Plan safety stock and reorder points around shelf life

In most categories, extra safety stock costs only capital. In beauty it also costs life, because every day on the shelf comes out of the product’s usable window.

The reorder point still follows the usual logic: demand during the lead time plus safety stock. Take a product with a 12-month shelf life, daily demand of 100 units, and a 90 day lead time. Demand during the lead time is 9,000 units. A 20-day safety stock adds 2,000, so the reorder point is 11,000 units.

Now add the expiry check.

If stock sits for 168 days, the figure eightx reports for e.l.f. Beauty, the last unit out of a 12-month batch ships with about 6.5 months left. A product with a usable life of about 3 months cannot wait that long at all. Shelf life therefore sets a ceiling on days on hand (DIO) for each product, and safety stock has to be capped at the days of cover that ceiling allows.

Three habits keep that ceiling visible:

• Store shelf life and minimum remaining life for each product in the planning system, next to lead time and minimum order quantity.
• Ship first-expired, first-out. Since 2023, MoCRA batch and lot traceability has made this a compliance matter as well as a planning one, according to the eightx analysis.
• Review DIO by product and variant, not for the whole range, because the average hides the shades and sizes that are aging.

Priya plans for a color cosmetics brand with 60 lipstick shades and a 24-month shelf life. The eight slowest shades each sold under 40 units a week, but their second production run, sized to the manufacturer’s minimum, covered 14 months of demand. The following year she cut the cover on those shades to 4 months, accepted smaller and more frequent runs at a higher unit cost, and watched expired stock fall from 6,000 units to 1,500. (Priya is also an illustrative example.) The higher unit cost was real, and she judged it cheaper than writing the stock off.

Measure forecast accuracy for beauty brands where you decide

Forecast accuracy for beauty brands has to be measured at the level where orders are placed. A single portfolio number can look healthy while individual variants run months off.

Use two error metrics side by side. MAPE (mean absolute percentage error) treats every product equally, which makes it noisy for slow variants with a handful of weekly sales. WMAPE (weighted mean absolute percentage error) weights errors by volume, so it describes the products that carry the revenue. Add bias, the tendency to over- or under-forecast, because a forecast that is consistently high produces expiry and one that is consistently low produces stockouts.

Report accuracy by channel as well. A blended figure hides the case where DTC is forecast well and wholesale is not. For the financial side, see what inaccurate sales forecasts cost a retailer, which applies directly to beauty brands selling through partners.

FAQ

How do you forecast demand for a new beauty product?

Start from analog products with a similar price tier, format, and launch support, group them by demand shape, and build low, base, and high weekly scenarios for each channel. Commit the first run to the low-to-base case, and track sell-through weekly against the scenario lines so you can trigger a repeat order in time.

Should DTC and wholesale be forecast separately?

Yes. Consumer demand and retailer ordering follow different patterns, so each needs its own forecast. Estimate shopper demand in each channel first, then convert the wholesale portion into purchase orders using the retailer’s weeks of cover and ordering history.

How does shelf life change inventory planning for beauty?

It sets a ceiling on days on hand for each product. Safety stock and production run sizes have to fit inside the usable window, which ranges from about 3 months for mascara to 36 months or more for fragrance, and stock should ship first-expired, first-out.

How accurate can a beauty demand forecast be?

It depends on the product. Established products with steady sales can be forecast well, while launches and creator-driven spikes carry wide error. Measure with WMAPE and bias at product and channel level, and judge the forecast by whether it improves your decisions, not by a single percentage.

Can a small beauty brand forecast demand without a data team?

Yes, to a point. A spreadsheet with weekly sales by product and variant, an analog table for launches, and shelf life per product covers the basics. Once the range grows to hundreds of variants across several channels, AI-driven inventory optimization takes over the recalculation so planners can focus on exceptions.

Conclusion

Demand forecasting for beauty brands starts with treating DTC and wholesale as separate signals, then adds analog-based launch ranges and a shelf life cap on days on hand. Each piece can be tested on a single product line before anything else changes. To see them applied to your own range, book a Metreecs demo and model your launches and variants.

Sources

• McKinsey & Company, State of Beauty 2026: TikTok beauty sales growth of about 260% a year since 2023, e-commerce at 28% of beauty sales, and about 5% annual market growth through 2030.
• Namazian, Stuckey, and Betts, Monash University, New Product Demand Forecasting for Beauty Retail using Seasonality (CIE51, 2024): clustering and gradient boosting for beauty launches.
• eightx, Beauty Brand Inventory Planning: The 168-Day Problem: days on hand for public beauty brands, shelf life ranges, minimum order quantities, lead times, and MoCRA traceability.