Question 1

Which inventory metrics should clothing brands track?

Accepted Answer

Track five, at SKU level, weekly: sell-through rate, days of supply, gross margin per unit, return rate, and aged inventory. Anything you cannot act on within a week is a reporting metric, not a management one — and the common failure is tracking twelve numbers and deciding from none of them. The five, with their formulas and their triggers: (1) Sell-through rate = units sold ÷ units received, over a defined window. This is the buy/no-buy signal; compare against your own plan and against the comparable product, not against an industry figure. (2) Days of supply = units on hand ÷ average daily unit sales. This is the reorder signal and the stock-age signal in one; it is the metric merchandisers actually plan against. (3) Gross margin per unit = retail price − landed unit cost − fulfilment and payment costs. Margin, not revenue, funds the next order; a high-volume, low-margin SKU can consume cash while looking like a success. (4) Return rate by SKU = returned units ÷ sold units. Returns are hidden inventory — stock that is paid for, in transit or in inspection, and unsellable until processed. (5) Aged inventory = units unsold beyond a threshold (commonly 90 or 180 days). This is your markdown trigger. Supporting measures worth computing monthly rather than weekly: inventory turnover = COGS ÷ average inventory at cost (use cost on both sides — mixing retail inventory with cost COGS overstates the ratio badly), and GMROI = gross margin ÷ average inventory investment, which pairs the two into one number. Read all of them against your own three-year history before comparing to anyone else; the benchmark ranges vary by model, so a luxury brand's low turnover and a fast-fashion brand's high one can both be correct. Evidence and benchmark ranges. ReadyRatios' compilation of US-listed companies puts apparel inventory turnover at 154 days in 2025 (122 days in 2023); analysts put typical annual turns at roughly 6–10+ for fast fashion , 4–6 for sportswear and branded chains , 2–4 for department stores , and 1–2 for luxury — so compare within your model, not across it. BoF–McKinsey State of Fashion 2025 confirms the strong correlation between inventory turnover and profitability that makes these worth tracking, and reports warehousing costs up ~10% in 2023 , which raises the cost of every extra day of supply. On returns: e-commerce apparel return rates average around 30% , so return-processing speed is part of stock accuracy — consistent with Gruen and Corsten's finding that 70–90% of stockouts come from internal process gaps .

Question 2

How can clothing brands improve inventory turnover?

Accepted Answer

Buy less breadth, buy more often, and clear aged stock on a schedule — but do not chase turnover as a target, because turnover bought with stockouts is worse than slow stock. Turnover improves as a by-product of better buying, not as a goal in itself. A brand that simply holds less will turn faster and sell less. What actually moves the ratio: (1) reduce SKU count — fewer styles and fewer colourways concentrate demand into inventory that turns; (2) shift from season-buying to cycle-buying — buy coverage for the reorder cycle, not for the season, and reorder on trigger; (3) re-cut the size curve every order — dead extreme sizes are the slowest-turning inventory most brands own, and they are usually bought out of habit; (4) set a markdown calendar — decide in advance that stock crossing 90 days moves to a defined action, rather than deciding case by case months later; (5) fix the products that cause returns — a high-return SKU is churning inventory without turning it; (6) shorten the replenishment cycle — a shorter cycle requires less coverage, which raises turnover for the same service level. Two cautions: diagnose before acting — turnover falling while margin holds is usually aged stock not yet marked down; turnover falling with margin falling is a demand or price problem; and do not compare across business models — a luxury brand holding stock for scarcity is not underperforming. The manufacturing lever here is the same one throughout: a low reorder MOQ plus a short, reliable repeat cycle lets you hold less stock at the same availability, which is the only way to raise turnover without raising risk. Evidence. BoF–McKinsey State of Fashion 2025 frames inventory management as a primary profit lever precisely because of the strong correlation between inventory turnover and profitability , and reports concrete results from doing it systematically: Hugo Boss reduced inventory-to-sales ratio by 3.4% year on year by mid-2024 while investing €150m+ in digital intelligence, and the report estimates cross-functional inventory management has 10–15% of retail revenue in cost-reduction potential. On the corrective levers, the same report notes 75% of fashion executives increasing investment in advanced analytics/AI for forecasting, with AI-supported forecasting estimated to reduce inventory 5–15% and improve stock-outs 15–25% , and Kering reporting a 20% improvement in forecast accuracy after AI implementation. Counterweight: only about 60% of fashion inventory sells at full price , so raising turnover by discounting is self-defeating — the gain must come from buying, not from markdown.

Question 3

How can brands tell whether a product is worth restocking?

Accepted Answer

Restock when the product clears four tests — it sells, it makes money, customers keep it, and it still fits the range — and not merely because it sold out. Selling out is the weakest of the four signals: it can mean demand exceeded supply, or it can mean you under-bought by accident and the demand was thin. The four tests: (1) Velocity and sell-through — units per week against plan, and sell-through inside the intended window; a product that sells 200 units in 10 weeks and one that sells 200 in 30 weeks are different decisions. (2) Margin per unit after returns and fulfilment — a bestseller with a 35% return rate and heavy discount dependence may be worth less than a slower seller with clean economics; compute contribution per unit, not revenue. (3) Retention signal — repeat purchase in the same category, low return rate, and review sentiment; a product customers keep and rebuy is a programme, a product they try once is a moment. (4) Range fit — does it still represent the brand, is it seasonal or carryover, and does it cannibalise something better? Then decide the form: straight repeat for a proven carryover; corrected repeat — same style, re-cut size and colour split (the usual answer, see Q24); revision — same style, updated fabric weight or fit, if returns or reviews point at a product problem; or drop it and redeploy the cash. A useful discipline: score every SKU on these four on a fixed quarterly date and force a decision, because the default in most brands is to keep restocking whatever sold last time until cash runs out. Evidence. The margin-versus-volume distinction is the point of the GMROI framework: gross margin return on inventory investment, which pairs turnover with margin precisely so that high-volume/low-margin products are not mistaken for successes. BoF–McKinsey State of Fashion 2025 supports the retention and fit test indirectly — it reports out-of-stock sizes as the top shopper complaint and up to 20% monthly profit loss from inaccurate size purchasing , meaning a restock must also be a size-corrected restock; and it cites Lululemon attributing its Q1 2024 US slowdown partly to stockouts in smaller women's sizes, i.e. demand that existed but was not served. On returns: at ~ 30% average e-commerce apparel return rates, a high-return SKU's true contribution is materially below its gross margin, which is why test (2) must be net of returns. Quality and consistency controls for repeats: ISO 2859-1 / ANSI ASQ Z1.4 inspection, AATCC 135 / ISO 6330 dimensional change.

Question 4

How can AI help clothing brands forecast inventory demand?

Accepted Answer

AI is good at pattern-finding across many SKUs and bad at predicting what has no precedent — use it to compress your forecasting cycle and to surface imbalance early, not to replace the judgement call on new product. The realistic value is narrower than the marketing: AI forecasting works where you have history and many similar items, and it degrades where the product is genuinely new or the demand is driven by an unpredictable external event. Where it reliably helps: (1) SKU-level demand curves — fitting size and colour distributions across a range, which is exactly the error that costs brands the most; (2) early-warning on imbalance — flagging SKUs whose velocity is diverging from forecast before the divergence is visible in a weekly report; (3) replenishment quantity suggestions — combining velocity, lead time variability and service-level targets into an order recommendation; (4) seasonality and regional patterns across a large catalogue; (5) scenario comparison — what changes if lead time moves from 35 to 55 days, or if the price moves 10%. Where it does not: brand-new designs with no analogue, demand driven by a creator moment or a weather anomaly, and thin data — with a handful of SKUs and one season of history, a spreadsheet with honest assumptions will usually outperform a model fitted to noise. Practical guidance: treat outputs as a recommendation with a confidence band , require the model to show its error on the last cycle, and keep human sign-off on new-product quantities. Also separate two uses that get conflated — forecasting (how many) and planning (when to order given MOQ and lead time); a better forecast does not help if your reorder MOQ is too high to act on it. Evidence. BoF–McKinsey State of Fashion 2025 quantifies the opportunity: AI-powered forecasting has the potential to reduce inventory by 5–15% and improve stock-outs by 15–25% compared with traditional planning methods, and reports Kering achieved a 20% improvement in forecast accuracy after implementing AI; 75% of fashion executives were increasing investment in advanced analytics and AI for demand forecasting. The same report is explicit about the limits: micro-trend search volume fluctuates up to 300% within 12 months and #fashion content grew 2.5x in three years , so models fitted to historical patterns face genuine non-stationarity — which is why the report frames AI as decision support alongside regular inventory reviews and reliable supplier communication, not as a replacement for judgement.

Question 5

Can AI help fashion brands reduce overstock and stockouts?

Accepted Answer

Yes — measurably, but only in combination with shorter replenishment cycles and supplier terms that let you act on the output. AI reduces forecast error; it does not reduce lead time or MOQ. A brand that gets a 20% better forecast but still needs 300 pcs and 70 days to reorder will still overstock or stock out — the improvement is real but bounded by how fast and how small it can respond. The split matters: on overstock , AI helps most by catching divergence early — flagging a colourway that is tracking 40% below plan in week two instead of week ten, which is when a markdown is still avoidable and the next order can still be changed; on stockouts , it helps most by computing reorder points per SKU against real lead-time variability rather than an average, and by catching size-level imbalance before the core sizes go. Documented magnitudes: AI-supported forecasting is estimated to reduce inventory 5–15% and improve stock-outs 15–25% versus traditional planning, with Kering reporting a 20% accuracy improvement . But the same sources are clear that AI cannot remove inventory risk: trends, marketing shifts, supplier delays and weather still move demand, and BoF–McKinsey recommends combining AI recommendations with regular inventory reviews and supplier communication. Note also that a large share of stockouts has nothing to do with forecasting — Gruen and Corsten attribute 70–90% to internal replenishment and process gaps , which no model fixes. So the sequence is: fix the process (reorder triggers, SKU visibility, returns handling), then add forecasting, then use the improved forecast to negotiate better supplier terms — lower reorder MOQ, reserved fabric, shorter repeat cycle — which is where the compounding actually happens. Evidence. BoF–McKinsey State of Fashion 2025 : AI forecasting potential of inventory −5–15% and stock-outs −15–25% ; Kering +20% forecast accuracy post-AI; 75% of executives increasing investment. Limits stated in the same report: out-of-stock sizes remains the top shopper complaint , inaccurate size purchasing still costs up to 20% of monthly profit , and Lululemon's Q1 2024 US slowdown was partly attributed to stockouts — i.e. sophisticated planners still fail on execution. Gruen and Corsten: 70–90% of stockouts from internal process gaps . Complementary investments: Hugo Boss reduced inventory-to-sales ratio 3.4% with €150m+ digital investment; ASOS targeted 10% of own-brand on test-and-react; Shein runs 100–200 piece test batches — all examples of changing the response capability, not just the forecast.

Question 6

How can clothing brands make inventory decisions with limited sales data?

Accepted Answer

Replace forecasting with testing: buy the smallest quantity that produces a readable signal, define in advance what result means scale or stop, and let the manufacturing terms decide how small "smallest" can be. With no history, the goal is not to predict — it is to design an order structure where being wrong is cheap. The method: (1) set the learning quantity — enough units that the size and colour read is statistically meaningful (a few units per core size per colourway, typically landing at 50–150 pcs per style), not enough to matter if it fails; (2) define the decision rule before launch — e.g. "if sell-through exceeds 60% in 21 days, reorder at coverage; if below 30%, clear and do not repeat"; writing it first is what stops the post-hoc rationalisation that keeps dead stock alive; (3) borrow a prior from comparable product — a similar silhouette, weight and price band in the market, adjusted for the differences you can name; (4) collect a committed signal where possible — pre-orders, deposits, or wholesale sample commitments, which are worth far more than engagement; (5) prefer products that share materials and blocks — a capsule built on one fabric and one fit block can be re-cut into whichever style wins, which converts a forecasting problem into an allocation problem; (6) run one variable at a time — if you change silhouette, fabric and colour simultaneously, a failure teaches you nothing. The manufacturing prerequisite is a low MOQ per style/colour and a short repeat cycle, because that is what makes the learning quantity affordable and the scale-up fast. DOLUXE works from 50 pcs custom (20 on blanks) with 3–15 day sampling and 20–35 day bulk , which is the structure this method depends on — enquire. Evidence. Documented screening approach: a Chinese down-jacket brand listed 12 styles as pre-sale samples and only opened production on those exceeding 300 paid deposits — 5 proceeded, 7 were cut, saving an estimated ¥4.2m in fabric; the winner was replenished with 32,000 units in 15 days (operator-reported; not independently audited). Shein's on-demand model launches in 100–200 piece batches and scales on real-time response. Dancing Leopard , post-administration, adopted on-demand production explicitly to avoid high MOQs, long lead times and "seasonally driven guesswork". Why testing beats predicting here: BoF–McKinsey State of Fashion 2025 documents micro-trend search swings up to 300% within 12 months and #fashion content up 2.5x in three years — conditions under which a thin historical base is close to noise. Related: low-MOQ programme detail in Issue 04; sampling and development in Issue 09.

Metrics, AI and Deciding With Thin Data

Inventory Performance & Data-Driven Decisions

Q25Which inventory metrics should clothing brands track?

Q26How can clothing brands improve inventory turnover?

Q27How can brands tell whether a product is worth restocking?

Q28How can AI help clothing brands forecast inventory demand?

Q29Can AI help fashion brands reduce overstock and stockouts?

Q30How can clothing brands make inventory decisions with limited sales data?

Planning an order against these numbers?

Sources & Further Reading

From the DOLUXE Journal

Related DOLUXE Q&A issues