Short answer: you forecast demand for a new collection by triangulating — borrowing evidence from comparable products, measuring real buying signals before production, starting with a controlled first order, and letting actual sales refine every order after that. No forecast of a brand-new product is ever "accurate." The goal is to be approximately right and cheaply wrong.
This is the question behind almost every difficult first order we've seen since 2007: how many do we make when we've never sold this before? The brands that get hurt are the ones who treat the question as a prediction problem. The brands that do well treat it as a risk-management problem.
1. Why demand forecasting matters more in fashion than almost anywhere
Most industries can produce to order or hold stable inventory for years. Fashion can do neither comfortably. Styles are perishable — a hoodie that's hot this season is dead stock next year — and production lead times force you to commit quantities months before demand is visible.
The cost of getting it wrong runs in both directions. Overestimate and you join the industry's excess inventory problem: the BoF–McKinsey State of Fashion report put 2023's overproduction at 2.5–5 billion garments, worth $70–140 billion. Underestimate and you stock out of the exact product customers wanted — Lululemon publicly attributed slowed US growth in 2024 partly to stockouts in small sizes, a pure forecasting failure on an otherwise strong product.
Every first order sits somewhere between those two cliffs. Forecasting is how you pick your position deliberately instead of by accident.
2. Start with historical sales — even if they're someone else's
If you've sold anything before, that's your anchor. Look at your last collection not as totals but as structure: which silhouettes moved, which fabric weights, which colors, at what weekly velocity, and how sales decayed after launch week. A new collection that shares DNA with a proven seller — same category, same price band, same customer — can reasonably be forecast as a fraction of that seller's early velocity. Fifty to eighty percent of a proven comparable is a defensible starting assumption for a genuinely new design; more than that requires evidence, not confidence.
If you've sold nothing yet, borrow comparables deliberately: pick two or three brands at your target scale (not Supreme — brands two years ahead of you), study what they restock, and treat their visible bestsellers as demand evidence for the category, not for your specific design. Restocks are the most honest signal in fashion — a brand only reorders what already proved it sells.
3. Research trends without becoming a victim of them
Trend data is useful for calibrating direction and terrible for setting quantities. Search trends, competitor sell-outs, and category-level growth tell you where demand exists; they cannot tell you how much of it will choose your product.
Use trends to answer three questions before any production decision:
- Is the category growing, stable, or decaying? Producing into a decaying trend means your inventory ages from day one.
- How crowded is the space? A trending silhouette with forty competing brands needs differentiation before it needs depth.
- Does the trend fit your customer's actual life? A cargo-pant surge means nothing to your audience if they buy from you for heavyweight fleece.
4. Test customer interest before you commit to production
Between "we think people want this" and "we're making 500," there are cheap instruments that convert opinion into evidence:
- Product previews. Show the design to your actual audience. But read the response correctly: likes and "fire" comments are interest, not intent. Intent looks like "when is this dropping" and "will you have my size."
- Waitlists and email signups. An email address is a micro-commitment. A few hundred signups for a specific product is real signal; it typically converts to purchases at a fraction of the list, so size accordingly.
- Small paid tests. A limited run of samples sold through your channel is the strongest pre-production signal that exists — real money changed hands.
The discipline is matching the production quantity to the strength of the evidence. Compliments justify a sample run. Signups justify a small first order. Only paid purchases justify depth.
5. Use pre-orders to turn forecasting into measurement
Pre-orders are the cleanest answer to the new-product problem: instead of predicting demand, you collect it. Customers commit before you cut fabric, and the order book is the forecast.
This is no longer a fringe tactic. Entire brands now run on it — UK label Dancing Leopard rebuilt around made-to-order after being bought out of administration, testing with runs of 100–200 pieces and scaling only what sells. Their founder told Drapers that forecasting 9–12 months out is "basically guessing," so they stopped organizing around the guess. Industry analysts have pointed to models like this as a structural answer to the sector's overproduction problem.
Pre-orders carry obligations, though: you must communicate honest production timelines and actually hit them. This is where manufacturing speed stops being a convenience and becomes the business model. A pre-order campaign promising 8-week delivery against a 12-week production reality is a refund wave waiting to happen. Our repeat-order production runs 20–35 days; brands running pre-order models need that number — and their shipping time — pinned down before the campaign opens, not negotiated after.
6. Set the initial quantity with downside math, not upside math
Here's the decision frame that works: ask not "how many could we sell?" but "how many can we afford to be wrong about?"
Price your worst case. If the new style sells at half your hoped velocity, what does holding that stock for six months cost you — in cash tied up, in storage, in the markdown you'll eventually take? If the answer is "it would hurt but not kill us," the quantity is defensible. If the answer is "we couldn't fund the next drop," the quantity is too big regardless of how good the design is.
Practical anchors for a first run:
- Cap unproven styles at 4–8 weeks of hoped demand, not a full season.
- Respect MOQ structure — at 50 pieces per style for custom work, a two-style test is a 100-unit exposure, not a 500-unit one.
- Weight the size curve toward your proven sizes instead of splitting evenly; McKinsey has estimated poor size allocation alone can cost up to 20% of monthly profit.
- Keep enough budget uncommitted to reorder fast when the test reads positive.
7. Treat the forecast as a living number
The brands that forecast well don't forecast once. They set an initial quantity, then revise weekly against reality: velocity versus plan, sell-through by size and color, returns and reasons. The first two weeks of sales data are worth more than every pre-launch estimate combined.
This is also where AI tools earn their place — carefully. McKinsey estimates AI-driven demand forecasting can reduce inventory levels by 5–15% and improve stockout rates by 15–25%, largely by detecting pattern shifts in SKU-level data faster than manual review. But those tools amplify existing data; for a first collection with no history, they have nothing to amplify. Early-stage forecasting stays analog: evidence, small bets, fast revision.
Frequently asked questions
How many units should a new clothing brand produce for a first drop?
Enough to test, not enough to sink you. For most new streetwear brands that's 50–150 units total across one to three styles — at a 50-piece custom MOQ, a tight first drop is genuinely affordable to test. Size it by downside: if the entire run sold at half speed, could you still fund your next move? If yes, the quantity is defensible.
Can I trust social media engagement as demand evidence?
Only as the weakest tier of evidence. Likes measure attention; they don't measure willingness to pay. Weight signals by commitment level: compliments < follows < email signups < deposits < paid pre-orders < repeat purchases. Production quantities should be calibrated to the strongest signal you have, not the loudest.
Are pre-orders risky for a small brand?
The risk shifts but doesn't disappear. Pre-orders remove inventory risk and add delivery risk: you owe customers a product on a stated timeline. That makes confirmed production lead times non-negotiable — lock the manufacturing schedule before the campaign, include buffer, and under-promise on dates. A late first pre-order campaign damages trust far more than a small one.
How do I forecast sizes for a product I've never sold?
Borrow a curve, don't invent one. Use the size distribution of your closest comparable product — or for a true first product, a standard men's streetwear curve weighted to M and L — then correct it with your first two weeks of sales data. Even-split orders across sizes are almost always wrong and create size-level stockouts and overstock simultaneously.
How often should I update a demand forecast?
Weekly during a launch window, monthly for stable sellers. A forecast is a hypothesis with a date on it; sales velocity, sell-through by SKU, and channel signals either confirm it or demand revision. The brands that get inventory right are not the ones with the best initial guesses — they're the ones that correct fastest.
Conclusion
Demand forecasting for a new collection is not about finding the right number in advance — that number doesn't exist. It's about building a process that starts small, measures honestly, and scales only what proves itself. Borrow evidence from comparables, collect real buying signals before you cut fabric, and let your worst-case math — not your best-case hope — set the quantity.
Once the collection is live, the next problem arrives fast: the winners start selling out. That's a better problem, but it's still a problem — here's how to time your reorders so you catch demand instead of chasing it.