How to Forecast Demand for a New Product

There's no sales history to lean on, the buy is due Friday, and "pick a number" is somehow the whole forecasting plan. Every brand that launches products lives this moment. The good news: no history doesn't mean no signal. It means the signal lives somewhere other than this SKU's sales curve.

To forecast demand for a new product with no history, base it on analog products that resemble it, layer in pre-order and waitlist signals, and apply structured judgment from your team. Start with a comparable SKU's launch curve, adjust for differences, and build a range rather than a single number. Treat the first forecast as a starting estimate you'll correct fast.

Key takeaways

• No history doesn't mean no signal: the signal lives in analog products, pre-orders, and waitlists instead of this SKU's empty sales column.
• The range is the forecast, not the point: a 620-to-1,150 band with named assumptions beats a confident 884 that hides them.
• Signals move you within the range: strong pre-order pace argues for the top of the band, a quiet launch list for the bottom.
• The re-forecast matters more than the forecast: schedule it for one to two weeks after launch, when real sales replace every assumption.

Why is new-product forecasting different?

Because every standard forecasting method extrapolates the past, and a new product doesn't have one. You can't trend, smooth, or seasonalize a blank column. So the discipline changes: you borrow signal from products that already lived the life this one is about to live, and you accept a wider error band as the honest price of launching. The goal isn't a precise number; it's a defensible range and a plan to correct it quickly.

The "cold start" problem

Planners borrow the term from recommendation systems: with zero data on a new item, any model's first guess is only as good as what it can infer from similar items. That's not a flaw to apologize for, it's the actual method. The whole craft of new-product forecasting is choosing what "similar" means (price, category, audience, channel) and being systematic about it instead of grabbing the nearest optimistic memory.

How do you pick an analog product?

Pick a past SKU whose launch conditions match the new one where it matters, then use its early-life sales curve (the first 4 to 12 weeks) as your baseline shape. Four matching criteria, in rough order of weight:

• Price point: demand behaves differently at $30 and at $120, even inside one category.
• Category and use case: the analog should answer the same customer need, not just live on the same collection page.
• Target customer: same audience segment and channel mix, or the curve shape won't transfer.
• Launch push: an analog that launched with a paid campaign can't baseline a quiet organic drop.

The analog's job isn't prophecy; it's giving you a realistic curve shape and magnitude to reason from, drawn from your own store's reality rather than hope.

Adjusting the analog for differences

No analog is a twin, so you adjust. Work through the differences one at a time and put a factor on each: launching to an email list twice the size the analog had? Nudge up. Price point 20% higher? Nudge down. Weaker launch marketing, extra colorway, different season? Each gets its own multiplier, written down so the assumptions are visible and arguable.

A worked shape of the math: the analog sold 800 units in its first 8 weeks. The new product launches to a bigger audience (×1.3) at a higher price (×0.85), with a comparable marketing push (×1.0). Point estimate: 800 × 1.3 × 0.85 = 884 units. Then honor the uncertainty by turning it into a range, say ±30%: roughly 620 to 1,150 units. The range, not the point, is the forecast. (The taxonomy of methods behind this, including judgmental and market-test approaches, lives in new product forecasting methods.)

How do you use pre-order and waitlist signals?

Pre-orders, waitlist signups, and back-in-stock requests are real demand you can read before the launch buy is final. They're the difference between forecasting in the dark and forecasting with a flashlight: a paid pre-order is a customer who already decided. Email signups for a launch list are softer but still directional, and their conversion behaves like your past launch lists did.

Turning signals into a number

Anchor each signal to its historical conversion. If launch-list emails have converted at around 5% in past launches and the new list holds 4,000 subscribers, that's roughly 200 launch-window units of evidence. Paid pre-orders count closer to one-for-one, minus your typical cancellation rate. Use the signals to move within your forecast range: strong pre-order takeup pushes you toward the top of the 620-to-1,150 band, a quiet list argues for the bottom. Signals sharpen the range; they rarely replace it.

What's the step-by-step approach?

Six steps, in order:

• Pick the analog: the past SKU that best matches on price, category, audience, and launch push.
• Set the baseline curve: its first 4 to 12 weeks of sales, as shape and magnitude.
• Adjust for differences: one written multiplier per difference (audience size, price, marketing weight).
• Layer in pre-order signal: convert committed demand and list size into evidence.
• Build the range: point estimate ±30% or wider, honestly reflecting how good the analog is.
• Schedule the re-forecast: one to two weeks after launch, on the calendar, before launch day.

The last step is the one teams skip, and it's the cheapest: the fastest way out of a wrong launch forecast is catching it while the reorder window is still open. (What quantity to actually order against the range, including sell-through targets and using pre-orders to de-risk the buy itself, is owned by planning inventory for a product launch.)

This whole workflow is also exactly what look-alike modeling automates. Conative AI builds a launch baseline from look-alike products (similar price, category, and audience, matched from your own catalog's history), so a new SKU starts with a forecast grounded in evidence instead of a blank cell, then corrects itself as real sales arrive. AI-powered demand forecasting handles the analog matching and the fast re-forecast; your team spends its judgment on the adjustments only a human can see coming. Start your free trial and watch a launch baseline build itself.

Frequently asked questions

How do you forecast a product with no data?

Borrow signal: base the forecast on an analog product with a similar price, category, and audience, using its early sales curve as the baseline. Adjust for known differences, add pre-order or waitlist evidence where you have it, and express the result as a range. Then re-forecast within weeks of launch, once real sales exist.

What's an analog product in forecasting?

An analog is an existing product whose launch resembles the new one closely enough to borrow its sales curve as a baseline: similar price point, category, target customer, and launch push. You adjust its numbers for the differences that remain. A good analog comes from your own catalog, where the audience and channel already match.

How accurate is a new-product forecast?

Meaningfully less accurate than a forecast for an established SKU, and that's normal: there's no history to anchor on. Plan for error bands of ±30% or wider rather than pretending precision. Accuracy improves quickly once launch sales arrive, which is why the re-forecast a week or two in matters more than the initial number.

Can pre-orders predict launch demand?

They're one of the strongest signals available, because a pre-order is a committed purchase rather than an opinion. Count them near one-for-one minus your usual cancellation rate, and read their pace against past launches. They cover the launch window best; demand after the initial spike still needs the analog-based forecast underneath.

How long until a new product has enough data to forecast normally?

Standard methods start working once a stable pattern emerges: often 8 to 12 weeks of sales for a steady product, longer if demand is noisy or seasonal. The practical approach is blending, weighting the analog baseline less and the product's own history more each week, rather than switching methods on a single cutoff day.

How do you forecast a brand-new category?

With no in-catalog analog, triangulate: use judgmental forecasting with structured inputs (multiple estimates, ranges, written assumptions), external reference points from comparable market launches, and where the buy is large, a market test such as a small pre-sale or limited drop. Order conservatively with a fast reorder path, and let early sales choose the trajectory.