Forecasting Inventory for New Products (No Sales History)

Every forecasting method on this site so far assumes one thing: that the product has sold before, so there's a sales history to calculate from. A brand-new SKU has none. Day one, it has zero average daily sales, zero standard deviation, zero anything a formula could use — and you still have to decide how many units to order.

The Cold-Start Problem

Operations research literature calls this the cold-start problem, and it's worth being honest about upfront: every method below is inherently lower-confidence than forecasting an established product with real sales history. That's a structural feature of the problem, not a shortcoming of any one method — you're substituting analogy, judgment, or a thin early signal for the direct historical data a moving average or smoothing model actually needs. A promise of accuracy here would be a lie; the honest goal is a defensible starting estimate and a fast way to correct it. This is also why seasonal forecasting doesn't apply yet — a seasonal index needs a year or more of the product's own history to build, which a brand-new SKU by definition doesn't have.

A peer-reviewed review of new-product forecasting groups the documented approaches into four families: judgmental methods (comparing the new product to similar existing ones, expert opinion, structured panels like the Delphi method), consumer research (concept testing, pre-launch test markets), diffusion models (adoption curves borrowed from comparable mature products), and machine-learning methods that need more training data than a small catalog has. Its own conclusion is the most useful part here: combining more than one method beats relying on any single one.

Using Analogous Products

The most practical version of this, and the one the research literature calls "product life-cycle analogy," is what merchants already do informally: find an existing SKU that's as close as possible to the new one — same category, similar price point, same season, same customer — and use its early sell-through curve as a template, not just its total volume.

Say you're launching "Sea Salt & Cedar, 250g" alongside your established "Cedar & Fig, 250g" line. Cedar & Fig has years of history and sells a steady 5 units a day. Sea Salt & Cedar shares the same customer base, the same price point, and the same 12-day supplier lead time — but it's unproven. There's no formula that tells you what fraction of Cedar & Fig's volume a new, related scent will open at; that discount is the judgment call analogous forecasting always requires. A merchant who knows their own audience might reasonably expect a closely related product to open somewhere between half and full established volume in its first weeks — that's a starting assumption to test against real orders, not a number to treat as settled.

Other Early Demand Signals

Analogy isn't the only signal available before launch. Pre-orders, wishlist adds, teaser-ad engagement, and "notify me when available" sign-ups all carry real information, even though none convert to sales at a known, fixed rate. Treat each as a directional adjustment to the analogous estimate — a strong wishlist response leans you toward the higher end of your range, a quiet one toward the lower end — rather than a number you plug into a formula. No source gives a defensible conversion rate for any of these signals; treat anything that hands you one specific figure as a guess, not a fact.

Sizing the First Order

Once you have a demand estimate, however rough, the same basic mechanics apply as any reorder calculation: you need enough stock to cover the supplier's lead time, plus a margin for being wrong.

Continuing the example: the range above was "somewhere between half and full established volume" — 2.5 to 5 units a day for Cedar & Fig's 5-a-day baseline. Say you land on 3 units a day, toward the conservative end, since a first order is exactly where caution earns its keep. At the same 12-day lead time, lead-time demand alone is 3 × 12 = 36 units. Because that 3-units-a-day figure is a guess rather than a measured average, the usual safety-stock math doesn't apply cleanly — there's no real standard deviation to plug in yet. The practical move most small stores make instead is to round up meaningfully beyond the bare lead-time number, enough to avoid stocking out if the guess runs low, without ordering so much that a wrong guess ties up cash in unsold stock. There's no formula that removes this judgment call; treat the final number as a starting position, not a target to defend.

Correcting Fast Once Data Arrives

The moment real orders start coming in, they're worth more than any pre-launch estimate, however carefully reasoned. Recalculate as soon as there's a meaningful handful of actual sales — don't wait for a fixed period to elapse first. No source this pack could verify gives a defensible universal number of days or weeks to wait before trusting early sales; treat any content that states one specifically as a convenient-sounding guess, not a documented standard. What matters is the habit of checking the estimate against reality often, not the exact interval.

Watching a new SKU's first days of orders and manually recomputing the estimate is the kind of tracking that holds up for a week and quietly stops after that. StockCue doesn't have a special cold-start mode either (nothing meaningfully does, for a product with zero history), but the moment a new SKU accumulates enough of its own sales data, its forecasting takes over automatically from your manual estimate, recalculating nightly from that product's own order history rather than the launch guess you started with.

Once a product has real history, comparing forecasting methods and the reorder point formula take over from the estimate in this post — or start from the complete guide to Shopify inventory forecasting if you want the full process end to end.

Frequently Asked Questions

How do you forecast demand for a product with no sales history?

You don't forecast it the way you would an established SKU: there's no sales history to calculate a velocity from. Instead you estimate: comparing the new product to similar existing ones (analogous forecasting), reading whatever early signals exist before launch, and sizing the first order conservatively so a wrong guess is cheap to correct.

What is analogous forecasting?

Analogous forecasting means finding one or more existing products that are as similar as possible to the new one (same category, price point, season, and target customer) and using their early sales pattern as a starting template, adjusted for any known differences between the two products.

How many units should you order for a brand-new product's first run?

Enough to cover your supplier's lead time at your estimated demand rate, sized deliberately conservative because the estimate itself is a guess, not a measured average. There's no formula that removes the judgment call: the goal is an order small enough that a wrong guess doesn't tie up excessive cash, and large enough to actually learn something from real sales.

When does a new product have enough data to switch to normal forecasting?

There's no universally agreed number of days or weeks: treat any source that states one specifically as a guess dressed up as a rule. What's defensible is the practice itself: recalculate your estimate every time meaningful new sales data arrives, rather than waiting for a fixed period to pass before you're "allowed" to update it.