Forecasting demand for new products when you have zero history

Short answer: you don't forecast a new product from nothing — you forecast it from its nearest existing relative, size the first buy for learning rather than coverage, and treat the first 60 days as data collection. The forecast will be wrong; the design question is whether it's wrong cheaply. Here's the zero-history toolkit and a fill-in worksheet to run it on your own launch.

The worksheet: fill these in first

Everything below assumes you've filled in five inputs. Do it now, in your spreadsheet, before reading the methods:

1. The analog SKU — your closest existing product by price band, use case, and audience. Not your bestseller: your most similar seller. Write its SKU.
2. The analog's 90-day run rate — units per week, in-stock days only. Why "in-stock days only": weeks where the analog sat unavailable record zero sales, which reads as zero demand — the correction is explained in the velocity method; the forecasting guide covers the same principle. Use whichever you already compute.
3. Your launch spend and audience delta — will this product get more attention than the analog ever got (email list, launch push, an influencer), less, or the same? Write a multiplier: 2x, 1x, or 0.5x.
4. Cash you can afford to strand — the dollar amount that could sit unsold for six months without hurting. Not the amount you wish you could deploy. The amount you'd genuinely forgive.
5. Supplier lead time and minimum order — weeks, and units. Write both.

Method and sizing follow from these five numbers. (If you already run structured forecasting for your existing catalog, the method ladder — moving averages through seasonality-aware models — lives in the forecasting guide; everything below is what you do before any of those methods have history to work with. Shopify's own Help Center materials on inventory forecasting make the same point: no method produces good output from zero sales data, accessed October 2026.)

Method 1: the analog forecast (the workhorse)

Take the analog's in-stock-only weekly rate, apply your attention multiplier, and you have a launch estimate. Here it is worked on a real-shaped example — a fishing-lure seller launching a new chatterbait style:

• Analog: their existing spinnerbait, which sells 14 units/week across its variants, in-stock days only.
• The new lure sits at the same $12.99 price point, same water conditions, same audience — a strong analog. Multiplier: they're launching to an email list of 9,000 that the spinnerbait never had, so 1.5x.
• Launch estimate: 14 × 1.5 = ~21 units/week, first 90 days.

Honest caveats, because the analog method's errors are predictable:

• Analogs overshoot when the new product competes with the analog itself. If the chatterbait is close to the spinnerbait, some of its demand is the spinnerbait's demand moving house — your total units don't grow by 21/week; the split does. This is cannibalization, and it's detectable from your own data a few weeks in (the detection method is covered in the cannibalization guide).
• Analogs undershoot when the launch has genuine new attention (a new platform, a retail account, a viral angle). No spreadsheet fixes that; the multiplier is a judgment input, so write your reasoning down next to it.

Method 2: the small test buy (what the forecast is actually for)

Here's the reframe that matters more than any formula: at zero history, the first buy is not a coverage decision — it's an experiment with a sample size. You're buying information as much as inventory.

The worked sizing, continuing the lure example:

• Analog-based estimate: ~21/week. Lead time: 6 weeks. Supplier minimum: 300 units.
• A naive coverage buy — 90 days of demand plus safety buffer — says 21 × 13 weeks + buffer ≈ 350+ units.
• The test-buy logic says: buy the 300 minimum, structure the launch so demand reveals itself before you need a reorder, and set the reorder trigger inside the launch window. If the first 3 weeks sell at 20+/week, reorder immediately — you'll ride out the 6-week lead time on the remaining stock if you trigger by week 3–4.

The arithmetic that makes the test buy safe: 300 units ÷ 21/week ≈ 14 weeks of stock — more than one full lead time of cover. Even if the forecast was 50% low (31/week), you still hold ~10 weeks, which outlasts the reorder cycle. Even if it was 50% high (10/week), you're carrying a slow mover — annoying, not fatal — and the next buy corrects to reality.

The discipline: never let a first buy exceed roughly two lead-times of cover unless the MOQ forces it — and if the MOQ forces you past that line, that's the moment to renegotiate terms or rethink the launch, not to close your eyes and order. (The full quantity logic for established SKUs is different — it's covered in how much inventory to order; the how-much-inventory-to-start article owns the first-line decision for a brand-new catalog.)

Method 3: preorders — demand signal you can actually measure

Preorders convert the riskiest weeks of a launch — before you know anything — into measurement. Offer the new product for preorder with a stated ship window, and the order count is not a forecast at all; it's demand, realized, with money attached.

Where it earns its keep:

• It sizes the first buy from evidence instead of analogy. If the preorder pulls 90 orders in two weeks from a 9,000-person list, you have a conversion-anchored baseline to extrapolate from — crude, but anchored to your own audience, not someone else's benchmark. (Shopify's own guidance on market validation makes the same case — sell before you stock, via preorders or small test batches; Shopify blog and Help Center, accessed October 2026.)
• It de-risks the MOQ. A 300-unit minimum hurts less when 90+ units are already spoken for.
• It has a failure mode worth naming: the ship window slips, and preorders are unforgiving of slips — buyers who paid up front have lent you money and expect the date. Only run preorders against a supplier date you'd bet on, and pad the promise. The mechanics of running preorders on Shopify — deposits vs. full charge, fulfillment windows — are covered in the preorder guide.

Method 4: when to abandon the forecast

Every zero-history forecast needs a kill clause — a point where you stop trusting the model and start trusting the data, even when the data says something the launch plan didn't. Set it before launch, in writing:

• Week 4 checkpoint: compare actual weekly sales to the analog estimate. Within ±40%: keep the model, adjust the reorder math to actuals. More than 40% low: stop reordering at the model's rate — re-forecast from the four real weeks, and check the cannibalization question. More than 40% high: reorder immediately at actuals, and treat the launch multiplier as validated for your next new product.
• Week 8 checkpoint: the model is now retired. From here, this SKU has its own velocity history and joins your normal replenishment routine like every other SKU. New-product forecasting is a bridge, not a residence — the goal is to hand the SKU to your standard method as fast as the data allows.

The single most expensive pattern in new-product forecasting is the one where the forecast survives contact with reality: the launch underperforms, the founder keeps "giving it time," and the reorder placed on model optimism arrives into demand that was never there. The kill clause exists to prevent you from being the person who does that. (Write the checkpoints into your ops notes the same day you fill in the worksheet — a checkpoint that lives only in your head is a checkpoint that moves.)

Common questions

How do you forecast demand for a product with no sales history? Forecast from an analog: your most similar existing SKU's in-stock-only weekly rate, adjusted by an attention multiplier for how much more or less the launch will be promoted. Then size the first buy as a test — enough to survive one reorder cycle even if the estimate is low — and replace the analog model with real data at a week-4 checkpoint.

What is analog forecasting? It's estimating a new product's demand from the closest existing product you already sell, matched on price band, use case, and audience. Its known errors: it overshoots when the new item cannibalizes the analog, and undershoots when the launch gets attention the analog never had. Both errors are correctable at the week-4 checkpoint.

How much inventory should I order for a new product? Enough to cover roughly two lead times of demand at your analog-based estimate — often just the supplier's minimum order. A 300-unit minimum at a 20-units-per-week estimate is about 14 weeks of cover, enough to trigger a data-based reorder inside the launch window. Treat the first buy as an experiment, not a coverage commitment.

Are preorders a good way to measure demand for a new product? Yes — preorder orders are realized demand with money attached, not an estimate, and they can offset a high MOQ. The failure mode is the slipped ship window: buyers who paid up front expect the date. Only run preorders against a supplier timeline you'd bet on, with the promise padded.

When should I stop forecasting a new product and use normal reordering? At the week-8 mark, or after roughly eight weeks of real sales, whichever comes first. By then the SKU has its own velocity history and belongs in your standard replenishment routine. The forecast is a bridge to that point — keeping the launch model alive past week 8 is how slow launches turn into dead stock.

Where to go next

Before your next launch, fill in the five worksheet inputs at the top of this article — analog SKU, its in-stock-only run rate, your attention multiplier, your strandable cash, and the supplier's lead time and minimum. That single row of numbers decides your first buy. And once the SKU is past week 8 and has real history, hand it to the method ladder in the forecasting guide — that's where established-SKU forecasting lives, and this article deliberately doesn't re-teach it.