New product demand forecasting without sales history

New product demand forecasting starts with comparable launches and explicit assumptions about who will buy, how often, and how much. Build those estimates separately, then compare them with the new product's first sales. An estimate without history becomes useful when its assumptions can be checked.

Define the launch you are forecasting

Specify the product, channel, geography, unit of measure, and launch window before choosing an analogue. A US distributor estimating cases ordered by dealers has a different question from a manufacturer estimating individual items purchased by end customers. Shipments into a channel and sales out of it need separate labels.

Use the same start point for both products: their first available selling period, not their first calendar month in a report. Record active locations and availability alongside sales. Otherwise, a comparison can reward the older product simply because it reached more locations sooner.

Forecasting: Principles and Practice discusses judgmental forecasting when history is unavailable, including analogies and customer intentions. Its practical requirement is documentation: preserve the reasoning so later observations can test it.

For products with an established sales record, use the separate existing-product demand workflow. This launch calculation should not replace a history that already exists.

Calculate the comparable-product estimate

A US safety-products distributor launches a new glove. The planning window is the first four selling weeks. The analogue's sales are already expressed as units per active door across that entire window; do not multiply by weeks again.

Put this table on an Analogue worksheet, starting in A1. Inputs are assumptions or observations supplied by the planning team. Actual sales are added only after the window closes.

In D2, F2, and G2, enter these formulas and fill down through row 3:

=B2*C2
=D2-E2
=ABS(F2)

The analogue forecast is 24 × 50 + 40 × 30 = 2,400 units. Its total overforecast is 500 units. The total signed error must also equal SUM(D2:D3)-SUM(E2:E3). This identity checks the arithmetic; it does not establish that the analogue is appropriate.

Keep a short reason for each rate: comparable shelf placement, planned selling support, and price differences. A convenient historical product with a different purchasing pattern is a weak comparison even when its export is complete.

Build an independent adoption estimate

Now estimate purchasing from reachable accounts. Eligible accounts here are distinct end-customer accounts served by the same active dealer or chain locations used above; accounts and doors describe different levels of the same launch coverage. The conversion assumption refers to buying within this launch window. Units per order means units per buying account here because the example assumes one order per buyer.

Put this table on an Adoption worksheet, starting in A1:

In E2, G2, and H2:

=B2*C2*D2
=E2-F2
=ABS(G2)

Fill down through row 3. The calculation is 500 × 20% × 8 = 800 units for dealers and 200 × 30% × 20 = 1,200 for chains. The eligible accounts and active doors measure different things; never add them together.

An independent check of total adoption demand is:

=SUMPRODUCT(B2:B3,C2:C3,D2:D3)

Microsoft documents SUMPRODUCT as multiplying corresponding array entries and summing the results. Its ranges must have matching dimensions. Covirage is not affiliated with Microsoft.

Compare both with early actual sales

Use absolute error as well as signed error. Across multiple channels, overforecasts can cancel underforecasts in the signed total.

Weighted absolute percentage error = Sum of absolute unit errors / Sum of actual units

The analogue estimate produces 500 / 1,900 = 26.3%; the adoption estimate produces 100 / 1,900 = 5.3%. Percentages are rounded to one decimal place. These results favor the adoption estimate for this completed window, not for every future launch.

On the adoption worksheet:

=SUM(H2:H3)/SUM(F2:F3)

Test the sensitive assumption before changing the plan. If dealer conversion alone falls from 20% to 15%, dealer demand becomes 500 × 15% × 8 = 600 units. Total adoption demand becomes 1,800, which is 100 below actual sales. A modest change reverses the direction of the miss.

Preserve the original assumptions when making that revision. The broader forecasting without historical data workflow explains how to add confirmed orders without double counting scenarios. Use forecast bias by SKU and site when enough observations exist to examine repeated directional misses. As the product develops, the product life-cycle workflow changes the evaluation window and assumptions.

Where it goes wrong

• Different launch exposure. Normalize for active locations and selling time before comparing products.
• Survey interest treated as orders. Conversion remains an assumption until buying behavior is observed.
• Channel loading treated as adoption. Initial stocking orders may not reflect repeat customer demand.
• Stockouts treated as weak demand. Preserve availability flags; observed sales can be limited by supply.
• A revised forecast replacing the original. Keep issue dates so an early estimate is never scored using knowledge gained later.

Compare your launch assumptions in Covirage

Bring the launch assumptions, comparable-product export, and early actual-sales file to Covirage. Covirage maps the columns; its tools compute and check the supplied estimates and comparisons, and the external AI model explains the results with every figure traced to its rows. Create a free Covirage account, then set up your business and bring a sample of your own data.

See Covirage for supply chain teams for the broader planning context.