The most useful sales history for a product that has never sold usually belongs to another product. That sounds odd until you look at what is already in your data: items with a similar price, the same channels, comparable marketing support. Meanwhile the launch date is set, the supplier wants a quantity, and your usual forecast has nothing to work with.
Without its own sales history, a new product’s first order can go wrong in two ways. Order too much and cash sits on shelves. Order too little and a product that works sells out in its first weeks. The way out we recommend is not a braver guess. It is a first order sized across low, base and high scenarios, built on comparable products you have tested, with a clear point where real sales replace the assumptions.
What new product demand forecasting is and which type of launch you are facing
New product demand forecasting estimates demand for a product before and during launch, when sales history is thin or missing. Time-series models project a product’s past forward; when no such history exists, use proxy data from comparable products, market signals and explicit scenarios. Before choosing a method, identify the type of launch you are planning, because that decides what data you can borrow.
Demand forecasting for new products: the three launch types
The first type is a genuinely new product: nothing in your catalogue resembles it. With no analog, you forecast from SKU-level forecasting based on attributes such as price point, features, target market size, distribution and marketing support.
The second is a new SKU in an existing family, such as a new colour, size or flavour. Its siblings are your proxy. The open question is how much of their demand it will take and how much it will add.
The third is a replacement. The old product’s sales are your baseline: estimate the share that transfers to the successor, then model any incremental demand separately. Adding the old product’s full volume to the successor’s transferred volume counts the same customers twice; adding incremental demand does not. Check the life cycle stage of the predecessor too. If the decline was specific to the predecessor, its recent sales may understate what the successor could sell; if it reflects the wider category, they may be a more realistic baseline.
Demand forecasting methods for new products when sales history is missing
With few past launches to learn from, analogy usually wins; as the number of launches grows, models start to earn their place. Our overview of forecasting methods covers the wider field. The comparison below is limited to launches.
Analogy, regression, transfer learning and ensembles compared
Analogy borrows the launch curve of a comparable product and adjusts it. It needs one good match and your own sales data, which makes it the most accessible method. Regression relates attributes such as price or distribution to early sales across your past launches. It works only if you have enough past launches to fit it.
Transfer learning trains a model on the histories and attributes of many products, then uses the new product’s attributes (and, once available, its first weeks of sales) to generate a forecast. It suits large catalogues and is harder to explain to a buyer. Weighted ensembles blend the outputs:
Forecast = w1 × F(analogy) + w2 × F(regression) + w3 × F(transfer learning), with the weights summing to 1.
Start with equal weights, then shift them toward whichever method tracked the first weeks best. With a small catalogue and few past launches, regression and transfer learning have too little data to learn from. In that case, analogy plus documented expert judgment will serve you better than a model with nothing to learn from.
Lookalike products: borrowing history and adjusting it for your launch
A lookalike is only valid if it matches the new product on price, features, distribution, target market size and marketing support. If it differs sharply on any one, either adjust for the gap or pick another product. Align the data by weeks since launch to compare launch trajectories, but adjust separately for seasonality and calendar timing. In Tannico’s case, the planning team faced high forecast errors across a large, seasonal catalogue, difficulty accounting for promotions, holidays and vintage-specific demand, and the risk of stockouts or overstock on high-demand SKUs. The company needed accurate, actionable forecasts across thousands of SKUs so its buyers could make confident, proactive purchasing decisions.
The calculation is: Adjusted forecast = analog units in the launch window × distribution factor × marketing factor × price factor.
Here is an illustrative case, with numbers you should replace with your own. The analog sold 600 units in its first 8 weeks on two channels. The new product launches on three channels (factor 1.5), with similar marketing support (1.0) and a price 10% higher (a judgment-based 0.9). The result is 600 × 1.5 × 1.0 × 0.9 = 810 units. In Excel, put the four inputs in B2:B5 and enter =B2*B3*B4*B5 in B6.
The 1.5 distribution factor assumes the new channel sells like the average of the existing two and brings in new customers rather than shifting them from those channels. Write each factor down with its reasoning. These are the first numbers you will test against real sales, and a single figure like 810 is not yet enough to size the first purchase.
How to size the first order across low, base and high scenarios
Treat the 810-unit result as a base or most likely scenario, not as a promise. For illustration only, a low scenario could use a 1.2 distribution factor, 0.9 marketing factor and 0.9 price factor, giving 583 units; a high or best-case scenario could use 1.8, 1.1 and 1.0, giving 1,188 units. These are not standard multipliers: set them from the uncertainties in your own launch. This scenario planning approach makes the worst-case, most likely and best-case outcomes explicit. Once the first weeks of sales arrive, replace the assumptions with actual demand and rework the order and replenishment plan.
Test the method on a launch you have already lived through
Take a product you launched last year and choose the lookalike you would have picked at the time. Apply the three factors with the reasoning you had then, and compare the result with the product’s first eight weeks of actual sales. The widest gap shows which assumption to document and challenge first.
As launches multiply, blending methods and reweighting them on early sales starts to pay off. The factor-by-factor comparison from this test, with the reasoning next to each number, is what to have in hand when you discuss your launch with the Intuendi team.
