Demand Planning for New Products With No History

New product demand planning is the hardest forecasting problem in a consumer product business, because there is no history to project and the stakes are high. This guide is part of demand planning and inventory optimization, follows demand sensing and consensus demand planning, and leads to common demand planning mistakes and what demand planning is. A launch that is stocked too lightly loses momentum and retailer goodwill in the first weeks. A launch that is stocked too heavily leaves cash in the warehouse and a markdown problem in the second quarter. The honest approach is not to find a perfect number but to build a reasoned estimate, express the uncertainty as a range, buy in stages where the supply chain allows and watch early sell-through closely. This guide covers the main methods for forecasting without history, including analogs and attribute-based approaches, works through a launch estimate with checked arithmetic, shows how to apply a ramp curve, explains how to size the first buy under uncertainty, and sets out early-warning rules and a decision process for adjusting. The same problem in a financial setting is covered in forecasting sales for a new product, and launch inventory ties up cash, as described in cash flow management.

Start with the questions

Before choosing a method, collect what you do know. What is the product, and who is it for? What is its price relative to similar items? Where will it be sold, in how many doors or channels, and with what support? What is the launch timing, the marketing plan and the promotion plan? Is it replacing an existing item, extending a line or opening a new category? What are the supply constraints: minimum order, lead time, whether reorders are possible and how fast? The answers determine both the forecast and how much risk the company can take.

Method 1: analogs

An analog is an existing or past product that resembles the new one in ways that drive demand: category, price tier, brand, channel, target customer, pack size, seasonality. The planner looks at how the analogs sold in their launch periods and scales for differences. Choose several analogs, not one, to avoid anchoring on a single lucky or unlucky launch. Record why each was chosen.

Worked example

Suppose three comparable launches sold the following in their first 13 weeks (illustrative): Analog A, 1,200 units across 40 doors; Analog B, 900 units across 25 doors; Analog C, 1,500 units across 50 doors. Units per door: A 30; B 36; C 30. Using total units over total doors gives (1,200 + 900 + 1,500) / (40 + 25 + 50) = 3,600 / 115 = 31.3 units per door. The new product will launch in 80 doors, so the unadjusted estimate is 31.3 × 80 = 2,504 units. The new item is priced about 10 percent above the analogs, and the team judges that this reduces volume by 10 percent, so the adjusted estimate is 2,504 × 0.9 = 2,254 units over the first 13 weeks. Check: 3,600 / 115 = 31.304; × 80 = 2,504.3; × 0.9 = 2,253.9.

Notice the structure: a base rate from comparable launches, scaled by distribution, adjusted for specific differences, each adjustment visible. This is far better than a number from a hunch, and it can be debated.

Method 2: attribute-based models

When you have many past launches, you can model sales as a function of product attributes, such as price, category, pack size, color or brand, and distribution. A simple regression or machine learning model can estimate demand for a new combination of attributes. This needs enough history of launches to learn from, and attention to avoid false precision.

Method 3: pre-launch signals

Signals before launch can sharpen estimates: retailer commitments and planogram space, pre-orders, waitlist sign-ups, sampling results, social engagement, search interest, test market results and early reviews. Treat these as evidence to adjust the analog estimate up or down, not as forecasts in their own right. A waitlist of 5,000 emails does not mean 5,000 sales, but a conversion rate learned from earlier launches can convert it into an estimate.

Method 4: structured judgment

Where data is thin, structured expert judgment is better than guessing. Ask several people independently to estimate a low, expected and high outcome, share the results, discuss the reasons for differences and revise. Averaging independent estimates tends to cancel individual biases. Write down the assumptions so that you can compare them with outcomes later.

Apply a ramp curve

Launch demand does not arrive evenly. There is usually a fill-in of the distribution pipeline, a peak, and a settling to a base rate. Estimate the shape from analogs and spread the total across weeks. Suppose the 13-week shape from analogs is: 5, 7, 8, 9, 9, 9, 9, 8, 8, 8, 7, 7, 6 percent of the total (illustrative). These add to 100: 5 + 7 + 8 + 9 + 9 + 9 + 9 + 8 + 8 + 8 + 7 + 7 + 6 = 100. Applied to 2,254 units, the first four weeks are about 113 (5%), 158 (7%), 180 (8%) and 203 (9%) units, a total of 654 units, or 29 percent of the 13-week estimate. The practical use is that the first purchase must cover the early weeks plus the pipeline fill, and the reorder lead time determines how much more is needed before replenishment can arrive.

Express the range

A single number conveys false certainty. Estimate a low and a high case. If past launches in the category have varied by roughly plus or minus 35 percent around the analog estimate (illustrative), the range around 2,254 is 2,254 × 0.65 = 1,465 to 2,254 × 1.35 = 3,043 units. The range is wide, and that is honest. It tells you that a first buy at the expected value may be 50 percent short if the product succeeds or 54 percent too much if it fails (3,043 / 2,254 = 1.35, and 2,254 / 1,465 = 1.54).

Size the first buy

The right first buy depends on the economics of being wrong. If the product is high-margin, reorders are slow and a stockout would harm the retailer relationship, buy closer to the high case. If margins are thin, shelf life is short, minimum orders are low and fast reorders are possible, buy close to the low case and replenish. Consider a product with a 12-week reorder lead time. The first buy must cover the 12 weeks before any reorder can land, plus the pipeline fill. Using the weekly shape above, the first 12 weeks represent 94 percent of the 13-week total, so a first buy at the expected case should be nearly all of it, about 2,119 units, while the low case suggests 1,377. Where lead times are long, the first buy is effectively the launch decision, and the discussion should involve finance, because the exposure is cash. Where lead times are short, buy the low case and add quickly. Options like split shipments, postponement of final configuration or reserving supplier capacity can reduce risk.

Monitor early and act fast

Define early-warning rules before launch. For example: review sell-through weekly; if cumulative sales after four weeks are more than 25 percent above the expected shape, trigger a reorder review; if they are more than 25 percent below, pause reorders and review the marketing and distribution plan. Look at sell-through per door, not only total sales, to separate distribution effects from demand effects. Track repeat purchase where possible, because trial and repeat tell different stories: a high first-week trial that does not repeat is a warning sign. Update the forecast with actual data quickly, using a method that blends the prior estimate with observed sales, giving more weight to observations as they accumulate.

Cannibalization and transitions

New items often take volume from existing ones. If the new product replaces an existing item, the demand plan for both should be linked: reduce the old item's forecast as the new one ramps up and plan the last buy and clearance of the old item so as not to leave stranded stock. If it is an extension to a line, estimate the share of new volume that comes from the existing items and the share that is incremental. Be skeptical of launch plans that count all new volume as incremental, since customers have limited demand for the category.

After the launch

Compare the plan with actuals at 4, 13 and 26 weeks. Record the assumptions that were wrong and the analogs that predicted best. Build an internal library of launches with their attributes and results, which becomes the base for analogs and models over time. A company that does this for a few years forecasts launches far better than one that treats each as unique.

Common mistakes

1. A single analog, chosen because it flatters the plan.
2. Treating the sales target as the forecast.
3. No range, so no sensible stocking decision.
4. Ignoring lead time when sizing the first buy.
5. Counting all new volume as incremental.
6. No early-warning rules.
7. No post-launch review.

FAQs

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

Use analogs of similar past launches scaled for differences in distribution and price, supported by pre-launch signals and structured judgment, and express the result as a range.

How many analogs should I use?

Three to five is common. Using several reduces the risk of anchoring on one unusual launch.

How much inventory should I buy for a launch?

It depends on margin, lead time, shelf life and the cost of being wrong. With long lead times, the first buy is close to the expected or high case, and with fast reorders, closer to the low case.

How soon should I adjust a launch forecast?

Weekly in the first weeks, using sell-through per door and repeat purchase. Define trigger rules beforehand.

What if the new product replaces an old one?

Link the two forecasts, reduce the old item's demand as the new item ramps and plan the sell-through of remaining stock.