A new-product forecast needs more than one method

A product without history does not justify a single confident number. It changes the job. Before launch, the team must decide whether it is estimating a market boundary, purchase potential, an adoption path, or an investment outcome. Each question exposes a different part of the future.

The practical answer is a method matrix. Use independent methods where they observe different things, reconcile them around one defined product concept, and keep every blind spot visible. Thomas combines multiple market-potential methods. Urban, Weinberg, and Hauser condition a really-new-product decision on market measurement, learning, engineering information, and judgement. Oren and Rothkopf add the market clock. Aydin, Kwong, Ji, and Law keep uncertainty in the estimate. None of these studies says that more methods automatically produce a better forecast.

Why must new product forecasting define the economic unit before modeling?

For an established product, a sales history can hide a large number of assumptions. A new product cannot offer that convenience. The first question is therefore not which model is most sophisticated. It is what the forecast is meant to support.

Is the team screening a market? Estimating purchase potential for an offer? Planning production and availability? Testing whether customers can learn the category? Deciding whether to fund the launch? These are different objects. A method that is useful for one can be a poor answer for another.

Why do triangulated forecasting methods require a unified demand concept?

Thomas’s contribution is not a vote-counting rule. It begins with a defined new-product concept, obtains independent operationalisations, and uses the concept to reconcile the forecasts. The independence matters because methods that share the same hidden assumption are not independent merely because they have different names.

Make the reconciliation record explicit. For each output, write the product definition it represents, the population or market it observes, the time horizon, the key assumptions, and the reason it receives the weight it receives. If two methods disagree, do not average first and explain later. Identify the condition that makes them disagree.

Thomas illustrates the approach with a new electronic mail service. That application is valuable as a method example, not as a modern benchmark. Its result does not establish a universal weighting rule. The useful transfer is the discipline of defining the object before reconciling its estimates.

How does category novelty alter the variables required for product forecasting?

Really-new products can ask customers to learn a category, imagine an unfamiliar use, or evaluate infrastructure that does not yet exist. Urban, Weinberg, and Hauser combine Information Acceleration with qualitative research, engineering information, and managerial judgement in a General Motors electric-vehicle application. The research environment tries to let respondents react to a more concrete future than a bare product description.

That does not turn the application into a general accuracy guarantee. It gives the team a design question: which missing experience, constraint, or learning step could make the premarket estimate misleading? Add a method only when it observes that missing condition or tests a rival assumption.

Why does diffusion timing govern commercial adoption rates?

A static market potential can be large while early sales remain small. Oren and Rothkopf model the path between preference and realized sales through awareness, availability, production, sales-force size, purchasing delay, and other changing market states. Adoption is not a switch from zero to one.

This matters for a launch plan. A forecast that contains preference but not availability may support an opportunity story while failing to tell operations when stock, coverage, or service must be ready. A dynamic market model can make the timing visible, but it cannot supply the local parameters by itself. The high-speed nonimpact printer application is a model and case, not a parameter library for every new product.

How can commercial forecasts display parameter uncertainty without losing utility?

Aydin, Kwong, Ji, and Law combine fuzzy regression, discrete-choice analysis, and fuzzy market-potential estimation in a tablet-PC application. Their worst, normal, and best scenarios keep uncertainty in the inputs and choice probabilities instead of compressing every assumption into a single point.

That is a useful representation choice. It is not a solution to uncertainty. The case is one product application, and fuzzy inputs do not remove sampling, model, or market assumptions. Use a market scenario to show what changes across states, then name which observation would move the team from one state to another.

How does product architecture determine appropriate forecasting investment?

Klinger argues that the work required to measure industrial market potential depends on product type, age, and use. An industrial component, a new category, and a mature replacement do not expose the same information problem. The eight-step approach and product typology are a reminder to scope the research effort to the object.

The implication is not that an old checklist becomes correct because it has eight steps. It is that a method plan should explain why its chosen evidence can observe this product, this buyer, and this stage of adoption.

How should leadership reconcile divergent forecasts before capital commitment?

Use the matrix in sequence:

1. Name the decision and the forecast object: market boundary, purchase potential, adoption path, or investment outcome.
2. Select methods that observe different conditions, and record the blind spot of each.
3. Put every method on the same product definition, population, unit, time horizon, and scenario frame.
4. Reconcile disagreements in writing. State the weight, the reason, and the condition that would change it.
5. Attach an observation to every major uncertainty. A scenario is useful when the team knows what it would learn next.
6. Separate a launch commitment from a research commitment. The first spends on execution; the second buys information.

This turns method count into method purpose. One method may be enough for a narrow decision. Several may be justified when the product is novel and each one covers a different blind spot. The test is not whether the final number looks precise. The test is whether the decision can show what the number saw and what it still assumed.

Why is pre-launch forecasting distinct from ongoing forecast value added?

Premarket forecasting constructs a view before a product has a settled sales history. Forecast value added evaluates a recorded baseline, intervention, and actual outcome after the relevant process has produced an observation. The forecast value added audit asks whether a change improved a defined forecast process. This piece asks how to build and reconcile a forecast when the outcome is not yet available.

The market allocation ledger supplies the adjacent boundary: market opportunity, firm reach, resources, and return are separate objects before the methods are reconciled. In parallel, empirical reviews demonstrate that new-product success factors are context-dependent, meaning that drivers of launch performance differ sharply across industries and development stages.

The two practices can meet later. A premarket method matrix can preserve the assumptions that a later outcome audit will test. It cannot borrow the later audit’s evidence before launch.

Where are the analytical boundaries of new product forecasting models?

This piece does not claim that more methods always improve accuracy. It does not provide a current electric-vehicle or tablet-PC estimate, a universal weighting rule, or transferable dynamic parameters. It uses five full-text studies to make the premarket forecast object, blind spots, and reconciliation logic visible before a launch budget becomes a commitment.