10.6 New-Product Forecasting: Analogs, Attributes & Diffusion Models

New-Product Forecasting: Analogs, Attributes & Diffusion Models

Every statistical forecasting method requires history. A new product has none. This is the introduction stage of the life cycle, where forecast error is at its worst and where the supply consequences — capacity commitments, tooling, long-lead material, launch inventory — are largest and least reversible. ISM's Product and Service section tests this directly.

---

Analog (Like-Item) Forecasting

The most widely used practical method: find a comparable prior launch and scale its demand curve.

Method:

1. Select analogs. Choose prior launches similar in category, price tier, channel, target customer, and seasonality. Two or three analogs beat one.
2. Extract the shape. Take the analog's demand by week or month from launch, indexed to its own peak or its first-year total. What you are borrowing is the shape of the ramp, not the absolute volume.
3. Scale the level. Adjust for the differences that matter:

1. Produce a range, not a point. Low, base, and high scenarios with explicit assumptions. A single-point new-product forecast is false precision and drives exactly one supply plan, which is the wrong response to this level of uncertainty.
2. Replace it fast. Set a defined checkpoint — often four to eight weeks of actual sell-through — at which the analog forecast is discarded and a data-driven forecast takes over.

---

Attribute-Based Forecasting

Where no single good analog exists, decompose the new item into attributes whose demand effects are known from the existing portfolio: size, colour, material, feature set, price tier, package configuration. Statistical relationships from the current range estimate the demand contribution of each attribute, and the new item's forecast is assembled from its attribute profile. This works well in categories with wide, structured variety — apparel, consumer electronics accessories, industrial fasteners, packaged goods — where the portfolio itself is the training data.

---

The Bass Diffusion Model

The Bass model describes how a new product spreads through a population by splitting adopters into two groups:

• Innovators — adopt because of external influence such as advertising and availability. Governed by the coefficient of innovation, $p$.
• Imitators — adopt because of internal influence: word of mouth from those who already own it. Governed by the coefficient of imitation, $q$.

$S(t) = m\left[\frac{(p+q)^2}{p} \cdot \frac{e^{-(p+q)t}}{\left(1 + \frac{q}{p}e^{-(p+q)t}\right)^2}\right]$

where $m$ is the total market potential (eventual cumulative adopters), $p$ the innovation coefficient, and $q$ the imitation coefficient.

You will not be asked to solve this on an exam. What is testable is the behaviour it predicts and the parameter meaning:

The resulting adoption curve is the familiar bell shape, with cumulative adoption following an S-curve. Parameters are estimated by analogy to prior product diffusions in the same category, which is why the Bass model complements rather than replaces analog forecasting.

The supply implication that ISM cares about: a high-$q$ product has a slow start and a sharp later peak. Building capacity to early-weeks demand guarantees a stockout at the peak, and reading the slow start as failure and cancelling capacity is the classic error. Conversely, a high-$p$, low-$q$ product peaks early and declines, so late capacity additions arrive after the peak has passed.

---

Pipeline Fill vs. Ongoing Consumption

This distinction causes more launch misreadings than any other.

• Pipeline fill (channel loading) is the one-time volume required to stock the distribution network — distributor warehouses, retail shelves, display units, service stock. It is a function of the number of stocking points times the presentation quantity, not of consumer demand.
• Ongoing consumption is the recurring sell-through rate.

Worked illustration. A product launches into 1,200 retail locations, each holding 24 units of shelf presentation, plus 30,000 units of distributor inventory.

• Pipeline fill = $(1{,}200 \times 24) + 30{,}000 = 28{,}800 + 30{,}000 = \mathbf{58{,}800\ \text{units}}$, shipped in the first weeks.
• If steady-state consumption is 9,000 units per month, month-one shipments look enormous and month-three shipments look like a collapse — even if consumption is exactly on plan.

Exam trap: interpreting the post-pipeline-fill shipment drop as a demand failure and cutting supplier orders. The correct diagnosis separates one-time fill from recurring consumption, and the correct signal to watch is sell-through, not shipments.

---

Cannibalization and Incremental Demand

A new product rarely creates demand purely out of nothing. Some of its volume is transferred from existing products in the same portfolio.

$\text{Incremental demand} = \text{New product demand} - \text{Cannibalized demand from existing products}$

Cannibalization must be forecast explicitly for two reasons: the capacity and material plan for the existing items must come down as the new item ramps, or total inventory balloons; and the financial business case must be built on incremental rather than gross volume. A launch that adds 40,000 units while removing 30,000 units from an existing item delivers 10,000 units of incremental demand, and any supply plan built on 40,000 units of net new volume will strand inventory on the older item.

---

Making a Wrong Forecast Survivable

Because launch forecast error is irreducibly high, the professional response is supply-side hedging, not forecasting heroics:

1. Scenario-based capacity. Contract base capacity to the low scenario and secure options on incremental capacity, paying an option fee rather than committing full volume.
2. Flexible supplier agreements. Upside and downside tolerance bands (for example, plus 50% and minus 30% against the plan) with defined notice periods, negotiated before launch.
3. Postponement. Commit to generic or semi-finished inventory whose aggregate forecast is far more reliable than the variant-level split, and configure late.
4. Staged launch. Roll out by region or channel, using early markets as a live test that recalibrates the forecast before full commitment.
5. Short-lead-time sourcing for the launch phase. Accept a higher unit cost from a responsive supplier during introduction and shift to a lower-cost efficient supplier at maturity — the deliberate agile-to-lean transition of the life cycle.
6. Pre-agreed exit terms. Cancellation windows, raw material liability caps, and tooling amortization treatment settled at contract signature, so a failed launch does not become an open-ended liability negotiation.