New products & phase-outs

A new product has no sales history, so time-series methods have nothing to work with. Planners build the forecast from other evidence instead:

• Analogs: a similar past launch, such as last year’s flavour variant, gives a starting launch curve (slow start, growth, then a plateau).
• Adjustments: scale that curve for what’s different this time, such as a stronger brand, wider distribution, a higher price or a bigger marketing push.
• Drivers and judgment: market size, distribution plan, pre-orders and sales-team input.
• Scenarios: conservative, expected and aggressive versions, because uncertainty is highest at launch.

Then the most important step: reforecast quickly. The first few weeks of real sales are worth more than any analog, so the plan should update fast and supply should stay flexible until the curve is clear. A classic model of how adoption spreads is the Bass diffusion model, which separates innovators who buy early from imitators who follow them.

The other end of the lifecycle needs the same care. When a product is phased out, demand winds down while a replacement ramps up. Good phase-out planning:

• links the old item to its successor, so the successor can inherit demand history,
• stops buying components early enough to avoid dead stock,
• times the last production run against remaining stock and the launch date of the new item.

Cannibalization ties the two ends together: the new product’s success often comes partly from the old one’s sales.

Analog forecasting step by step

1. Choose analogs: past launches with a similar category, price tier, channels and season.
2. Align and normalize: line them up by weeks since launch and express sales per store (or as a share of their first 12 weeks), so launches of different sizes can be compared.
3. Build the curve: average the normalized analog curves.
4. Scale it: multiply by the new product’s distribution (stores or reach), and adjust for price, marketing support and brand strength.
5. Add scenarios: for example, the expected curve plus versions 30% lower and 30% higher.

Weighting matters: one week of data is noisy, so blend the analog curve with the actuals, giving the actuals more weight as the weeks pass.

The Bass diffusion model

For products people adopt once, such as durables and technology, the Bass model describes how adoption spreads. New adopters in a period are:

Here $m$ is the market potential, $N_{t-1}$ the adopters so far, $p$ the innovation coefficient (people who adopt on their own) and $q$ the imitation coefficient (people who follow others). Sales peak at about $t^{*} = \ln(q/p)/(p+q)$. With $p$ = 0.03 and $q$ = 0.38, close to averages reported across many published studies, the peak comes after about 6 periods. With no history of its own, a product borrows $p$ and $q$ from similar categories.

Run-out planning at end of life

For the product being replaced:

1. Agree the switch date with marketing and sales.
2. Forecast the remaining demand to that date, which usually declines once the replacement is announced.
3. Stop ordering components and materials once stock plus open orders cover that demand, allowing for their lead times.
4. Plan the last production run and decide what happens to leftovers: rework, sell-off or scrap.

Sell-in is not demand

At launch, the first shipments fill the pipeline: every store, distributor and warehouse needs opening stock. Those sell-in volumes can equal several weeks of consumer demand, and they happen once. A forecast built on early shipments, rather than on sell-through to consumers, overestimates ongoing demand, and the expected second wave of orders never comes. Track consumer sales or distributor sell-out from day one.

Learning fast from early sales

For short-lifecycle products, early sales predict the season remarkably well. In fashion retail, Fisher and Raman (1996) showed that producing part of the season’s volume first, observing early sales and then committing the rest (quick response) sharply reduces the cost of demand uncertainty. The same idea applies to any launch: commit part of the volume up front, keep capacity or supplier flexibility for the rest, and reforecast every week.

Managing product transitions

When a new model replaces an old one, the switch can be:

• a solo roll, a hard cutover on one date: simple, but risky if the new model is late or the old one has leftovers, or
• a dual roll, where both are sold for a while: safer, but it needs more stock and creates cannibalization to forecast.

Billington, Lee and Tang (1998) describe how the right choice depends on how risky the new product’s launch is and how different the two products are. Planning systems support transitions by linking the old item to its successor, so the new one inherits demand history (scaled as needed) while the old one’s forecast phases out along a profile.

Supply strategy for launches

• Smaller first production runs with fast replenishment.
• Postponement, so generic stock can become whichever variant sells.
• A separate launch buffer, released as the curve becomes clear.
• Pre-agreed decision points: dates and sales thresholds for scaling up, holding or cutting the plan.