Demand Cannibalization When You Launch a New SKU

A new flavour rarely creates demand. It moves it. How to forecast a launch that takes sales from a product you already sell, how to measure how much it took, and what to do about the buy on both sides.

7 min read

A launch forecast is usually built as though the new SKU arrives into empty space. It rarely does. Most consumable CPG launches are a fourth flavour, a smaller size or a new format aimed at people already buying from the range — and if the launch does well, some of that success arrived from a shelf you already own.

What is demand cannibalization?

Demand cannibalization is when a new product takes sales from one you already sell instead of adding sales you would not otherwise have made. It is the difference between the new item's units and the change in the category's units, and only the second of those two numbers is money.

What it looks likeWhat it does to the buy
Fully incrementalCategory grows by the new SKU's whole volumeBuy the new SKU to forecast, leave the rest alone
Partly cannibalizingCategory grows by less than the new SKU sellsBuy the new SKU, cut the donor SKUs by the moved portion
Fully cannibalizingCategory flat, mix has shiftedTreat it as a replacement and plan the old SKU down to exit
Cannibalizing across channelsOne channel grows, another fallsThe plan changes per channel, not in total

That last row catches brands out most often, because the two halves are usually measured by different people. A wholesale listing that quietly moves subscription volume looks like growth in one report and churn in another.

Why do CPG launches cannibalize?

Because substitution is the normal shopper behaviour in a consumable category, not an unusual one:

  • Fixed consumption. Someone drinks a certain amount of the beverage a week. A fourth flavour changes which one they drink, not how much.
  • Fixed shelf space. In retail, a new facing usually comes out of an existing facing, and the category's total space is the constraint the buyer is managing.
  • Fixed budget per trip. A basket has a ceiling. A new size at a higher price often replaces the old size rather than joining it.
  • Your own promotions. Launch support pulls attention from the rest of the range, so the donor SKU's decline can be caused by the launch marketing rather than by the product.

None of that makes a launch a bad idea. It makes an incremental forecast a bad assumption.

How do you forecast a launch that moves demand?

There is no statistical model for this, because the new SKU has no history and the donor SKUs have history from a world where it did not exist. What works is an explicit, written split that can be graded afterwards.

1. Name the substitute group

List the existing SKUs a shopper would treat as alternatives: same shelf, same format, similar price, same need. Be strict — a group of three real substitutes is more useful than a group of twelve that includes everything you sell.

2. Forecast the new SKU by analogy

Use the most similar launch you have already run, adjusted for the support behind this one. Forecasting demand for new products covers this in full; the short version is that an analogous product's first weeks beat any model fitted to no data.

3. Split the forecast in two, on the record

Divide the new SKU's forecast into a portion that is genuinely new demand and a portion moved from the group, and write down which SKUs it moved from. A rough split you can check beats a precise number you cannot. If nobody will commit to a split, that itself is the finding — the launch case is resting on the new SKU being incremental without anyone having said so.

4. Reduce the donor forecasts by the moved portion

The step that gets skipped, and the one that costs cash. If the new SKU takes sales from two existing products, those two products' forecasts have to come down by the same units, on the same dates, in the same channels.

5. Hold the group's total, then sanity-check it

Add the group up. If the group total after the launch is far above its pre-launch trend, either the launch really is opening new demand — which needs a reason you can name, like a new channel or a new occasion — or the split is optimistic. What demand forecasting is covers laying explicit adjustments over a statistical baseline, which is exactly the shape this is.

6. Plan the buy from both sides

The new SKU needs a deliberately small first order and a fast reorder path. The donor SKUs need their next purchase order cut, and if their lead time is long, that cut has to be made before the launch, not after the first weeks of data arrive.

How do you measure cannibalization after launch?

Compare the group against its own pre-launch forecast, not the new item against nothing.

  • Group actuals against group forecast. Take the substitute group's units for the launch weeks and compare them with what the group was forecast to sell before the new SKU existed. The shortfall against the new SKU's own sales is the moved portion.
  • Same channels, same weeks. A launch that ships DTC first and retail later cannot be measured in total until both are live.
  • Watch for the borrowed-forward effect. Launch promotions pull demand forward from the weeks after the launch, so the period immediately following often dips on both the new SKU and the donors. Adjusting a forecast for promotions covers separating lift from pull-forward.
  • Re-measure the variability. Both the new SKU and the donors are more volatile than their history says while the group settles, which is a safety stock question rather than a forecasting one.

Do this at a fixed point after launch — the end of the first full replenishment cycle is a defensible one — and write the answer next to the split you committed to before launch. Two or three launches graded that way and your pre-launch splits stop being guesses.

Which launches cannibalize most?

Launch typeExpected substitutionWhy
New flavour in an existing lineHighSame occasion, same shopper, same shelf
New size of an existing productHighUsually a swap rather than an addition
New format of an existing productMediumCan reach a new occasion, often does not
Same product, new channelLow to mediumNew buyers, but existing buyers may switch where they buy
Genuinely new category for the brandLowDifferent need, different shelf, different budget

The pattern is simple: the closer the new item sits to something you already sell, the more of its sales came from there. A launch that cannibalizes heavily can still be right — trading a low-margin SKU for a higher-margin one, or replacing an ageing item before a competitor does — as long as the plan says so in advance and the buy on both sides reflects it.

Where Planster fits, and where it doesn't

Planster builds one demand forecast across DTC, Amazon, retail and wholesale, and nets it against what is on hand, on order and already committed to produce a reorder point, an order quantity and an order-by date per SKU. That is the engine. Because the plan is one forecast across channels rather than a set per channel, a launch that moves volume from one channel to another shows up as a shift in the same plan instead of as growth in one report and a decline in another.

For the launch decision itself, Master Plan prices a scenario before you commit to it — the bigger first order, the earlier cut on the donor SKU — and the overnight run brings you a ranked list each morning with the purchase orders already drafted. You open Planster, change what you want changed, and approve. Nothing reaches a supplier until you do. Flat $1,000/month, for consumable CPG brands between $10M and $50M.

Where it does not help: Planster does not decide what share of a launch is incremental. No system can, because the answer does not exist in your sales history — it exists in a judgement about shopper behaviour that you make and then check. What the plan can do is hold that judgement explicitly, apply it to both sides of the launch, and show you the variance afterwards. See promotional planning for how launch and promotion adjustments sit on top of the baseline, and pricing for the whole number on one page.

Common questions

What is demand cannibalization?

Demand cannibalization is what happens when a new product takes sales from a product the same company already sells, rather than adding sales the company would not otherwise have made. In consumable CPG it is most visible at flavour, size and format launches, where the buyer was already going to purchase something from the range and simply chooses the new option. The total for the category moves far less than the new item's sales suggest, which is why a launch can look successful and leave the business flat.

How do you forecast cannibalization for a new product?

Forecasting cannibalization starts by naming which existing SKUs the new one will take from, before the launch rather than after. Pick the products a shopper would treat as substitutes — same shelf, same format, similar price — and split the new item's forecast explicitly into a portion that is genuinely new demand and a portion moved from those SKUs. Reduce the donor SKUs' forecasts by the moved portion. Writing the split down is what makes it checkable later, even when the first estimate is rough.

Is cannibalization always bad?

Cannibalization is not automatically bad, and treating it as a failure leads to worse decisions than accounting for it does. Moving demand from a low-margin SKU to a higher-margin one improves the business at flat volume. Replacing an ageing item before a competitor does it for you protects shelf space. What causes damage is unplanned cannibalization: the new SKU forecast as fully incremental, the donor SKU bought at its old rate, and cash tied up in inventory on both sides of a launch that only ever moved sales around.

How do you measure cannibalization after a launch?

Measure cannibalization at the level of the group rather than the item. Compare the whole substitute group's actual units against what the group was forecast to sell before the launch existed, over the same weeks and the same channels. The gap between the new SKU's sales and the group's increase is the portion that moved. Run it at the group level because item-level comparisons cannot separate a donor SKU declining from a donor SKU being replaced, and the buying decision depends on which one happened.

Does cannibalization affect safety stock?

Cannibalization affects safety stock on both sides of a launch, in opposite directions. The new SKU has no history, so its demand variability is unknown and its buffer is a judgement call rather than a calculation. The donor SKUs become more volatile than their own history suggests, because demand is being redistributed while shoppers try the new option, so a buffer sized on pre-launch variability will be wrong for a period. Re-measure both once the group has settled.