When a new SKU eats an existing one

A launch report that shows the new SKU doing eighteen thousand units a month is not evidence of anything. The question is where those units came from. If nine thousand of them used to be sold by the SKU sitting next to it, you have run a large operational project to move volume sideways, and you have added a code, a forecast, a stock position and a slot on every dark store shelf to do it.

This is a measurement problem first. Get the measurement right and the decision that follows is usually obvious.

The mistake is judging a launch on its own sales

Two things hide substitution. Categories in India are often growing anyway, so a segment that grows six percent while a new SKU adds twelve percent of segment volume still looks fine on a total line. And launch periods are noisy in the new SKU’s favour: trial, launch media, new-listing visibility and a hero banner slot all inflate the first six weeks, and none of it is a repeatable rate.

The unit to read is not the new SKU’s sales. It is the segment’s volume, and inside it, the volume of the specific SKUs a buyer would otherwise have chosen.

What to compare against what

Total business against last year is the wrong frame, because too much else changed. Build the comparison so that only the launch differs.

• Same channel. Do not blend quick commerce, marketplace and D2C. A launch that lands on Blinkit and Instamart first will show a national marketplace number that has nothing to do with the effect you are looking for.
• Same city set. If the rollout was phased, you already have a hold-out. Compare the donor SKU’s units in launch cities against the same SKU’s units in cities that have not received the new one yet, both indexed to their own pre-launch four weeks. If the donor falls eleven percent in launch cities and one percent in hold-out cities, you have measured the effect, not guessed at it.
• Same store where you can get it. A store-level view on quick commerce removes the biggest confound in the exercise, which is that the platform added dark stores during your launch window. Total platform volume can rise while every store you were already in went backwards.
• Buyer level where you own the data. On your own site the cleanest read is overlap. Of the buyers of the new SKU, how many bought the old one in the previous ninety days? That figure is the source of volume, stated plainly.

Index the donor against its own seasonal position rather than a flat prior month, or a launch that happened to land in a category ramp will read as expansion. That indexing is the job of a category demand calendar.

Good cannibalisation and destructive cannibalisation

Substitution inside your own range is not automatically a loss. Split it on the buyer, not the unit.

It is fine, and often the point, when the buyer moves up. A shopper who bought a 200ml pack twice a month and now buys a 500ml pack once has cut your unit count and raised your margin per buyer per month. A trial pack that converts to a full size has done its job even though the trial line drops. A premium variant taking volume off a mainstream one has upgraded the buyer.

It is destructive when volume moves to a worse position. Same buyer, same frequency, lower contribution per unit. That happens when the new SKU carries a higher trade margin, a heavier platform commission slab, a worse weight-to-value ratio for shipping, or a launch price that never came back up. The comparison to run is contribution per buyer per month before and after, not gross revenue, and the per-SKU version of that arithmetic sits in profitability per SKU and contribution margin.

Where it shows up most: pack size and price point

Two conditions produce most of the substitution you will ever measure.

• Adjacent pack sizes on the same shelf. A 400g introduced between an existing 250g and 1kg takes from both, and on a dark store shelf where all three appear in one scroll the effect is larger than in general trade. How the ladder gets designed is covered in pack size strategy and price pack architecture.
• Price per unit within roughly fifteen to twenty percent. When two of your SKUs land that close on price per gram, the shopper reads them as the same decision with a different label. Test how sensitive that gap is rather than assume, using price elasticity testing.

Flavour and variant extensions cannibalise more predictably than format changes. A new flavour in an existing pack size almost always splits an existing buyer base. A genuinely new format or occasion is the only reliable way to recruit.

What you do with the answer

The response to confirmed cannibalisation is a range or pricing decision. It is not a marketing decision, and spending more media on the new SKU when the segment is flat only buys the same volume twice. The three decisions that follow are widening the price gap so the two SKUs stop reading as substitutes, redesigning the ladder so each step has a distinct job, or changing what the range holds. Those decisions are documented separately in range architecture and SKU rationalisation. This post ends at the measurement.

Predict it before launch, then check

Before a launch goes into a forecast, write one sentence: of the first three months of volume, X percent will come from buyers switching off SKU Y and the rest will be new to the brand. Anyone can produce that estimate, and the estimate being wrong is not the failure. Not having written it down is, because a source-of-volume number agreed after the fact is never disagreed with, and the forecast quietly double-counts the same demand in two lines.

Then check at week eight and again at week twelve, using the hold-out comparison rather than the total line. The same discipline belongs in the quick commerce new launch playbook, alongside the conversion mechanics in trial pack sampling. Rule out one thing before you call a donor’s decline cannibalisation: a SKU that was out of stock for nine days did not lose to your new one, and that case is out-of-stock substitution.