How to Calculate Cannibalization Rate: Formulas and Examples

Cannibalization rate is the share of a new unit's sales that came out of something you already own. The formula is (sales lost by the existing unit / sales of the new unit) x 100. If an existing product loses 65 units a month after a launch and the new product sells 250, the rate is 26%.

Two decisions do more damage to that number than the arithmetic ever will: which base you divide by, and how long you wait before you measure. Both are covered below, along with the store version of the math most write-ups skip.

Here is the quick-reference version for a product launch:

I'm Clyde Christian Anderson, Founder and CEO of GrowthFactor.ai, where my team and I have helped retailers evaluate sites at scale. Cannibalization math is a big part of that work, because nobody wants to sign a ten-year lease and then find out they moved sales from one store to another.

How to calculate cannibalization rate for a new store

For a store opening, the rate is transferred sales divided by the new store's sales. Getting the numerator right takes one extra step that product math doesn't need: you have to remove whatever the existing store would have lost anyway, or you'll book a soft quarter as cannibalization.

Work it through with real shape to it. Store A's averaged $400,000 a month for the twelve months before anything changes. You open Store B two and a half miles away, and it settles at $200,000 a month. Store A settles at $348,000.

The raw drop is $52,000. That's not your numerator. Over the same window, a matched set of your stores far enough away that Store B can't reach them fell 3%, which on a $400,000 base is $12,000 of decline that had nothing to do with the new location. Subtract it, and $40,000 actually transferred.

Store cannibalization rate = ($40,000 / $200,000) x 100 = 20%

That matched set is a control group, and building one is the step teams skip most often. The five-step method for running the analysis covers how to pick control stores and what breaks when you don't. Comparable-store metrics are the same instinct applied to a whole fleet, so same-store sales get read alongside a cannibalization estimate rather than instead of one.

Which base you divide by changes the answer

Both of these sentences describe the identical opening, and both are correct. "The new store cannibalized 20%." "The existing store lost 10%." Nothing about the stores changed between them. The denominator did.

Dividing by the new unit's sales answers "how much of this opening is not new business," and that's the question a growth plan asks. Dividing by the existing unit's prior base answers "how much did the neighbor give up," and that's the question the person running that store asks. Both belong in the packet. Neither is the cannibalization rate on its own.

Write the convention down before anyone sees a number, and use the same one on every site. A 15% from one convention read against a 25% from the other is how a portfolio review turns into an argument about arithmetic.

When to measure, and why the number keeps moving

A cannibalization rate calculated in month one is mostly measuring the opening. Grand-opening promotion and curiosity traffic pull hard on the nearby store, then customers sort themselves out and settle into whichever location actually suits them.

The number stops being a moving target when it flattens. Until it does, you have a trend, not a result. Hold the first serious read until at least two quarters of post-opening sales are in, and treat anything computed before then as directional. Teams that lock a number at 90 days and never revisit it end up defending a figure the data has since walked away from. GrowthFactor's own advice is to backtest the forecast against the realized rate at four quarters, because the size of that gap tells you how your model reads this kind of market.

The two product formulas

Product launches are the easier case, because the decision's reversible and the data arrives in weeks rather than quarters.

Using sales volume

This is the version inventory and supply chain teams use, and it tells you how many units of the old product got displaced.

(Old product unit loss / new product unit sales) x 100

Say Product A sold 1,000 units a month before the launch and sells 800 now. That's a loss of 200 units. Product B, the new one, sells 500. Divide 200 by 500 and multiply by 100 for a 40% rate. Read plainly: 40% of the new product's success came from people who would've bought the old one anyway.

Using revenue

Finance asks in dollars, and dollars can flip the verdict. If the new product carries a higher price, a scary-looking unit rate can still be a good trade.

(Revenue loss on old product / total revenue of new product) x 100

A $10,000 revenue loss on the old line against $50,000 of new product revenue is 20%. Whether that's a win depends on margin, not on the percentage.

Don't mix the two inside one calculation. Units in the numerator and dollars in the denominator produces a number that looks like a percentage and means nothing.

Forecasting the rate before you open

Before a store exists there's no sales dip to measure, so the estimate has to come from trade area overlap instead. The standard approach is a gravity model, which comes from David Huff's 1963 paper A Probabilistic Analysis of Shopping Center Trade Areas in Land Economics. Sixty years on, it's still the backbone of how the industry forecasts overlap.

The idea is that a household's choice between competing locations is a probability, driven by how attractive each one is and how far away it is:

P(i to j) = (Aj^α ÷ Dij^β) ÷ Σ over all locations k of (Ak^α ÷ Dik^β)

Aj is the attractiveness of location j, usually selling area or another size proxy. Dij is the distance or drive time from origin i. The exponent α scales how much size matters and β scales how fast interest decays with distance, and both get calibrated to the category. A convenience run and a weekend furniture trip decay at very different rates. The reference version of the notation is worth a look if you're implementing it, and researchers have calibrated Huff models against large-scale location data to check that they still hold up against how people actually move.

The forecast falls out of running it twice. Compute each existing store's share of the trade area without the new location, compute it again with the new location included, and the drop is your predicted transfer. Divide by the new store's forecast sales and you have a rate before anyone signs anything. This is also where trade area overlap can be reduced by sequencing rather than accepted as fixed.

Whatever you feed it, the boundary matters more than the exponents. A radius drawn around a pin will happily reach across a river with no bridge and count households that can't drive to you, and every overlap number downstream inherits that mistake.

What the published benchmark ranges are actually worth

Search for a normal cannibalization rate and you'll find the same four bands everywhere: 0-10% is great, 10-20% is acceptable, 20-30% is caution, 30% and up is high. They appear on calculator sites, accounting glossaries, and consulting explainers, all with the same numbers and none with a citation.

We went looking for the primary source and didn't find one. No trade association publishes those thresholds. No peer-reviewed study establishes them. The academic work that does exist, like Pancras, Sriram and Kumar's study of retail expansion and cannibalization in Management Science, models how transfer behaves in a specific chain rather than handing down a universal cutoff. Meanwhile retail vendors circulate ranges that contradict both the generic bands and each other, with quick-service tolerances quoted anywhere from 25% to 35% and specialty retail from 15% to 20%. Numbers that disagree by a factor of two aren't a benchmark. They're a guess that got repeated until it sounded official, and some of that repetition traces back to earlier versions of this page.

That doesn't make the rate useless. It makes the percentage the wrong place to stop. The question a real estate committee is actually asking is whether the new unit adds profit, and that math doesn't need an industry threshold:

Net contribution = (net-new sales x contribution margin) − (transferred sales x contribution margin at the existing unit)

Run that and a 25% rate on a site pulling heavy competitor share can clear easily, while a 12% rate on a site that adds nothing but a shorter line at your existing store doesn't. The rate ranks the risk. Margin decides the deal. If you want somewhere to start, our thresholds for acceptable overlap are labelled as what they are: a first cut you adjust to your own margins, not a standard.

One more figure worth retiring while we're here. The claim that 95% of new consumer products fail gets attributed to Clayton Christensen everywhere. The Harvard Business School article usually cited for it, Carmen Nobel's February 2011 write-up of his milkshake research, contains no failure percentage at all. What it says is that "each year 30,000 new consumer products are launched and many of them fail." The page now opens with an editor's note reading "Updated to clarify a failure rate figure included in an earlier version," which is about as close to a retraction as this kind of stat ever gets. We couldn't find a primary source for 95% anywhere, so it's not in this article any more either.

What the data has to look like

Clean inputs matter more than formula choice, because every version of this math is a subtraction and subtraction is unforgiving.

You need at least twelve months of sales history before the change, so seasonality is visible rather than baked into the answer. You need a control set of units the new one can't reach. For stores you need customer origin data, because if half the new store's customers live in the existing store's core trade area, the overlap is already telling you the answer. And you need loyalty or basket data if you want to know whether the same individual switched, which is the difference between transfer and a coincidence.

Anything missing from that list doesn't stop you from producing a number. It stops you from defending it. That distinction is the whole reason internal cannibalization is easier to argue about than to measure, and why market saturation work belongs upstream of the site-by-site question.

When we ran the Party City auction for Books-A-Million, the #2 book retailer in the US, the overlap math was the filter, not the afterthought. Roughly 700 sites went through scoring and revenue forecasts in 72 hours, and BAM entered two new markets with zero cannibalization while passing on 15 sites that did not clear. Same formula everyone else has. The difference was having the control set and the trade area data ready before the bidding started.

Frequently Asked Questions about Cannibalization

Putting the number to work

The arithmetic here takes a minute. Naming the denominator, waiting for the rate to settle, and holding a control set are what turn it into something you can put in front of a committee.

At GrowthFactor we pull demographics, foot traffic, competition, zoning, and drive-time analysis into one place, and you can open any score and see which inputs moved it. No black box, no separate spreadsheet to reconcile against it. Cavender's evaluated 2,000+ sites that way and avoided three poor locations worth about $2M.

If you want the overlap math running before the lease rather than after, see how retail expansion teams use GrowthFactor.