Cannibalization Analysis: Protect Retail Revenue 2026

Retail cannibalization analysis measures how much of a new store's or new product's sales are transferred from something you already own instead of added to the business. It turns a vague worry about overlap into a number you can weigh against incremental profit, before the lease is signed or the SKU ships.

What retail cannibalization analysis measures

The analysis has four moving parts, and they apply whether the new unit is a storefront or a shelf item.

• Product cannibalization: a new item pulls demand from an existing one.
• Location cannibalization: a new store draws customers from a store you already run.
• Cannibalization rate: the sales an existing unit lost, divided by the sales of the new unit or by the existing unit's prior base, expressed as a percentage.
• Net impact: incremental sales minus cannibalized sales. This is the only number that tells you whether the business grew or just reshuffled.

The distinction that matters is not whether cannibalization happened, because it almost always does. The question is whether it was a trade you made on purpose. A brand that retires an aging product for a better one chose the hit. A brand that signs a lease three miles from its own store because the corner looked strong did not. If you want a plain-language primer on the concept before the mechanics, start with our guide to internal cannibalization.

The stakes are concrete in this market. Coresight Research projects roughly 7,900 US store closures against about 5,500 openings in 2026, a net loss of some 2,400 locations. Growing brands are picking sites in markets where somebody just handed one back, and every one of those calls carries an overlap question.

I'm Clyde Christian Anderson. At GrowthFactor I've worked alongside retailers evaluating more than 2,000 potential locations, and the cannibalization question comes up on nearly every one. My background runs from retail operations to building the platform that puts overlap on the table before a lease gets signed.

How to run a cannibalization analysis in five steps

A cannibalization analysis runs the same five steps whether the unit is a store or a SKU: define what you are measuring, build a control set, estimate transfer against that control, convert transfer into profit, and fix the decision rule before you see the result. Most teams skip the second one, which is where the analysis quietly stops being evidence.

1. Define the unit, the window, and the affected set

Name three things before you pull any data. The unit is the new store or the new item. The window is the comparison period, long enough to clear the opening bump and short enough to stay inside one season, commonly 8 to 13 weeks after launch measured against the same span a year earlier. The affected set is the specific existing stores or items you expect to take the hit.

Choosing the affected set in advance is what keeps the analysis honest. If you pick the affected stores after looking at which ones declined, you have found a narrative, not an effect. Write the list down, date it, and hold yourself to it.

2. Build a control set, because before-and-after lies

A store's sales move for a dozen reasons in any 90-day window: weather, a road closure, a competitor opening, a category-wide slowdown. Compare the affected store's sales before and after your new opening and call the gap cannibalization, and you have credited your new store with every one of those effects.

The fix is a control set. Pick stores or items that resemble the affected ones on the variables that matter (format, volume band, trade-area profile, seasonality) but sit far enough away that the new unit cannot reach them. Measure the change in the affected set, measure the change in the control set over the same window, and take the difference. That difference is your cannibalization estimate. Analysts call the design difference-in-differences, and it is the same logic behind a matched-market test. When no single unit is a clean match, a synthetic control built as a weighted blend of several donor units does the same job. PyMC Labs walks through the Bayesian version, which returns a credible range for the cannibalization share instead of one brave number.

Two practical rules. Use several control units per affected unit where you have them, because a single control carries its own noise. And freeze the control set before the launch, for exactly the same reason you froze the affected set.

3. Estimate transfer, not correlation

A control set tells you that demand moved. It does not tell you that it moved to your new unit. That is a separate measurement, and the method splits by unit type.

For stores, the evidence is customer origin. Map where the affected store's customers actually live or commute from, project the new site's draw, and calculate what share of the existing store's origin volume falls inside it. Loyalty records, transaction ZIP codes, and foot traffic patterns all work as inputs. What does not work is a distance ring, because two stores five miles apart can fish the same commuter pool while two stores a mile apart serve entirely different crowds.

Then weight it. Two trade areas can share a thin slice of area and a thick slice of demand, because customers are not spread evenly across a map. CARTO's cannibalization tutorial measures overlap as the share of an existing store's catchment population that also falls inside the new site's catchment, rather than the share of map area the two catchments happen to share. Lead with the demand-weighted figure, and label which convention the slide is using.

For products, the evidence is the basket. The volume is what makes this worth systematizing: NielsenIQ counted more than 3,500 new brands and sub-brands launched across five European markets in 2025, and only about a third reached 1 percent household penetration. Look at customers who bought the existing item before the launch and see what they bought after: the new item, the old one, or nothing from the category. Switching rates from that group are your transfer estimate. Category-level basket analysis fills in the rest, separating customers who traded across from customers who were newly attracted to the category, which is halo rather than cannibalization.

Halo is the part teams forget. A new unit that pulls some demand from a neighbor while lifting total category volume is a different decision from one that only splits an existing line. Measure both directions or the net number is wrong.

One more correction while you are in the data: stop treating the answer as two buckets. A new store's forecast demand arrives from several distinct places, and lumping them together hides the deal. Transfer from your own nearby units. Customers taken from a competitor. Demand in the trade area nobody was serving. Relief for a unit already running at capacity. And a residual whose origin you honestly cannot attribute, which deserves its own line rather than a quiet rounding into one of the others. Geod's site-selection framework breaks candidate demand out along those lines. An 18 percent transfer rate on a site that also wins heavy competitor share is a very different deal from 18 percent transfer and nothing else, and a cannibalized-or-incremental read cannot tell the two apart.

4. Convert transfer into net profit

Transfer is a sales number. Decisions get made on profit, so run it the rest of the way.

1. Cannibalization rate (%): for products, (sales lost on the existing item / sales of the new item) × 100. For locations, (estimated sales lost by the existing store / existing store sales before the opening) × 100.
2. Incremental sales: estimated new store or product sales, minus cannibalized sales from the existing network.
3. Net profit impact: (incremental sales × profit margin) minus the lost profit on cannibalized sales.
4. Proximity overlap (%): for locations, shared customers divided by total customers in the region, times 100.

For a step-by-step walkthrough of each formula with worked examples, see our guide to calculating cannibalization rate. To ask the broader question of whether a market can hold another unit at all, see market saturation analysis.

Here is the arithmetic on a sample store:

5. Set the decision rule before you see the number

Decide what result would make you walk away, and decide it before the model runs. A rate you set in advance is a gate. A rate you interpret after the fact is a negotiation you will lose, because a great-looking corner is very good at talking a team past a threshold nobody wrote down.

A workable rule names three things: the overlap rate above which the site needs a different answer, the minimum net profit the opening has to clear anyway, and who gets to override it. Ten to 20 percent of new unit revenue is a common tolerance band, but treat it as a starting point rather than a law, because margin and category move the line. Our prevention playbook covers how to build those gates into site selection.

Product cannibalization analysis vs. store cannibalization analysis

The two analyses share a skeleton and almost nothing else. Same difference-in-differences logic, different data, different clock, and very different consequences for getting it wrong.

The reversibility row is the whole reason store cannibalization gets the pre-commitment treatment. A product launch that eats its neighbor is a bad quarter. A store that mostly serves customers who were already yours is a bad decade, and no amount of post-opening analysis un-signs the lease.

What triggers cannibalization

Cannibalization has a short list of usual causes, and knowing them tells you where to look before the numbers come in.

• Pricing: a new item priced just below a comparable, higher-margin one gives customers a reason to trade down. Discounts do the same thing on a shorter timescale.
• Assortment overlap: new items with trending features make older ones look redundant. Fashion sees it every season; packaged goods see it whenever a line extension replaces demand instead of creating it.
• Insufficient differentiation: two units, on a shelf or on a map, that do roughly the same job for roughly the same customer will split that customer rather than add one.
• Proximity without a plan: a strong-looking site close to an existing store is the most expensive version, because the commitment is long and the overlap is baked in on day one.

The cost of missing one shows up in the closure column a year or two later. Grocery Outlet opened 42 stores in fiscal 2025 and then identified 36 with no viable path to sustained profitability, two dozen of them on the East Coast, or 30 percent of that region's store count. CEO Jason Potter told investors on the March 2026 earnings call that "it's clear now that we expanded too quickly." Starbucks ran its own correction in September 2025, closing about 400 US locations concentrated in large metros, including 42 shops in New York City alone.

Not every cause is a mistake, though. Procter & Gamble knew Tide would take sales from its own soap brands and launched it anyway, because the category shift was the point. Apple has run the same play for two decades. The difference between those and the closure lists is that somebody priced the trade in advance instead of finding it in a quarterly report.

Where cannibalization analysis goes wrong

Four failure modes account for most of the bad cannibalization numbers we see, and none of them are about math.

Ring-based trade areas. A three-mile radius is a circle drawn for convenience, and it agrees with real customer origins less often than teams expect. One veterinary group we built a custom model with found that actual customer data redrew their trade areas entirely, and that variables like household income frequently correlated in the opposite direction from what the ring analysis assumed. If the trade area is wrong, every overlap number downstream of it is wrong too.

The wrong denominator. Cannibalization rate can be quoted against the new unit's sales or against the existing unit's prior base, and those produce very different-looking percentages from identical facts. Pick one, define it in writing, and make every site use the same one. Comparing a 15 percent from one convention against a 25 percent from another is how a portfolio review goes sideways.

No control set. Covered above, and worth repeating because it is the most common single defect. A before-and-after with no control credits your new store with the weather.

A number with no owner. An estimate that nobody can open is an estimate nobody will defend when a CFO pushes on it. If the analysis cannot show which inputs produced the overlap figure, the meeting turns into a debate about the tool rather than the site. That is a solvable problem, and it is mostly a question of whether the method is inspectable.

Measuring what actually happened after you open

The analysis you run before signing is a forecast. The backtest is what nobody runs, and it is the step that makes every forecast after it better.

Set a review at 90 days and again at four quarters. Pull the same affected set and the same control set you froze before the opening, and compute the realized cannibalization rate the same way you computed the projected one. Then put the two numbers side by side. If you projected 12 percent and got 21, the gap tells you something specific about how your model reads that market type, and you can correct it before the next five sites.

Most expansion teams never close this loop, which is why cannibalization estimates in year five are usually no better than the ones in year one. The comparison costs a morning. Skipping it costs an assumption you keep re-using.

Strategic cannibalization: when the trade is worth making

Some overlap is bought on purpose. As Steve Jobs put it, "If we don't cannibalize ourselves, someone else will." The strategic version shows up in four recognizable shapes: replacing an aging product with a better one, blocking a competitor from a corner you would rather not concede, moving customers to a higher-margin format, or reaching a customer segment the existing unit cannot serve.

What separates strategic from accidental is not the size of the hit. It is whether the number was on the table when the decision got made. A 20 percent overlap you priced into the business case is a trade. The same 20 percent discovered in a quarterly review is a miss.

From analysis to decision

Understanding cannibalization only pays when it changes what gets approved. The teams that get value from this work share one habit: the overlap number arrives with the site packet, not after it.

That is the workflow GrowthFactor was built for. Overlap between a proposed site and every existing unit is modeled as part of scoring the location, with each input visible and adjustable rather than buried behind a score. Cavender's Western Wear cut analyst time per site evaluation in half this way and avoided three locations the analysis flagged, worth roughly $2M. Across customers, roughly 80 percent fewer underperforming locations show up once the workflow is in place (JAN2026 customer survey).

The same data runs the other direction too. A high cannibalization rate is not automatically disqualifying, and some operators find that closing an underperforming unit lifts network profitability as sales consolidate into the stores that convert them better. That is portfolio optimization working off the overlap figures you already built, fed by site selection that reads network effects rather than one address at a time, retail location analysis on the inputs, and demand forecasting that accounts for what a new unit takes as well as what it adds.

GrowthFactor does the analysis. You make the call. Plans run from GrowthFactor Pro for teams evaluating sites week to week up to Labs for network-wide optimization, and our retail expansion planning software covers the workflow around the decision.

Ready to put the overlap number on the table before the committee meets? See how expansion teams use GrowthFactor.

Frequently Asked Questions about retail cannibalization analysis