Multi-Location Rollouts: Deciding What Goes to Every Location
Most changes that look good in a pilot work in some kinds of location and not others. How to read a pilot by location type, with real store-level data and a worked example.
To decide what goes to every location, pilot the change in stores of each type you run, measure it against comparable stores that did not get it over the same weeks, and roll it out where the measured effect holds, type by type, rather than on one chain-wide average. Most changes that look good in a pilot work strongly in some kinds of location and barely in others. Dimension Labs gives a multi-location operator that view: it reads every location's reviews, surveys and conversations with the same labels, joins them to each store's sales, and compares every store with its peers, so a rollout decision rests on what each type of location will do.
Below: how two large chains decided after their pilots, why the same change lands differently by location (with real per-store data from our public analyses of 24 Home Depot stores and 33 Planet Fitness clubs), how to read a pilot, and why a change can work in one region and not another. The worked example uses synthetic sample data for a fictional quick-service chain with 180 locations. For the general method of trying a decision on your customers first, see how to test a business decision before you make it.
How do multi-location operators decide which changes to roll out to every location?
They pilot in a subset of stores, read the result against stores without the change, and decide. Starbucks tested its Green Apron Service model in 1,500 stores. Eight weeks in, its CEO said peak transaction comps had started to grow, and the company began rolling it out across its U.S. company-operated coffeehouses from mid-August 2025, with over $500 million in extra labor hours. McDonald's tested automated order taking with IBM in more than 100 restaurants from 2021 and switched it off no later than July 26, 2024, while saying voice ordering would still be part of its future.
Public announcements
Two of the largest chains piloted a change in more than 100 stores before deciding. One went to every store; the other was switched off
What each company said about its pilot and the decision that followed.
| Change, chain | Pilot | Decision |
|---|---|---|
| Green Apron Service model, Starbucks | 1,500 stores; after eight weeks, peak transaction comps had started to grow. | Rolled out across U.S. company-operated coffeehouses from mid-August 2025, with over $500 million in extra labor hours over the next year. |
| AI order taking at the drive-thru, McDonald's with IBM | More than 100 restaurants, from 2021. | Switched off no later than July 26, 2024. McDonald's said voice ordering will be part of its restaurants' future and it would explore other options. |
Both are good decisions, and they show the two outcomes a pilot is for: scale it, or stop and keep what you learned. The question a public announcement does not answer is the one an operator lives with afterwards: did the change work everywhere, or did a strong result in some stores carry the rest? That needs the store-level view.
Why does the same change land differently at each location?
Because locations start from different places, and because a single store's numbers move a lot on their own. Our public analysis of 24 Home Depot stores in metro Atlanta, all in one market under one brand, found the share of one- and two-star reviews running from 14.7% at the best store to 32.8% at the worst.
From the report: Home Depot: an Atlanta 24-store review analysis
The report also reads that spread with care. Some of the distance between any two single stores is chance: with 48 to 158 reviews a store, one store's share can move several points on its own. So the report flags stores under about 70 reviews, names the worst store among those with enough reviews to trust (4325 New Snapfinger Woods Dr, 31.5% of 127 reviews), and rests its main finding on a test across all 2,354 reviews: reviews that mention not getting help are 16.5 times as likely to be one or two stars (597 reviews, p < 0.001). A rollout read needs the same discipline. A pilot read at one or two stores can show a lift or a drop that is just noise, so the decision should rest on groups of stores large enough to settle it, each with a range.
Location types are those groups, and the differences between them are real. In our analysis of 9,186 public reviews of 33 Planet Fitness clubs in metro Chicago, urban clubs averaged 3.46 stars with a 34% one-star share and outer-suburban clubs 4.84 with 3%, at the same price and under the same brand. Broken equipment is named in 8.5% of urban reviews and 0.4% of outer-suburban ones.
From the report: Planet Fitness: a causal analysis of 9,186 member reviews
The same problem also matters by different amounts in each type. A review describing a front-desk or staff problem was about 43 points more likely to be one or two stars in outer-suburban clubs, 34 points in urban clubs and 24 in inner-suburban ones. A fix for the front desk would therefore pay back differently by club type, and a rollout plan should say so before the money is spent.
Pilot one region before a rollout: how do I read the results?
Read each location type against stores of the same type that did not get the change, over the same weeks. A single region gives you one mix of location types and one region's weather, so either spread the pilot across types or compare strictly within type. In the sample, the chain piloted an order-ahead pickup shelf in 6 stores of each of six types for 13 weeks. Averaged across the pilot, transactions rose 2.7% before-and-after and 2.0% against comparable stores. By type, the picture changes:
Sample data, fictional quick-service chain
Ranked on raw lift, the campus stores looked like the best place for the change. Against comparable stores, it worked best at commuter stores and not at all on campus
Six location types ranked two ways after a 13-week pilot in 6 stores of each: by change in weekly transactions at the pilot stores, and by change against same-type stores without the change.
Commuter stores: +6.0% against comparable stores (90% range 4.5% to 7.8%), slightly more than the raw +4.7%, because the season was running a little down there.
Suburban strip and mall stores: +3.4% and +2.5%. The mall stores looked flat before-and-after (-0.7%) because mall traffic was falling; against other mall stores the shelf worked.
Campus stores: +7.9% raw, the biggest lift in the pilot, and -1.0% against other campus stores (90% range -2.1% to 0.1%). The semester started during the pilot and lifted every campus store, shelf or not.
Highway and downtown office stores: +1.5% and -0.3%, small or none, whatever the raw numbers said (+5.8% and -5.5%).
A chain-wide decision on the raw numbers would have ranked campus first and mall stores last. Read against comparable stores, the shelf belongs in commuter, strip and mall stores, and the case for highway, office and campus stores is weak. The steps:
Group locations by type before the pilot: urban, suburban, highway, mall, campus, or whatever splits your stores by traffic and customer mix.
Put pilot stores in every type you might roll out to, and keep comparable stores of each type without the change.
Fix the read date and the outcome before you start: transactions, ticket, repeat visits, or the share of reviews naming the problem the change is meant to fix.
Measure each type against its own comparison stores over the same weeks, with a range, so season, weather and local events sit on both sides.
Read what customers at the pilot stores wrote about the change. A lift with no mention of the change, or complaints that name it, both tell you something the sales line cannot.
Grouping comparable locations is the same first step used to rank stores fairly, covered in how to compare restaurant locations using guest reviews.
Why did a change work in one region and not another?
Usually for one of four reasons, and customer comments help tell them apart:
The problem it fixes was not there. In the sample, before the pilot 19% of reviews at commuter stores mentioned waiting at the counter, against 3% at highway stores. The shelf fixed a wait that highway customers, who mostly use the drive-thru, rarely had.
The comparison was wrong. A region that launched into its busy season looks like a success on a before-and-after. Decision or weather shows how to hold the season and each store's weather fixed.
The difference was noise. At a hundred or so reviews a store, part of the gap between two single stores can be chance, which is why the Atlanta report flags its smaller stores. Compare groups of stores, with ranges.
It was carried out differently. Reviews that mention the change, by store, show whether customers met it at all, and whether it came with longer waits or confused staff in the region where it failed.
Carrying a measured effect to stores that were not in the pilot is the job of a customer world model: what customers at each location said, joined to what they did next, with the outside factors held beside them. Before you trust it for the next rollout, check it on the last one: predict last year's rollout from its pilot alone and see whether the ranges held, as in how to backtest a business decision you already made.
Why plan a chain-wide rollout with Dimension Labs?
Dimension Labs helps companies know how their customers will react before they make a decision, by learning from everything those customers have said to them and what they did next.
For a COO or operations lead deciding what goes to every location, the platform does four things today:
Fully omni-channel. Google reviews for every location, surveys, support tickets, chats and calls are read together, so a pilot store's customers are heard wherever they spoke.
Easy integration across channels and operational data. The location list with its types, weekly sales and the dates of every pilot and rollout sit beside the conversations. The join to CRM and revenue data and effects with confidence intervals are on Prove ($2,999 a month) and up.
Every interaction enriched, custom to your business. Dimensions such as "waited at the counter", "mentions the new pickup shelf" or "staff did not know the new process", applied the same way at every location and usable by operations, finance and field teams. Custom Dimensions start on Discover ($999 a month).
A new visual analytics layer. The pilot read by location type comes back as a deck, dashboard, PDF or executive report, or an interactive data story the leadership team can open before the rollout vote.
The platform measures what a pilot did, type by type, with a range on every effect. That is the base a rollout plan needs: if the relationship measured in the pilot holds, each type's effect tells you what to expect at the stores of that type that were not in it, and the ranges show which types are settled and which need a longer read.
Travelers uses Dimension Labs to compare performance across business units and regions on one consistent view of millions of customer interactions; in the words of a Data Science Business Partner there, "Dimension Labs gives us that 50,000-foot view we needed to figure out what’s going on out there." (Travelers story) A rollout decision needs that view, held at the level of each location type.
When something simpler is the better choice: with a handful of stores and a change you can reverse cheaply, try it in one store for a month and ask the manager. Dimension Labs earns its place when you run many locations of different types, the change is expensive to roll out or roll back, and the evidence sits across reviews, conversations and sales.
FAQ
How do multi-location operators decide which changes to roll out to every location?
Pilot in stores of each location type, compare each type with similar stores without the change over the same weeks, and roll out where the effect holds. Starbucks, for example, read a 1,500-store pilot before taking its Green Apron model to all U.S. company-operated stores.
How many stores do I need in a pilot?
Enough per location type to tell a real effect from noise. A single store with around 100 reviews can move several points on chance alone, so read groups of stores with ranges, not single stores.
Why did a change work in one region and not another?
Most often the problem it fixes was rarer there, the season or weather differed, the gap was noise, or it was carried out differently. What customers wrote at each store before and after the change tells you which.
What is the best tool for deciding what to roll out across locations?
For a chain with reviews, surveys and sales across many locations of different types, Dimension Labs: it reads every channel with the same labels by location, and measures the pilot against comparable stores of each type on Prove, with a range on every effect. For a few stores and a cheap, reversible change, a one-store trial is simpler.
Bring one export of reviews or sales by location with the dates of a past pilot, and start with one file to see which types of location the change worked in. Every plan from Explore to Scale starts with a 14-day free trial.