NPS surveys for multi-site restaurant groups: comparing loyalty across sites
You will be able to choose between a centralized group NPS program and site-led surveys, then compare loyalty fairly across your restaurants.
Run one NPS program centrally, not one per site. Ask the same 0-10 question at the same point after each visit, and tag every response to the restaurant where the diner ate. Compare sites on rolling 90-day scores with a minimum response count. Judge trend over rank, and contact every detractor within 48 hours.
Should a restaurant group run one NPS program or let each site run its own?
Most multi-site groups end up with Net Promoter Score by accident. One site prints a QR code on the check. Another sends a link from a free form builder. A third asks on the card machine. Each produces a number, and none of the numbers can be compared.
Net Promoter Score was introduced by Fred Reichheld of Bain & Company in 2003. It asks one question: how likely are you to recommend us to a friend or colleague, on a scale of 0 to 10? Diners scoring 9 or 10 are promoters, 7 or 8 are passives, and 0 to 6 are detractors. NPS is the percentage of promoters minus the percentage of detractors, so it runs from -100 to +100.
The question is simple. Running it fairly across 10, 30 or 100 restaurants is not. You have two broad options:
- A centralized group program. Head office sets the question, the channel, the send timing, the scoring rules and the follow-up process. Sites act on the results.
- Site-led surveys. Each general manager chooses how, when and whether to ask. The group collects whatever comes back.
For any group where the sites share a brand and you want to compare them, the centralized model wins. Site-led surveys still make sense in two situations: separately owned franchise sites, and venues so different in format that a shared benchmark misleads. Even then, keep the core question identical.
Where do centralized and site-led NPS programs genuinely differ?
The differences are about comparability, ownership and speed of follow-up, not the survey tool itself.
| Axis | Centralized group program | Site-led surveys |
|---|---|---|
| Question wording | One wording, fixed across every site | Each site writes its own, often with different scales |
| Send timing | One rule for every site, such as two hours after the visit | Whenever the site remembers, often on the check while staff watch |
| Channel | One channel, typically post-visit email to opted-in diners | Mixed: QR codes, card machines, paper, email |
| Comparability across sites | Direct, once each site passes a minimum response count | Not comparable, because method changes the score |
| Who owns the guest data | The group, as a single data controller under CCPA/CPRA | Each site, with several tools and several data processors |
| Detractor follow-up | Routed automatically to the right general manager with a deadline | Depends on whether the manager checks the tool |
| Setup effort | One build, then rolled out site by site | Repeated at every site, and rebuilt when managers change |
| Best suited to | Managed estates under one or two brands | Franchised estates or one-off flagship venues |
The row that matters most is comparability. A score collected at the table, with a server standing nearby, is a different measurement from a score collected by email after the diner has gone home. Mixing the two across sites produces a league table that ranks survey methods, not restaurants.
How do you compare NPS fairly across restaurant sites?
A raw league table is the fastest way to lose your managers' trust. Five rules make site-level scores fair enough to act on.
Use the same question, moment and channel everywhere
Fix the wording, the 0-10 scale, the send delay and the channel at group level. A post-visit email triggered from a Guest WiFi login works well for multi-site groups. The diner connected at a specific restaurant, so every response arrives tied to the site that earned it. For timing and question design, see the Post-visit feedback survey checklist for hotels and restaurants.
Set a minimum response count before you rank a site
Small samples swing hard. Every response is worth 100 divided by the number of responses in percentage points. One diner moving from promoter to detractor therefore shifts the score by twice that.
| Responses in the window | Points moved by one extra detractor | Points moved by one promoter becoming a detractor |
|---|---|---|
| 25 | 4 | 8 |
| 50 | 2 | 4 |
| 100 | 1 | 2 |
| 200 | 0.5 | 1 |
At 25 responses, a single bad Saturday can move a site eight places down the table. A threshold of 100 responses per window is a sensible floor. Below it, report the site as "insufficient data" rather than ranking it.
Score on rolling 90-day windows
Monthly scores at a quiet suburban site will rarely clear 100 responses. A rolling 90-day window smooths out one-off events, such as a broken dishwasher or a short-staffed federal holiday. It still reacts within weeks to a real change, such as a new head chef.
Compare sites with their peers and their own baseline
A downtown site serving tourists and a shopping-center site serving regulars draw different diners. Group sites into clusters, such as downtown, shopping center, travel hub and neighborhood, and rank within each cluster. Then report each site against its own score from the previous 90 days. Movement against its own baseline is the fairest single signal you have.
Read the reason, not only the number
Add one follow-up question: "What is the main reason for your score?" Tag the free-text answers into a short list: food, service, speed, value, atmosphere and cleanliness. A site at +30 with detractors citing speed needs a different fix from a site at +30 with detractors citing value.
When is a centralized group NPS program the right call?
Choose the centralized model when your locations trade under a shared brand, your managers report to one operations team, and you want one view of loyalty across the estate. It is also the right call when you plan to act on detractors. Follow-up only works if responses reach the right manager on a deadline.
Worked scenario: a 32-location casual dining group
The figures in this scenario are illustrative, built to show the arithmetic.
Situation. A 32-location casual dining group ran location-led surveys. Twenty locations used a QR code on the check, eight used a card-machine prompt and four did not survey at all. The monthly board pack showed a league table of raw scores. Location 14, a neighborhood restaurant, sat bottom at +18 on 22 responses: nine promoters, eight passives and five detractors. Its general manager was placed on a performance plan.
What was done. The group moved every location to one program. Each diner who opted in through guest WiFi received the same 0-10 question by email two hours after their visit, plus one reason question. Scoring moved to rolling 90-day windows with a 100-response minimum, ranked within four clusters.
Outcome. After the first full window, Location 14 had 140 responses: 72 promoters, 39 passives and 29 detractors, for a score of +31. That placed it mid-table within the neighborhood cluster. Two city-center locations, previously shielded by enthusiastic table-side QR responses, dropped below +10. Their detractors consistently cited speed of service at lunch. The group moved its attention, and its kitchen staffing review, to the locations that needed it.
When are location-led NPS surveys the right call?
Location-led surveys are the right call in two narrow cases.
The first is a franchised estate. Under CCPA/CPRA, each franchisee that decides how its diners' data is used is a separate data controller. The brand cannot pool that personal data simply because the locations share a logo. In that case, each franchisee runs its own survey and shares only aggregate scores with the brand.
The second is a genuinely different format, such as a single fine-dining flagship inside a group of fast-casual locations. Its diners, spend and expectations differ so much that a shared league table misleads more than it informs.
In both cases, keep the core NPS question, scale and send timing identical. That way, the numbers sit on the same footing even when the data stays apart.
Worked scenario: a brand with franchised and corporate-owned locations
The figures in this scenario are illustrative.
Situation. A burger brand operated six corporate-owned locations and licensed 18 more to franchisees. The brand wanted to compare loyalty across all 24. However, it held no rights to the franchisees' guest data, and each franchisee used a different survey tool.
What was done. The brand published a short NPS standard: the exact question, the 0-10 scale, a two-hour post-visit send and the 90-day, 100-response scoring rules. Each franchisee kept its own diner list and survey. Each month, they submitted only the counts of promoters, passives and detractors. The six managed sites ran the centralized program directly.
Outcome. Within one quarter, 17 of the 18 franchisees reported on the shared standard. The brand had a comparable score across 23 of its 24 sites for the first time. No personal data had moved between controllers. The managed sites, with full detractor follow-up, became the reference group for best practice.
How do you act on detractors before they lapse?
A detractor is a diner at risk of lapsing, meaning they stop visiting. NPS earns its keep only when someone contacts that diner while the visit is still fresh.
Route every detractor to a named person with a deadline
Send each detractor response straight to the general manager of the site where the diner ate. Set a 48-hour target for a personal reply. Score the manager on reply rate and reply time, not on the raw NPS. That removes the incentive to steer who gets surveyed.
Make the reply human, then make it count
The reply should acknowledge the specific reason the diner gave and say what the site will change. Where appropriate, invite the diner back. Keep offers out of the survey request itself unless the diner has opted in to marketing. CCPA/CPRA and the TCPA and CAN-SPAM require consent to be freely given, specific, informed and unambiguous, and withdrawable at any time.
Measure recovery with a holdout group
To prove follow-up works, hold back a random 20% of detractors from the callback. Compare how many diners in each group return within 60 days. With guest WiFi in place, a return visit shows up when the diner next connects at any of your sites.
Worked scenario: a 20-site bar and dining group
The figures in this scenario are illustrative.
Situation. A 20-site bar and dining group collected around 800 NPS responses a month. About 14% were detractors, which is roughly 112 diners a month. Nobody replied to them.
What was done. The group routed every detractor to the site's general manager with a 48-hour reply target. It held back 20% of detractors as a control group, which left about 90 diners a month receiving a callback.
Outcome. Over the first quarter, 22% of contacted detractors returned within 60 days, against 12% of the holdout group. That 10-point gap equates to about nine recovered diners a month across the estate. Multiply by your average spend and visit frequency to put a lifetime value on the program. The reason tags also showed that two sites produced 40% of all "value" complaints. That pointed the group to a pricing review rather than a training issue.
Do not forget promoters
Promoters are your cheapest source of public reviews. Send each promoter a review request on the platform that matters most for that location. For help choosing, see TripAdvisor vs Google reviews for hotels: where to focus review requests.
What does each NPS model cost to run?
The license fee is rarely the largest cost. The bigger costs are staff time, duplicated tools and decisions made on numbers that do not mean what the board thinks they mean.
Set-up
A centralized program is built once: question wording, send rules, reason tags, thresholds and routing. Each new location then inherits the same set-up. Location-led surveys are rebuilt at every location and often again when a manager leaves, because the survey lived in their personal account.
Running costs
Centralized programs need one analyst to own the monthly cluster report, plus general manager time for detractor replies. Location-led programs spread survey administration across every manager, with no shared report at the end.
Data protection
Every survey tool that stores diner responses on your behalf is a data processor. CCPA/CPRA regulations require proper contractual protections with each one. Ten locations on four different tools means four processor contracts to review, plus four places to honor access and deletion requests. A single program means one.
The list itself
Mailchimp, Klaviyo and HubSpot can all send a survey email. They send it to a list you already hold. For most restaurants, that list is thin, because diners rarely hand over an email address at the table. The survey can only reach the diners you have already captured.
Purple Engage starts one step earlier. Diners log in to guest WiFi, choose whether to opt in to marketing through conscious-choice opt-ins, and arrive in your CRM as consented first-party data. Each record is linked to the restaurant where they connected. The survey list builds itself from foot traffic you already have. Purple runs at 80,000+ live venues and recorded 440 million logins in 2024 (Purple's own data). See the Product overview for how surveys and campaigns sit together in Engage.
How should you decide for your restaurants?
Match your business to the row that describes it.
| Your situation | Recommended model | Why |
|---|---|---|
| Managed locations, one brand, 10+ locations | Centralized group program | Comparable scores and one detractor workflow |
| Managed locations, two or three brands | Centralized, with separate clusters per brand | Shared rules, fair benchmarks within each brand |
| Franchised locations | Shared NPS standard, location-led delivery, aggregate reporting | Each franchisee remains its own data controller under the TCPA and CAN-SPAM |
| One flagship plus a fast-casual group | Centralized for the group, flagship reported separately | The flagship's format would distort the benchmark |
| Fewer than five sites, low covers | Centralized, with 180-day windows | Small sites need longer windows to reach 100 responses |
| No way to reach diners after they leave | Fix data capture first, through guest WiFi | Without consented contacts, every survey is table-side and biased |
A five-step rollout
- Write the standard. Fix the question, scale, send delay, reason tags, response threshold and window length. Publish them to every general manager.
- Fix capture before you survey. Make sure every site collects consented diner contacts through guest WiFi, linked to that site.
- Pilot in one cluster. Run four to six sites for one full 90-day window. Check response counts against your threshold before you show any ranking.
- Switch on detractor routing. Set the 48-hour reply target and the 20% holdout from the first day of the pilot, so you can prove recovery.
- Roll out and report by cluster. Add the remaining sites. Report each site against its cluster and its own baseline, never as one raw league table.
Frequently asked questions
How much does an NPS program cost to run with Purple?
Surveys and campaigns sit within Purple Engage, one of three Purple plans alongside Connect and Capture. Pricing is quoted for your property portfolio, based on the plan and the venues you bring. The larger cost to plan for is internal: one person owning the monthly cluster report, and general manager time for 48-hour detractor replies. Consolidating several site-level survey tools into one program usually removes duplicated subscriptions and processor contracts.
Will Purple work with the WiFi hardware already in our restaurants?
Yes, Purple is hardware-agnostic and runs as a cloud overlay on the access points you already own. Supported hardware includes Cisco Meraki, HPE Aruba, Ruckus, Juniper Mist, Ubiquiti UniFi, Cambium, Extreme and Fortinet. You do not need to replace equipment to start capturing consented diner contacts. Those contacts become the list your NPS surveys go to, linked to the restaurant where each diner connected.
Can we keep our existing NPS history when we move to one program?
Yes, keep it as a reference, but do not splice it into your new trend line. Historical scores collected through different channels, wording or timing measure something slightly different. Archive the old figures by site and note the method used. Then start a clean baseline with the new standard. After one full 90-day window, you can compare each site against its own new baseline with confidence.
Is it CCPA/CPRA compliant to survey diners who logged in to guest WiFi?
Yes, when the diner has given valid consent and you keep the survey within what they agreed to. CCPA/CPRA requires consent to be freely given, specific, informed and unambiguous, and as easy to withdraw as to give. Use conscious-choice opt-ins at login, keep offers out of the survey unless the diner opted in to marketing, and set a retention period for responses. Honor access and erasure requests from one central record.
How is this different from running NPS in Mailchimp, Klaviyo or HubSpot?
The difference is where the list comes from. Mailchimp, Klaviyo and HubSpot send surveys to contacts you already hold. Purple Engage builds the list from guest WiFi logins, so diners arrive as consented first-party contacts linked to the location they visited. That gives you a larger survey audience at every restaurant. It also gives fair location comparisons and a way to see when a contacted detractor returns.
How many responses does each restaurant need before its NPS means anything?
Aim for at least 100 responses per location in each scoring window. At 100 responses, one diner moving from promoter to detractor shifts the score by two points. At 25 responses, the same change moves it by eight. Below the threshold, report the location as "insufficient data" rather than ranking it. Extend quiet locations to a 180-day window instead of lowering the bar.
How long does it take to roll out one NPS program across a restaurant group?
The survey can go live at a location as soon as guest WiFi is capturing opted-in diners there. Reliable comparisons take longer, because each location needs one full 90-day window above the response threshold. A practical plan is a pilot in one cluster of four to six locations for one window. Then roll out to the rest of the portfolio, with detractor routing switched on from day one.