A Happy Cog field guide · Paid media and performance marketing practice
The Franchise Marketing Field Guide.
How restaurant brands with two hundred or more units turn marketing spend into measured visits, now that the recommendation is increasingly assembled inside an AI assistant before anyone opens a browser. Your sales are up and your traffic is not, and the gap between those two numbers is the conversation you're about to have with your CFO.
This is the second edition, and the shifts since the first all carry deadlines. Discovery started moving into AI assistants, the delivery platforms turned themselves into media networks that hold your purchase data, and Google began folding local advertising into automation you don't control, on a timetable that runs through 2027. A short section near the end names the calls we got wrong the first time.
The argument in four sentences.
01
Restaurant sales are tracking up about four percent this year while real growth sits closer to one percent, which means most of the headline is menu price rather than traffic, and your CFO already knows it.
02
The question of where to eat is increasingly answered inside an AI assistant that builds its recommendation from two hundred separate location records, and the weakest of those records is usually the one nobody at the brand actually owns.
03
At the same time, the delivery platforms that take a cut of every order have turned themselves into media networks that can reach your own customers on the open web for a competitor, and Google is moving local advertising into automation that runs on the same listing data, on a timetable through 2027.
04
The way through is to measure incremental same-store sales lift per marketing dollar at the unit level, to make your menu and location data readable by the machines now doing the recommending, and to fund that work out of the three to five percent of budget nobody has defended in two cycles.
Part one · The premise
The franchise marketing math has changed.
For most of the last decade, a restaurant marketing program could point to a sales number and call it growth. That's harder now. Total restaurant and foodservice sales are tracking up roughly four percent in 2026, but once you strip out menu price increases, real growth is closer to one percent, with traffic flat to down across most segments. A team can post a number that looks like growth without being able to say whether it bought a single additional visit, and that is the gap you'll be asked to close in your next budget review.
Three things have changed since we published the first edition of this guide, and each one carries a deadline.
i
Discovery moved inside the assistant.
A customer now asks an assistant where to eat before they open a browser, in a sentence shaped like a conversation: somewhere near Cherry Hill to take my parents, not too loud, open past nine, not Italian again. The assistant returns two or three names, and your brand is one of them or it isn't. It builds that answer from structured location data, review text and how recent it is, and what your records say across Google Business Profile, Apple Business Connect, and the directories that syndicate from them, and it resolves all of it for one specific location. So your visibility in AI search is two hundred numbers, one per location, and the median matters more than the best.
ii
Your aggregator became a media network.
On June 4, 2026, DoorDash consolidated DoorDash, Wolt, and Deliveroo into a single commerce media platform spanning more than four hundred thousand advertisers, with offsite audience extension and a clean room for measurement. Read that as a delivery company telling the market it's now in the media business. The platform that takes twenty to thirty percent of every order it delivers can now reach the customers who placed those orders out on the open web, using data you generated, on behalf of whoever is buying the placement. Defending your position inside the app used to be the whole job. Now that the app is also a demand-side platform, it's half of it.
iii
Google is automating your local advertising.
In August 2026, Google began moving Local Services Ads out of their own console and into Google Ads as a pay-per-lead Performance Max campaign type, in phases running through 2027. The ads still appear only on Search and Maps and you still pay for valid leads rather than clicks, but the old dashboard stops being your management home, historical reporting doesn't migrate (export it before your account moves), and because the campaigns now sync from Google Business Profile in real time, your listing hygiene becomes a paid media dependency as well as an organic one. The same franchisee-maintained record that decides whether an assistant recommends you now also decides what your paid local ads say.
Diagram A
Your AI visibility is two hundred numbers.
Weak record Strong record The median record
One square per location record. Corporate owns the brand page, the franchisee owns the location record, and the assistant reads the location record. The median across all of them decides how often you get recommended.
Sales are up because prices went up, and traffic hasn't moved. That is the sentence your CFO is about to say back to you.
Part one · Continued
Which channels deserve funding now?
Two arguments from the first edition are settled, so this section spends the space on how to buy rather than whether to. Connected television is a performance channel with footfall measurement, and digital out-of-home earns a line in a franchise plan. What's worth your attention now is buying each channel so you get a report you can defend.
Connected television is worth funding, and you should write footfall measurement into the plan as a purchase criterion before launch rather than accept it as a report afterward. Every platform will tell you it measures store visits. What varies is whether the design reaches statistical significance inside your actual footprint, and that depends far more on how geographically concentrated your units are than on how much you spend. A brand with sixty units across four neighboring states can get a defensible read at a modest budget, while a brand with two hundred units spread across forty-seven markets often can't at any budget it would sensibly commit, until it narrows to a test region first.
Digital out-of-home is still the most underused channel in restaurant marketing. Programmatic activation and dynamic creative have made it a different product from the billboard buys of a decade ago, so a rainy afternoon in Seattle pushes a soup creative and a ninety-degree day near a high school pushes a milkshake thirty minutes before dismissal. Measurement stays directional, and it pairs well with connected TV where you want presence and frequency at a moderate budget. Streaming audio and host-read podcasts remain underweighted, and the right way to target them is by audience and trust rather than by topic, so a drive-time news podcast read by a local anchor in a city where you have units will outperform a national food podcast by a measurable margin.
Figure 1
The channel stack, re-priced.
Most of this table has moved since the first edition, one row is new, and two channels no longer need arguing for.
Addressable reach with footfall measurement$28–$51 CPM$5.6K / month
Mature
How to buy it now
Write footfall measurement into the plan as a purchase criterion before launch, and size the test by exposed households per trade area. Whether the design reaches significance depends far more on how concentrated your units are than on how much you spend.
Programmatic DOOH
Geo-targeted presence with dynamic creative$4.50–$9.00 CPM$3.4K / market
Directional
How to buy it now
Buy it on triggers: weather, daypart, and proximity, so a rainy afternoon pushes soup and a hot day near a high school pushes a milkshake before dismissal. Measurement stays directional, so pair it with connected TV for presence and frequency.
Streaming audio
The commute decision, targetable by daypart$17–$28 CPM$3.4K / month
Mature
How to buy it now
Target by audience and daypart rather than by topic, and weight the drive-time windows when the where-to-eat decision actually gets made.
Host-read podcasts
Highest-trust format in the modern stack$28–$56 CPM$11.3K / pilot
Directional
How to buy it now
Buy on trust rather than topic. A drive-time news podcast read by a local anchor in a city where you have units will outperform a national food podcast by a measurable margin.
Reaching marketplace audiences on the open web, measured in a clean room$12–$22 CPM$11.3K / pilot
Emerging
Retail media (grocery and convenience)
Targeted shopper reach and co-branded placements$9–$17 CPM$5.6K / pilot
Maturing
Direct mail
Underperforming-store rescue and trade-area lift$0.45–$1.01 / piecePer-store basis
Mature, on offer codes
Opted-in SMS
Daypart promotion with weather and event triggers$0.011–$0.034 / sendList-build phase first
Mature
Mobile rewards apps
Acquisition with clean attributionCPA modelOffer underwriting
Native to platform
Cost and investment ranges reflect typical restaurant-category media costs at the time of writing. Measurement maturity reflects each channel's ability to support footfall and incrementality measurement at modest investment. The aggregator offsite extension row is new to this edition, and its range is directional.
Part two · AI visibility
Who owns your visibility inside the assistant?
This is the section we'd read first. Customers have changed how they decide where to eat, and almost no franchise brand has changed who's responsible for the data that decision now runs on. That gap is the largest unfunded opportunity in franchise marketing right now, and it's unfunded precisely because it doesn't belong to anyone: it's too technical for the field team, too local for the brand team, and too unglamorous for the agency that sold you a campaign.
What the assistants actually weigh
An assistant answering that question reads four things, and a beautiful location page with none of them underneath it is close to invisible.
01
Structured attributes
It matches the query against attributes it can read, so price band, cuisine tags, service options, accessibility, hours by day, and increasingly individual menu items all count.
02
Review recency and specificity
A 4.3 with fifty reviews this quarter beats a 4.6 with nothing since last year, because repeated concrete specifics like gluten-free options, parking, quiet, and good for groups are what let it match a request.
03
Consistency across sources
Conflicting hours between your site, your Google listing, and a delivery platform lower its confidence rather than averaging out, and a model with low confidence recommends somebody else.
04
Machine access to your menu
A menu that's an image, a PDF, or rendered by JavaScript a crawler can't execute doesn't exist to an assistant at all.
A diagnostic you can run before your next leadership meeting
You don't have a number for this yet, and you can have one in an afternoon. Pick your ten largest markets, write twenty-five prompts a real customer would use, phrased as conversations rather than keywords and spread across occasions like family dinner, quick lunch near the office, late night, a dietary constraint, and a group booking. Run each one in ChatGPT, Gemini, and Perplexity, and score two things: whether your brand appears at all, and whether it holds the first recommendation. Report it as two percentages by market. That's your baseline, it's repeatable every month, and it's the first marketing number in years that a franchisee will immediately understand and care about.
83%
of restaurants effectively invisible in AI search, according to one 2026 vendor study.
Vendor-sourced. The figure comes from the vendor's own crawl of a sample it chose, so read it as an order of magnitude rather than a measurement. Your own prompt panel produces a number you can defend.
Appearance rate and primary-recommendation rate, reported by market and re-run monthly. The series matters more than the first reading.
The figures filled in above are placeholders. Switch to the blank template to score your own panel. Nothing you type leaves this page.
Discovery happens in the assistant, and checkout stays in your app.
In April 2026, Starbucks launched an app inside ChatGPT that turns a description or a photo into a drink recommendation, Little Caesars shipped a comparable app across the US, Canada, and Mexico the same month, and in July, Square automatically enrolled eligible US food and beverage sellers into ordering through ChatGPT and Claude at standard card rates. Look at what those moves share: discovery happens inside the assistant, and checkout happens inside the brand's own app. Starbucks was explicit about why, and the reason is the loyalty program, because handing checkout to a third party is the aggregator trade repeated in a newer channel, and the brands who lived through the first one aren't volunteering for the second.
The dependency underneath it is a live menu feed, resolved to a specific location and exposed through an interface a machine can call, with the money settling to the correct franchise entity without someone reconciling it by hand at month-end. Order volume through assistants is small today and will stay small for a while, so nobody should move a media budget on the strength of it this year. The reason to build now is timing, because the fee structure is the best in the off-premise mix by a wide margin, and this is a channel where the build takes longer than the demand takes to arrive, which is the opposite of how most channels are timed.
We built our own restaurant ordering platform, Happy Food, partly for this, with multi-entity fund routing through Stripe Connect and live integrations across the major point-of-sale, payment, delivery, and loyalty systems. We'd rather you understood the requirement than bought ours.
The designed edition runs to 36 pages.
Same argument, with the full channel stack, the aggregator renegotiation playbook, the five-audience reporting matrix, and the complete 2027 outlook. Yours to take into a budget meeting.
Every franchise marketing program needs one primary metric, and we recommend incremental same-store sales lift per dollar of marketing spend, measured at the unit level, attributed by channel, and rolled up to the brand. It solves three problems at once: it's the number the CFO already cares about, so it survives the board meeting, it's the number the franchisee cares about, so it survives the advisory council, and it's the number the marketing team can directly influence, so it drives the right behavior. Building it takes three feeds, a clean unit-level sales feed from the point of sale, a clean unit-level spend allocation that includes the co-op dollars franchisees spend locally, and an incrementality method, which is geographic holdout testing for channels with no native measurement. The method takes three to six months to build, and it's the single highest-value investment a franchise CMO can make in this category.
The decomposition your CFO is about to ask for
A same-store sales number is three things wearing one coat. Sales equal transactions multiplied by average check, and average check moves for two separate reasons, because you raised prices and because customers bought a different mix of items. Marketing owns transactions, pricing owns price, and menu strategy and operations own mix. When a brand reports a two percent same-store increase without separating those pieces, marketing takes the credit in a good year and the blame in a bad one, and neither is earned. Report it decomposed and the conversation changes, because you can show that marketing delivered a traffic increase in a quarter where price carried the headline, or admit that it didn't, and either way you're discussing something specific. With nominal growth near four percent and real growth near one, that's the difference between a budget that gets defended and one that gets trimmed.
Interactive
The same-store sales decomposition.
Enter one period's same-store sales growth and the points contributed by price and by mix. Everything calculates in your browser, and nothing you type is sent anywhere, stored, or tracked.
Traffic contribution marketing owns
+1.0 pts
25% of the headline
Price Mix Traffic
Traffic is a small slice of this number, so most of the growth came from price and mix. Report the traffic line on its own, because that's the part marketing can defend as its own work.
Figure 3
One metric, five audiences.
Audience
What they care about
What to show them
Cadence
CEO and board
Same-store sales growth, marketing return, defense of the ad fund
Incremental lift per dollar rolled up to brand level, with the price and traffic decomposition
Quarterly
CFO
Margin, cash conversion, capital allocation
Incremental lift per dollar plus contribution margin by channel
Monthly
Franchise advisory council
Where their stores' ad-fund dollars went and what they produced
Unit-level dashboard with trade-area spend, measured footfall and lift, and market-level share of assistant recommendations
Monthly
Marketing team
Channel and campaign performance for tactical decisions
Channel-level incrementality, attributed conversions, click-through, cost per thousand, footfall
Forecast footfall lift by daypart and market, two weeks forward
Weekly
Every audience looks at the same underlying data, and only the view changes.
The demand that leaves no click path
The demand from the assistant arrives without a pixel. The assistant recommends you, the customer opens your app directly, and nothing in any channel report will ever connect those two events. We'd rather say plainly that this measurement is immature than pretend otherwise, which is exactly what this guide criticizes vendors for doing. What works today is a three-part read.
Leading indicator
The prompt panel
Run it monthly and report it as share of recommendations by market.
Lagging indicator
Direct and branded search
This is the series that moves first when assistant visibility improves.
The number that survives scrutiny
A geographic holdout
Choose the markets where you improve listings and feeds so comparable markets are left alone.
Server-side tracking
A measurement stack built on browser pixels will fail a little more every year, and the platforms have already decided to move server-side, so you're only choosing whether to migrate on your own schedule or have it forced on you. Done through a tag manager's server-side container, it captures events browser pixels miss, removes tag weight from the page, and gives you control over which data goes to which platform. Budget three to four months, and expect the benefit to compound from the day it goes live.
Part four · National to local
The national-to-local handoff.
A great national program produces national impressions, and local execution turns them into store visits, and the handoff between the two is where most franchise brands lose twenty to thirty percent of their program's value. Marketing development funds, co-op programs, and brand-fund contributions all express the same tension, because the brand wants consistent execution on activities it controls and the franchisee wants flexibility on activities they control, and neither side is wrong.
The brands that resolve it well share three habits
01
They publish a clear allowed-use list, so a franchisee knows before spending which activities are reimbursable, at what rate, and under what creative approval, since ambiguity is the single biggest source of friction in a co-op program.
02
They provide turnkey local creative a franchisee pulls from a portal and customizes within bounded parameters.
03
They report on co-op performance with the same rigor as national, which earns them the standing to be more prescriptive about future allowed uses.
The migration you didn't ask for.
The strongest program in this category is a brand-managed local digital playbook, where the franchisee opts in, pays a contribution, and the brand runs Google local advertising against their trade area at scale. That recommendation stands, and the mechanics under it changed in August 2026 when Google began migrating Local Services Ads into Google Ads as a pay-per-lead Performance Max type, in phases through 2027. Export your historical campaign reporting before the account moves, because it doesn't migrate, and treat listing hygiene as a paid dependency now that the campaigns sync from Google Business Profile in real time.
Two more programs earn their place. A geofenced grand-opening package of display, digital out-of-home, and connected TV deploys automatically when the brand learns of an opening, which is a predictable trigger and a standard build. A weather and events response system deploys pre-built creative when conditions are met, which most franchisees lack the bandwidth to run in real time, so the brand should run it for them. One program doesn't work at scale: nationally managed social accounts for individual locations, because each account has too few followers and too local a content need for central management to beat benign neglect.
Location-level AI visibility is the new row, and it will come up at a franchise advisory council meeting in 2027 whether you've prepared for it or not. Centralize the data layer and the tooling, because the listing platform, the structured-data standards, the menu feed, and the measurement are single systems that benefit from one owner, and localize the human judgment, because review responses and local photography need a local voice and local knowledge. Then name it in the allowed-use list before a franchisee who already pays a local listings vendor asks whether that's reimbursable and who's responsible when the two disagree.
Figure 5
What to centralize, and what to localize.
Activity
Centralize
Localize
Why
National brand campaigns (CTV, out-of-home, audio)
Yes
No
Scale and consistency exist only at the national level
Local digital playbook (Google local, Performance Max)
Yes
Opt-in
The brand manages, the franchisee contributes and benefits
Location records, structured data, and menu feeds
Yes
No
One system, one standard, and it now gates both organic and paid local performance
Review response and local photography
Tooling
Yes
The brand provides the platform, the franchisee supplies the local voice
Grand-opening packages
Yes
No
Predictable trigger, standard package, brand efficiency
Weather and event responses
Yes
No
Real-time triggers, and franchisees lack the capacity to execute
Local social media accounts
No
Yes
Local voice, hyperlocal content, low scale
Loyalty program
Yes
No
Single system, single dataset, brand-managed
Treat this matrix as the starting point for a conversation with your franchise advisory council rather than a verdict.
The plan
The next 90 days.
A sequence for a CMO who's read this far and is wondering where to start. It builds evidence before it asks for money, which is the order that works when budgets are frozen. Your progress is saved in this browser only.
0 / 12 steps complete
Days 01–30
Audit your measurement stack.
Map every conversion action across every channel and find the gaps, because most brands discover that thirty to fifty percent of their reported conversions fire incorrectly or not at all. The audit costs nothing and tells you more about your program than any other single activity.
Days 30–60
Run the prompt panel.
Twenty-five prompts, ten markets, three assistants, scored for appearance and for primary recommendation, which gives you a finding you can take to a board meeting rather than a media test you wait a quarter to read.
Days 60–90
Build the franchisee dashboard, even roughly.
Use it once with your franchise advisory council and watch the conversation change, because a rough version used once is worth more than a polished version built for nobody.
Where the three to five percent comes from.
Three to five percent of a brand-level media budget funds everything in this guide, and the honest part is what you stop. These are the four places we most often find the money, and the specific mix depends on your footprint, so run the exercise before the budget locks.
Figure 6
The 2027 reallocation.
Stop or reduce
National linear TV weight in your least concentrated markets
You're paying to reach households outside any trade area, and CTV can buy the same reach addressably.
What it funds
Location records, structured data, and menu-feed work
Stop or reduce
Marketplace sponsored placement above your defensive cap
The return-on-ad-spend figure flatters itself by measuring the last click before an order that was already coming.
What it funds
One channel pilot with footfall measurement designed in advance
Stop or reduce
Retainers on channels with no incrementality read
If nobody can tell you what the channel produced incrementally after a year, that is the answer.
What it funds
The incrementality method itself, which then prices every other decision
Stop or reduce
Print and legacy lines nobody has defended in two budget cycles
Print can work, but it is almost never the thing being measured.
What it funds
The franchisee dashboard
The record
What we got wrong.
We publish this guide with opinions in it, so some of them don't hold. A guide willing to show its record is worth more than one that quietly edits its old calls, so here are the four worth naming.
We wrote a dollar floor for connected TV measurement when we should have written a design requirement.
Five thousand dollars a month works for a geographically concentrated brand and fails for one spread across forty-seven markets, where the same spend won't reach significance in any single trade area. The right question is how many exposed households per trade area the design needs, and whether your footprint can supply them.
We treated first-party data as a prerequisite for AI, and it's now a prerequisite for access.
We said thin data would make your bidding less efficient. What has actually happened is that brands unable to supply clean server-side signal are being excluded from inventory and newer campaign types, not merely charged more. We understated the stakes.
We didn't see assistant-led discovery arriving this fast.
A year ago, AI in this category meant bidding, allocation, and creative production, and within a year it also meant distribution. Two of this edition's chapters exist because of a shift we had filed as a bidding story.
We were right about content-recommendation networks, and we'd say it again.
The one call we're happy to reprint without amendment.
Into 2027
What we expect in 2027.
Four calls we expect to hold by the time the 2027 budget locks, drawn from the shifts this edition tracks. Each one carries a confidence level, so you can hold us to it in the next edition.
01High confidence
Assistant-led discovery keeps accelerating.
The brands that haven't made their menu and location data machine-readable will compete for a shrinking share of recommendations, and the ones that have will pull ahead at the exact intersection a regional operator can win, which is a specific occasion in a specific place.
02Moderate confidence
Thin data stops costing efficiency and starts costing access.
As pay-per-lead automation, ads inside AI answer surfaces, and minimum-return auto-bidding spread, a brand that can't supply clean server-side signal is increasingly shut out of the inventory rather than merely charged more for it.
03High confidence
The Local Services Ads migration completes through 2027.
Franchisee-maintained listings become a brand-wide paid dependency, so the record a franchisee updates (or doesn't) now decides what your local ads can say as well as whether an assistant recommends you.
04Lower confidence, high consequence
Location-level AI visibility lands on the advisory council agenda.
The question of who owns it gets asked in a 2027 council meeting whether you've prepared for it or not, and the brands that have already centralized the data layer will answer it instead of opening it.
Frequently asked.
Why are our sales up but our traffic flat?
Most of the increase is menu price rather than visits. Restaurant sales are tracking up about four percent in 2026 while real growth sits near one percent, with traffic flat to down across most segments. The fix is to report same-store sales decomposed into transactions, price, and mix, so marketing is measured on the transactions it actually owns instead of taking the credit or the blame for a number it only partly moved.
How do we tell whether we show up in AI search?
Run a prompt panel. Pick your ten largest markets, write twenty-five conversational prompts a real customer would use, run each in ChatGPT, Gemini, and Perplexity, and score whether your brand appears and whether it holds the first recommendation. Report two percentages by market, re-run monthly, and treat the series as your baseline. It's the first AI-visibility number a franchisee will understand and act on.
Should we build ChatGPT or Claude ordering now?
Build the feed now, and expect the volume later. Order volume through assistants is small today, so don't move a media budget for it this year, but the fee structure is the best in the off-premise mix and the build takes longer than the demand takes to arrive. Keep discovery in the assistant and checkout in your own app, which is the decision Starbucks, Little Caesars, and Square all made in 2026.
Which channels should a 200-unit brand fund first?
Connected television with footfall measurement written into the plan, digital out-of-home for geo-targeted presence, and streaming audio or host-read podcasts targeted by audience rather than by topic. Size connected TV by exposed households per trade area rather than by a flat dollar floor, and pair it with out-of-home for frequency. The one channel we advise against scaling is paid placement on content-recommendation networks.
What's the right north-star metric for franchise marketing?
Incremental same-store sales lift per dollar of marketing spend, measured at the unit level, attributed by channel, and rolled up to the brand. It's the number the CFO and the franchisee both already care about, and the one the marketing team can directly influence. Building it takes a unit-level sales feed, a unit-level spend allocation that includes local co-op dollars, and a geographic holdout method for channels with no native measurement.
What's changing with Google Local Services Ads?
Google began moving Local Services Ads into Google Ads as a pay-per-lead Performance Max campaign type in August 2026, in phases through 2027. The ads still run on Search and Maps and still charge for valid leads, but the old dashboard goes away, historical reporting doesn't migrate (export it first), and the campaigns now sync from Google Business Profile in real time, which makes listing hygiene a paid dependency as well as an organic one.
How should we split marketing between corporate and franchisees?
Centralize the systems that benefit from one owner and localize the work that needs a local voice. Corporate should own national campaigns, the local digital playbook, the listing and menu data layer, grand-opening packages, and loyalty, while franchisees own review responses, local photography, community sponsorships, and their own social accounts. Publish the split as an allowed-use list before the advisory council asks.
Where do we find the budget for all this?
Three to five percent of a brand-level media budget funds everything in this guide. We most often find it by reducing national linear TV in your least concentrated markets, capping marketplace sponsored placement above the defensive line, cutting retainers on channels with no incrementality read, and dropping print or legacy lines nobody has defended in two budget cycles.
The full edition
Take the whole guide into your next budget meeting.
Thirty-six pages, second edition, with the full channel stack, the aggregator renegotiation playbook, the five-audience reporting matrix, and every figure at full size. No form, no gate. The PDF downloads from every button on this page.
Happy Cog has built and run marketing programs for franchise and multi-location restaurant brands since 1999. Our paid media and performance practice runs media, search, AI visibility, and measurement for brands with two hundred or more units. Nice people with interesting problems. If the questions in this guide describe your situation, the next thirty minutes with us is probably worth your calendar. Tell us what your same-store number is doing and where you think the gap is, and we'll tell you plainly what we'd look at first.
