# The AI Operating Model for Franchise and Multi-Site Operators

- Link: https://xanadu.co/ai-operating-model-franchise/
- Published: 2026-07-27T13:16:44+04:00
- Author: Daragh McGrath

A promotion goes live on the first of the month. Head office ships the artwork on
the twenty-eighth. By the fifth you have forty versions of it out in the network:
the old logo lockup on three, last quarter’s price on a handful more, Arabic that
has been run through a free translator and reads like a warning label, and somewhere
a light box displaying a JPEG that was compressed by WhatsApp four times before 
it got printed.

You cannot fire anyone for it. The people producing it do not work for you.

That is the structural fact of a franchise business and it is the thing every AI
conversation in this sector skips. You own the brand. You do not employ the people
who put it in front of customers. Every other industry solving an AI adoption problem
gets to mandate the tooling and mandate the training. You get to negotiate.

So the standard AI playbook, roll out a platform, train the staff, enforce a policy,
does not survive contact with your operating model. Not because franchisees are 
difficult, but because they are independent businesses with their own P&L, their
own staffing and their own priorities, and your franchise agreement is a commercial
contract, not a management line.

This playbook is about what works instead. What an AI operating model is, the four
dimensions that decide whether AI sticks in a multi-site business, how each one 
fails when you do not employ the operator, and a 90-day sequence you can start without
a network-wide mandate.

## What an AI operating model actually is

An AI experiment is one person doing one task faster. An AI operating model is the
business doing work differently, on purpose, at scale, in a way that survives staff
turnover and does not depend on goodwill at site level.

The two look identical for a quarter and then diverge completely. The experiment
produces a good story for the next franchisee conference. The operating model produces
a change in cost, speed or consistency that shows up in a number you already report.

An operating model answers four questions that experiments never have to:

**Who owns the outcome?** Not who owns the tool. Who is accountable for the thing
AI is meant to improve, whether that is time from promotion sign-off to live in 
every site, cost per campaign, or the share of network assets that are actually 
on brand.

**What is the work, redesigned?** Not the existing process with a faster step in
the middle. The process rebuilt around what is now cheap. If a seasonal campaign
still goes out as a zip file to a distribution list and comes back as forty interpretations,
you have not redesigned anything.

**Who can actually do it?** Not head office marketing. The people at site level 
producing local assets on a Thursday afternoon, who did not go to your conference
and will not read your deck.

**What is allowed?** Which tools are sanctioned, what data can go in, what must 
stay out, and, uniquely in your case, what you can actually require of an independent
operator versus what you have to make so easy they choose it.

Miss any one and AI stays a head office hobby. Get all four working and it becomes
the infrastructure your network runs on.

Only one of the four is really a technology question. The other three are leadership,
capability and control. In a franchise business the fourth is not a compliance footnote.
It is the product.

## The four dimensions

The four dimensions below are the same ones our [AI readiness assessment](https://xanadu.co/ai-readiness/)
scores against, in the same order. Read each one, be honest about where you sit,
and take the 30-day action even if you do nothing else this quarter.

### 1. Leadership

**What good looks like.** The exec team uses AI in their own work most weeks, visibly.
AI is owned by a business leader with an outcome attached rather than by IT or a
project group. And at least one leader has built something with it, an agent or 
a connected workflow, rather than only ever having a conversation with it.

**How it fails in franchise.** AI gets owned by marketing at the franchisor and 
never crosses the network boundary. That is a harder failure here than in any other
sector, because the boundary is legal and commercial rather than just organisational.
Head office gets more capable every quarter. Site level gets nothing. The gap widens,
and it widens fastest in exactly the places where the brand is most exposed.

Then there is the conference problem. AI turns up as a slide at the annual franchisee
event, gets a round of applause and a photo, and is never mentioned again until 
the next one. Franchisees have seen a lot of head office initiatives arrive that
way. They have correctly learned that most of them do not survive twelve months,
so they wait, and the waiting is rational.

The third failure is quieter and more expensive. Because you cannot mandate, leadership
concludes there is nothing to lead, and defaults to producing more central assets
faster. That is a real gain and it is a fraction of the available one, because the
volume problem in a franchise network is not central production. It is local adaptation.

**Do this in the next 30 days.** Pick one outcome you already report, time from 
campaign sign-off to live across the network, cost per local campaign, or the share
of network assets on brand, and put a named leader on it with AI as the lever and
a quarter to show something. One owner, one number. Then, personally, build something.
Connect Claude to a report you hate producing and watch where it breaks. Two hours
of that will change what you ask for more than any vendor briefing, and in a business
where you cannot compel anyone, knowing precisely what the tools can do is what 
stops you asking for the impossible.

### 2. People and skills

**What good looks like.** Capability sits broadly rather than in a few enthusiasts,
people have had structured role-specific training built around what they actually
do, and enablement is ongoing because the tools move every quarter.

**How it fails in franchise.** Every other sector treats training as a project. 
In multi-site operations it is a flow problem, and that difference breaks most rollout
plans.

Your site-level turnover is high. It is meant to be. So any capability you build
by training individuals starts decaying the day you build it. Train forty site managers
in March and by November a meaningful share of them have moved on, taking the capability
with them and leaving you with a slide that says the network is trained. Anything
that depends on trained people staying put will quietly unwind, and it will unwind
invisibly, because nobody reports the decay.

The second failure is the mandate you do not have. You cannot compel an independent
operator’s staff onto a training programme. You can offer it, and offering it competes
with everything else on a site manager’s Tuesday.

So the answer is not a bigger training programme. It is to move capability out of
people and into tooling. If the on-brand path is also the fastest path, a brand 
new hire who has never heard of your AI programme still produces the right thing
on their first day. That is the only form of enablement that survives turnover, 
and it is a design problem rather than a training one.

**Do this in the next 30 days.** Take your highest-volume local task, and in almost
every network that is producing a local promotional or seasonal asset, and make 
the compliant version of it the easy version. Locked templates in a controlled brand
kit, correct assets already in place, prices and legal lines where they belong, 
so the fastest route to a finished thing is also the on-brand route. Then write 
ten prompts for that specific job, tested against your real brand and your real 
menu or product set, and put them where the work happens rather than in a portal
nobody opens. Measure how long the task took before and after, at one site, honestly.

### 3. Process and workflow

**What good looks like.** AI shows up in real workflows rather than ad-hoc chats,
ideally across several core processes, and at least one process has been genuinely
redesigned rather than accelerated, running in production rather than sitting in
a pilot report.

**How it fails in franchise.** Task acceleration mistaken for transformation, in
the most expensive possible place. Your central studio now produces the campaign
master in two days instead of five. Real, and it is the wrong end of the process.
The campaign master was never the bottleneck. The bottleneck is the two hundred 
local adaptations that happen after it, across sites, in languages, at different
sizes, by people with no design training and no time.

Then there is everything nobody has looked at, because the AI conversation stopped
at marketing. Menu and price change propagation across every site, every channel
and every format. Franchisee onboarding packs. Compliance and audit documentation.
Food safety records. Mystery shopper reporting. Supplier and operations manuals 
that get revised centrally and then have to reach every operator in a form they 
will actually read. All high volume, structured, rule-governed, and currently eaten
by people or, worse, not done consistently at all.

Localisation deserves its own line, because in this region it is where most brands
quietly leak. Running your English campaign through a translator gets you Arabic
that is technically accurate and commercially dead. That is not a translation problem,
it is a [localisation problem](https://xanadu.co/translation-changes-the-language-localisation-changes-the-communication/),
and it is one of the clearest places AI pays in a Gulf network operating across 
two languages at volume.

This is where the two layers earn their keep, and in a franchise business the creative
layer carries more of the load than in any other sector. Canva is [the creative and communications layer](https://xanadu.co/services/canva/),
where the locked template, the brand kit and the controlled local adaptation live,
so a site can produce something fast without producing something wrong. Claude is
the reasoning, workflow and agent layer, where the price change gets propagated,
the compliance pack gets drafted, the local copy gets written and adapted, and the
audit gets summarised. Prompt to presentation. The value is not in either layer 
alone. It is in connecting them to a network that does not report to you.

**Do this in the next 30 days.** Map one process on one page. One hour, not a workshop.
Campaign sign-off to live in every site, or price change approved to correct on 
every menu and channel, or new franchisee signed to open. Mark every handoff, every
wait and every re-key. Do not mark where AI could help. Mark where the time actually
goes and where the errors actually enter. In a multi-site business the answer is
almost always the last mile, not the first.

### 4. Governance and data

**What good looks like.** People have sanctioned tools and clear guidance rather
than working it out themselves. There are real rules about what data goes in and
what stays out. And leadership is confident, not hopeful, that the network can use
AI without putting the brand or customer data at risk.

**How it fails in franchise.** In the other verticals in this series, governance
is the wedge. Here it is the whole proposition, because brand consistency across
sites you do not own is not a governance overhead. It is the thing a franchisee 
is paying you for.

And AI has just made the failure mode effortless. Generating a plausible-looking
promotional asset used to require a designer, which was itself a control. Now anyone
at any site can produce something in ninety seconds that looks close enough to your
brand to go out and wrong enough to cost you. Multiply that by every site and every
promotion. The cost is real and almost nobody measures it, which is the subject 
of a longer piece on [the cost of off-brand content nobody is measuring](https://xanadu.co/the-cost-of-off-brand-content-that-nobody-is-measuring/).

The legal position makes it sharper. A rogue asset at one site is, contractually,
somebody else’s business. Commercially, it is entirely yours. The customer does 
not know which locations are corporate and which are franchised, and neither does
the regulator when the claim on the poster is not one you can stand behind.

The data question is real too. Customer data, loyalty data, supplier terms and franchisee
commercial performance, sitting across systems in businesses you do not control,
with staff who have never been told what may go into a free AI tool.

So the honest answer on governance in a franchise network is that policy alone will
not carry it. Publishing a rule you cannot enforce is theatre. What works is a combination:
a short written standard that says what is sanctioned and what data never goes in,
and, doing most of the actual work, a controlled brand system where the easy path
is the compliant path. You are not policing behaviour. You are removing the reason
for it.

**Do this in the next 30 days.** Write a one-page AI and brand standard for the 
network. Which tools are sanctioned, what data never goes into an unsanctioned tool,
what is fixed in the brand system and what a site may adapt, and who to ask when
it is unclear. One page, in the language of the franchise agreement, not the language
of IT. Publish it with your name on it. Then lock the brand kit and the templates
so the fastest route to a finished asset is the correct one, and audit one promotion
across the network to get a real number on how far the drift currently goes. That
number is the business case for everything that follows.

## The journey: five steps, scoped one at a time

Once you know where you stand, sequence matters more than ambition. One path, five
steps, each scoped and bought on its own.

**Step 1. Assess.** A free self-serve assessment across the four dimensions. Three
minutes of honesty before anyone spends anything.

**Step 2. Blueprint.** The plan, the licence case, and a date. The whole business
scanned for where AI actually pays, prioritised into a heatmap, then one function
taken deep: every valued task mapped to a specific recommendation, the blockers 
named honestly, a right-sized licence plan, adoption targets set before anything
is deployed, and a business case your CFO can underwrite. In a franchise network
the first function is almost always marketing and local campaign production, or 
the operational end of menu and price propagation. Three to four weeks. One function
deep beats twenty shallow, because depth is what makes a plan executable.

**Step 3. Enable.** The go-live. Blueprint is the strategy, Enable is the launch:
deployment and configuration, the project run properly, a change campaign your network
actually notices, role-based training built around real workflows, prompt libraries
for each team, named champions, and adoption tracked against the targets you already
set. Most AI deployments [fail quietly](https://xanadu.co/most-ai-projects-quietly-fail-heres-how-to-know-if-yours-will/).
Licences get bought, a pilot impresses, usage fades within a quarter. In a franchise
network the launch also has to answer the question every operator will ask, which
is what this does for their P&L rather than yours.

**Step 4. Govern.** Risk, brand and spend under control as usage scales. A usage
baseline covering every AI tool in use, sanctioned or not. A data exposure inventory.
A spend baseline and cost guardrails. A governance workshop with leadership, IT 
and legal. A usage policy and brand standard written to your sector and your jurisdiction.
A board-ready readout with a remediation plan.

For most franchise businesses this is the step to start with, not the third one 
you get to. Your brand is your product, your control is contractual rather than 
managerial, and AI has just handed every site a content factory. Govern is sold 
standalone for exactly this reason, and a governed network is a faster network to
deploy into.

**Step 5. Build.** Agents in production, function by function, off the heatmap. 
The price propagation, the localisation pipeline, the compliance pack generation,
the connected route from campaign master to correct local asset in every site. Built,
tested, deployed to real users, with a runbook and a handover. Eight to twelve weeks
per function.

**Executive coaching runs alongside any of it**, for the leader the board is asking.
Five sessions, on Claude personally, building your own agents around how you actually
work. It is the fastest way to fix a Leadership score, and a low Leadership score
caps everything else.

Why one step at a time: each has to earn the next, and in a franchise business that
discipline does double duty. Every step that produces a visible result at site level
is a step that makes the next one easier to sell into the network. You are not just
de-risking the investment. You are building the internal case with people you cannot
instruct.

You can read more about [how the journey works in practice](https://xanadu.co/services/ai-transformation/).

## The 90-day starting sequence

None of this needs a network-wide mandate. Start it centrally and let the results
travel.

**Days 1 to 30. Get honest and get an owner.** Score yourself across the four dimensions.
Take the weakest, not the most interesting, because the weakest is your ceiling.
Name a leader with one operational outcome and one number. Publish the one-page 
AI and brand standard. Audit one promotion across the network and get a real number
on brand drift. Have one leader build one thing personally. Baseline everything 
now, before anything changes.

**Days 31 to 60. Prove it at a handful of sites, properly.** Do not go network-wide.
Pick five or six sites, ideally a mix of corporate and franchised, and rebuild one
process end to end for them: locked templates, the prompt library, the local adaptation
path made fast and correct. Corporate sites let you move quickly, franchised sites
prove it works where you cannot mandate. Keep the scope small enough to finish inside
thirty days.

**Days 61 to 90. Show the number and let the network pull it.** Report before and
after using something you already reported: cycle time, cost per local campaign,
share of assets on brand. Then show it to the network, with the participating operators
in the room rather than a head office presenter. A franchisee will believe another
franchisee’s numbers. That is the whole mechanism, and it is why the pilot site 
mix in the previous step matters more than the technology in it.

The point is not the 90 days. It is that at the end you have an owner, a standard,
a controlled brand system, one proven process and a set of operators who will vouch
for it. In a business you do not control, that last one is the asset.

The same four dimensions play out differently by sector. In [real estate](https://xanadu.co/ai-operating-model-real-estate/),
the constraint is a distributed frontline with no training and no rules. In [higher education](https://xanadu.co/ai-operating-model-higher-education/),
the committee absorbs AI as a risk to contain rather than an opportunity to run.
The pattern holds; only the failure modes change. Underneath all three sits the 
same operating model for the content itself: the four tiers and five control points
set out in [the AI Content Supply Chain](https://xanadu.co/ai-content-supply-chain/).

## Where to start

You probably already know which of the four is your weak spot. Most operators do.
But a rough sense is not a straight answer, and a straight answer is what you need
before you take anything to the network.

So we built a way to get one.

Nine questions, three minutes. An honest score of where your business stands and
the single next step that will move you fastest. It scores you across the same four
dimensions this playbook is built on, names your biggest gap, and tells you the 
fastest way to close it.

It is not a quiz that tells everyone they are doing brilliantly. Some of the results
are uncomfortable. That is deliberate, because an uncomfortable accurate number 
is worth considerably more than a flattering vague one.

**[Take the AI Readiness Assessment](https://xanadu.co/ai-readiness/)**

Whatever you score, you will know where you stand and what to do next. Which, given
how most AI initiatives in this sector end up, already puts you ahead.

## Frequently asked questions

### What is an AI operating model in a franchise business?

An AI operating model is the way a business runs work with AI on purpose and at 
scale, rather than individuals using AI ad hoc. It answers four questions: who owns
the outcome, how the process is redesigned rather than just accelerated, who is 
capable of doing the work, and what is allowed with which data. In a franchise network
there is a fifth consideration running through all of them, which is that you cannot
mandate behaviour at sites you do not own, so the model has to work through tooling
and incentives rather than instruction.

### How do you roll out AI when franchisees are independent businesses?

You do not mandate it, because you cannot. Start centrally, prove one process at
a small mix of corporate and franchised sites, and report the result in a number
the operators care about, which is their own margin and their own time rather than
head office efficiency. Then let franchisees who took part present it to the network.
Adoption in a franchise system travels operator to operator far better than it travels
top down.

### How do you keep brand consistent across franchised sites now that AI makes content easy to produce?

Policy alone will not do it, because a rule you cannot enforce is theatre. What 
works is making the compliant path the fastest path: locked templates and a controlled
brand kit where the correct assets, prices and legal lines are already in place,
so a site produces the right thing by default. Pair that with a one-page standard
naming what is fixed and what a site may adapt, and audit one promotion across the
network to get a real number on current drift.

### Can AI help with Arabic and English localisation across a Gulf network?

Yes, and it is one of the clearest returns available to a multi-site brand in this
region, provided you treat it as localisation rather than translation. Running English
campaign copy through a translator produces Arabic that is accurate and commercially
flat. Adapting it, with your brand voice, local context and the offer intact, is
the job, and it is exactly the kind of high-volume, judgement-light-but-not-judgement-
free work that AI does well when it is given proper brand context.

### How long before AI delivers anything real in a multi-site business?

You can have one process rebuilt, a handful of sites running it and a measurable
before-and-after inside 90 days without a network-wide programme. Scale takes longer,
and it depends less on the technology than on whether a named leader owns an outcome,
whether capability is built into tooling rather than into individuals who will turn
over, and whether the brand system makes the correct path the easy one.