The Franchise AI Maturity Model: Five Stages You Can Test This Week

The franchise AI maturity model runs from a network nobody has mapped to one where the system tells people things before they ask. Each stage is defined by an observable you can check this week — not an adjective, a percentage or a timeline — and the step out of it is smaller than the stage sounds.

Maturity models are usually a way of selling the top rung. This one is written to be used against its author, which means every stage has to be verifiable.

A stage has to be a test, not an adjective

Most published ladders here fail on the same three things, worth naming so you can reject mine on the same grounds.

• Adjectives instead of observations. "Experimental", "strategic", "transformational". Nobody has ever placed their own network with those words, because they describe a mood.
• A distribution nobody could have measured. No register of where franchise brands sit on any AI ladder exists. FRANdata's forecasting model tracks approximately 4,000+ US franchise brands, on the IFA and FRANdata outlook methodology, and none of it grades AI posture. There is no honest way to say where most brands sit, so I will not.
• A timeline. Stages arrive when the underlying work is done, and that work is document work of unpredictable length.

What survives is narrower and more useful: for each stage, one thing that is either true of your network or is not, and a test taking minutes rather than a quarter. Run them before reading the names — the stage you would have picked for yourself is almost always one rung above where the test puts you.

The franchise AI maturity model, stage by stage

One: unmapped. Observable: nobody at headquarters can say which AI tools the network uses, or whose accounts they sit in. Test: ask two field coaches separately to list what their operators use for brand work. Blank lists, or two that do not match, put you here — and this is not a stage of low usage but of high, invisible usage.

Two: sanctioned but ungrounded. Observable: there is an approved tool and it answers from general knowledge rather than your material. Test: ask it something whose answer exists only in your manual — a threshold, a named exception, the rule written after the incident in March. A fluent, plausible, uncited answer is the tell. This rung feels like progress and is the most dangerous of the five, because an answer about businesses like yours reads exactly like an answer about you until somebody acts on it.

Three: grounded. Observable: an answer names the section it came from, and the system declines when your material is silent. Test: ask something the brand has deliberately never written down. If it improvises, you are at stage two with better marketing. Then ask the same question from two markets and check that the same section comes back.

Four: scoped, metered and recorded. Observable: three artefacts headquarters can produce without calling a vendor. Test: ask for all three, and time it. What a shift lead is answered against what an owner is answered on the same question. Consumption by location last month. What the network was told about a named policy on a named date. Ten minutes is stage four; a week is stage three with a roadmap.

Five: compounding. Observable: the system tells somebody something they did not ask for, and they read it. Test: count the unprompted notices last week that changed what someone did. Then the harder one: a question the corpus could not answer produced an edit to a document. If nothing in your material changed because of what the network asked, the loop is open and the top rung is decoration.

The step out of each stage is smaller than the stage sounds

Each move is one piece of work, and only one is a purchase.

Out of one: find out. An anonymous survey, an expense scan and a conversation with managers rather than owners — an audit of actual use, and it takes a week. What it produces is not a compliance finding but a list of questions your manual failed to answer.

Out of two: the documents, not the software. One current version of everything, the appendix graveyard cleared, a page saying which document wins. Brands skip this because it looks like admin, and it is the whole difference between the second and third rungs — the checklist is getting the documents ready.

Out of three: decide who sees what, what a question costs and what is kept. The third arrives late and unwelcome, because a recorded network is a network with a file — the questions in franchise communications discoverability belong here rather than after a matter starts.

Out of four: connect a number to the documents, then decide what is worth interrupting someone for. Push without a threshold is noise, and noise is how a network learns to ignore you.

Out of five: nothing. There is no sixth rung. The work becomes maintenance, less satisfying to plan and the only reason it holds.

Where the ladder lies to you

Four ways, the last of them about me.

Stage three is a legitimate destination. A brand whose material is clean and whose answers cite sections is well served and may never need the rest. Ladders imply obligation; this one does not.

Order beats height. Stage five reached without stage four — a system pushing unprompted, unscoped and unrecorded — is an unattributed message reaching the wrong people at scale. Worse than stage one, because it carries your name.

Your network is not one climber. Multi-unit operators are 19.3% of US franchisees and hold 58.8% of franchised locations, on FRANdata's research — so most of your estate belongs to companies with their own systems and often another brand's ladder to climb, the part of multi unit franchise ownership trends that decides whether a rollout lands.

The budget runs ahead of the rung. Three in four franchisors expect to increase capital spending on technology and innovation, while a narrower 28% mentioned incorporating AI and increased automation, on the FRANdata and IFA survey. Money moves before capability, which is how a brand at stage one ends up owning stage five.

And the disclosure: I sell in this category, so read the ladder as written by an interested party. The defence is that these are the tests I would want run against us, and stage four is where a vendor produces artefacts rather than describing them.

Every rung is a fact about your documents and your roster before it is a fact about your software, which is why buying the top of the ladder does not move you up it — and why a brand at stage three having bought nothing sits higher than most that have. The question underneath is still governed AI for franchises, answered with a corpus and an ownership table long before a contract.

Every rung above the first rests on one prior argument: whether a network can govern AI at all.