The Real Cost of Ungoverned AI in Franchising, in Three Lines

The cost of ungoverned AI in franchising is rarely a single invoice. It is untracked subscription spend spread across independent operators, operational variance created when locations get different answers to the same question, and the slow export of the documented system a buyer would one day pay a multiple for.

I have never watched a franchisor produce the cost of ungoverned AI in franchising as a number. Not because the estimate is hard — it takes an afternoon — but because nothing in the accounting system asks for it. No vendor to negotiate with, no renewal date to force the question.

So here is a model, deliberately crude. If the term is new, start with the four conditions governed AI has to meet; this post is only about what the alternative costs.

Why it never arrives as a bill

Every other technology cost in a network announces itself. Someone signs a contract, approves an invoice, writes a budget line and defends it next year. This one does none of that, for reasons structural rather than sneaky.

• The default price is zero. A free tier that answers a policy question in ten seconds never produces an expense report.
• The buyer is an independent business. When a franchisee does pay, they pay from their own P&L, where you have no visibility and limited standing to ask.
• It is bought in twenty-dollar increments. No single purchase crosses an approval threshold. The aggregate is the only interesting number and nobody holds it.
• The benefit is immediate and private. The person who gains is the person who bought it, so there is no internal advocate and no critic either.

Meanwhile the budget process looks elsewhere. 75% of franchisors say they expect their technology and innovation capital spending to increase, per FRANdata's franchisor technology research. Capital spending is planned, approved and visible. What I am describing is none of those, which is why it shows up in no survey and no forecast.

Line one: the seats nobody counted

Take a 40-unit network and load it with conservative assumptions.

Assume one person in half your locations pays for a consumer AI plan at $20 a month, and another six people at headquarters do the same. That is 26 seats, $520 a month, $6,240 a year — spent on your brand's work, by people doing your brand's work, in accounts nobody at headquarters administers.

Notice what that figure is not. Not a saving you capture by banning anything, because the work those seats do still needs doing tomorrow. Not a cost you can invoice, because most of it sits on franchisee P&Ls. And not stable, because the assumption most likely to be wrong is the low one.

What it is good for is a comparison. Before you evaluate a single vendor, your network is already spending that much, unmanaged, on the category you have not decided to buy. Any procurement conversation that opens at zero opens on a number you know to be false — which is the dynamic underneath the franchise operations dashboards nobody opens. People pay, in money or attention, for whatever answers fastest.

Run the line with your own unit count and your own guess at penetration. If the guess is uncomfortable, that is itself a finding.

Line two: the variance that reaches the floor

This is the line that turns a security memo into a business case. Almost nobody computes it.

An ungoverned answer is rarely wrong in an obvious way. It is plausible, confident, uncited, and assembled from how businesses like yours generally operate rather than how yours does. Ask the same operational question from four private accounts and you get four defensible replies, none of them your policy and none traceable afterwards.

Put a price on it. Assume two locations a quarter act on something that is not your standard — a discount threshold, a scheduling rule, a hold time. Each costs a field consultant a day to find and unpick, and the location spends a fortnight running it the wrong way first.

Eight consultant days a year at a fully loaded $600 a day is $4,800, before counting whatever the location gave away while it was wrong. For 40 units where nobody is looking, two a quarter is not pessimistic.

What makes this line matter more than its size is what it does downstream. Each episode is invisible while it happens and looks like an execution failure when it surfaces, so it gets managed as a people problem. You pay either way. The variable is whether you know what you are paying to fix.

Line three: the asset with no invoice

The third line cannot be computed, and I would rather say so than invent a figure.

What sits on it is the documented system itself. Advisers price emerging franchisors — under about $3M of EBITDA — in the 5–9x band, and scaled systems above $10M at 10–16x and higher. That franchise business valuation multiple attaches to a royalty stream, and what keeps the stream durable is a written way of operating every market executes the same way.

That document is the only asset in franchising deliberately handed to hundreds of independent businesses in a format built to be read, copied and followed. Trade-secret protection generally rests on demonstrating reasonable efforts to keep a thing confidential. Ask your franchise counsel what that means when sections of the manual are pasted, week after week, into accounts you cannot name.

So line three is a blank with a question mark beside it. Leave it blank. A blank is the row that makes a leadership team argue about what number belongs there, which is a more useful hour than any figure I could supply.

What the cost of ungoverned AI in franchising adds up to

On those assumptions the two computable lines total $11,040 a year across 40 locations. About $276 a location. Nobody restructures a business over $276 a location.

That is the right reaction and it is also the trap. The two lines you can price are the two that matter least, and the output was never the total. It is the shape: a small measurable cost, an unbounded unmeasurable one, and nothing in the business that would tell you if either grew tenfold before Christmas.

Four inputs make the estimate yours rather than mine:

1. Penetration. Ask general managers what their teams use, not what is permitted.
2. Seat price. Consumer plans, business plans, and whatever your marketing agency bills back.
3. Variance episodes. Ask field consultants how often last year they found a location running a rule nobody at headquarters wrote.
4. The blank. Whatever your counsel says about reasonable efforts.

None of it requires buying anything. It moves the conversation from security posture to budget line — which is the practical argument for governed AI for franchises. Not that the ungoverned version is dangerous, but that it is already a cost centre with no owner, no meter and no line on anybody's plan.

Meanwhile the sanctioned budget keeps funding the franchise operations dashboards you already pay for and nobody opens.