Your AI Vendor Looks Like a Franchisor

The pitch usually arrives in a conference room. The supplier will put their own engineers inside your building, on your floor, working your problems. They bring the system: the model, the standard operating procedures, and the orchestration layer — the software that decides which task runs when, with what data, and what happens when a step fails. You bring the site. Proprietary data, domain expertise, business context, the authority to actually do anything with the output. Suppliers have collectively committed billions of dollars to this arrangement over the past few months, and every announcement lands on the same recognition for me. I have seen this economic shape before, and it wasn't in software procurement.

In a typical U.S. franchise, the franchisee pays maybe $40,000 upfront and then remits roughly 8% of gross revenue for the life of a 10-to-20-year contract. The entry fee is the cheap part by an enormous margin. Enterprise AI isn't priced as a revenue royalty, not yet anyway, but franchise economics keeps asking a question that enterprise buyers mostly skip: across the whole length of the relationship, who captures more of the value that both parties had to be present to create?

Franchise research draws a clear line between portable assets and relationship-specific ones. A prime location holds its value no matter whose sign hangs over the door. A building fitted out to one franchisor's exact spec, with their equipment and staff trained on their procedures, is worth considerably less the morning after the agreement ends.

Enterprise customers tend to assume their proprietary data is the prime location here, valuable no matter whose system sits on top of it. That's partly right. But the operating knowledge you generate during deployment — the tuned workflows, the exception handling, the orchestration configs — behaves like the fit-out. Its value is bound to the supplier's system. The Australian government's guidance on agent architecture says the quiet part plainly: models get swapped, and the orchestration and control layer is the investment that sticks around. If that's where the durable value lives, and the supplier owns it, your data advantage weighs less than you think.

Then there's the drift in who sets the rules. A longitudinal study of over 2,700 franchise chains found that across fifteen years, franchisors steadily tightened restrictions on sourcing, pricing, product offerings, and post-term competition, without trimming fees to match. Exclusive territories went from a majority of chains down to about 20%. Whether AI suppliers follow the same path is an open question. The incentive, though, is identical. Once your operations run on someone else's system, new requirements can be added without concessions, because the cost of switching has already been paid and nobody pays it twice on purpose.

And in franchise relationships, that cost is not primarily contractual. Your staff knows their system, your workflows are shaped around their tools, your processes assume their orchestration. I have run the migration off a vendor's stack, and the contract language was the easy part. Strike every restrictive clause and leaving is still brutally expensive, because the business grew into the shape of the relationship.

I'm not predicting territory maps and 8% royalties. The parallel is useful for what it makes visible that a software-procurement frame does not: what the relationship extracts after the invoice, how much of what you own is only worth something inside the supplier's system, and which party ends up deciding how the work gets done. Worth a hard conversation while the engineers are still in the parking lot.