Transactions for multi-unit operators: the same rules, every site.

The business decision is shared even when the systems aren't

Three garden centre shops, one returns policy, one loyalty scheme, three separate till systems holding three separate sets of local stock. That's not three problems for a growing retailer working out how an AI agent should be allowed to transact with any of them. It's one problem, connected in three places, and treating it as three separate setups is where most of the actual risk lives.

What counts as a valid order, what needs a person to confirm, how a deposit or a return works: these are decisions the business makes once, not per location. An agent checking site two shouldn't get a subtly different answer to the same question than an agent checking site one, just because the underlying stock systems happen to be separate. The rules are the constant across all three shops. The inventory is what varies from one to the next.

Drift is the real risk, not the effort of setting each one up

The danger with three separate setups isn't that each one takes work individually, it's that they quietly stop agreeing with each other over time. A returns policy tightened at one site and forgotten at the other two is the kind of inconsistency a person might never notice and an agent checking two of the three sites would surface immediately, because it's comparing exact stated terms against each other, not general impressions of what a garden centre chain is usually like.

This is the kind of drift that happens gradually and for entirely reasonable local reasons: one site's manager tightens a return window after a run of disputes over wilted plants, meant only for that site, and nobody tells the other two. A year later three customers at three sites get three different answers to the same question, and each one, taken alone, looks like a perfectly sensible local decision. Taken together, checked by something comparing all three at once, it looks like a business that doesn't agree with itself.

What's genuinely local should stay local

None of this means pretending the three shops are identical. Stock levels, seasonal ranges, even opening hours can and should vary site to site, plants and compost are exactly the kind of stock that changes week to week, and an agent needs to see the real, current version for whichever location it's checking. The shared layer is the rules and terms; the local layer is the facts those rules get applied to. Keeping the two separate is what lets one connection cover the rules while each site's own facts stay accurate on their own.

Once the shared rules are connected properly, adding a fourth site is a smaller job than the first three were, because the actual business logic, what's a valid order, what needs a person, how a return works, already exists in one place and just needs a new set of local facts attached to it, not a fourth version of the same rules written from scratch.

Connect the shared rules once, to whichever local system holds each site's current facts, and that's the layer Selfe builds: the same rules an AI agent checks everywhere, applied to real, current, local stock at every site.

Does this mean unifying our till systems?

No, the systems can stay separate. What needs to be shared is the layer above them, the rules and terms an agent checks, connected once to whichever local system holds the current facts.

Is this only relevant once a business has many locations?

It matters from two locations onward. The drift risk that makes this worth solving properly starts the moment there's more than one place for the rules to quietly diverge.