What a standard actually buys you over a one-off integration.

A one-off integration answers one question, once

Get a data feed working well for one particular AI shopping assistant, the right fields, the right format, exactly what that one assistant wants, and it's tempting to call the job done. The specialist bike shop that built it still has to answer the harder question: what happens the next time a different AI agent, built by a different company, wants the same underlying facts in its own, differently shaped format.

Built well, a bespoke feed for one assistant is genuinely good at serving that one assistant. What it doesn't do is generalise: the next AI surface asking about bike stock and sizing doesn't automatically benefit from the work already done, because the format was shaped around one platform's specific preferences, not around the underlying facts themselves. Each new surface starts its own version of the same project.

A standard answers the shape of the question, not just today's asker

An open standard for describing a product's stock, size and fit exists independently of any one platform choosing to use it. Once a bike's real facts are held in that shape, any surface that speaks the same standard can read them directly, including ones that don't exist yet. The value sits in solving the general shape every future version of the request will take, a much better return than solving only the specific request in front of it today.

The practical difference shows up at the second connection

The first connection, whether bespoke or standard-based, costs about the same either way, someone has to do the initial work of connecting real facts to something external. The difference shows up at the second: a standard-based setup connects a new surface in a fraction of the time, because the facts were never shaped around the first platform's preferences in the first place. A bespoke setup starts the second connection from close to scratch, repeating work that a standard would have made unnecessary.

Three or four surfaces in, the gap stops being marginal and starts being the whole story. A bike shop with four bespoke integrations is maintaining four separate versions of the same underlying facts, any one of which can quietly drift from the others. A shop with one standard-based connection serving four surfaces is maintaining one version, which is a fundamentally easier thing to keep honest, not just a smaller amount of the same kind of work.

Choosing a standard isn't choosing a side

Adopting an open standard for how the shop's facts are described doesn't mean picking a favourite AI platform, and it isn't a bet on any one of them succeeding. Whichever platforms end up mattering, a shop working this way already has its facts speaking something they can understand, without having to guess in advance which ones will still be relevant next year.

Selfe holds a shop's real stock, sizing and fit in exactly that open, standard shape, so the next AI surface is a connection to existing facts, not a fresh integration project.

Isn't a bespoke integration faster to build right now?

Often, yes, for the first one. That speed advantage disappears, and reverses, from the second surface onward.

Which standard should a small shop use?

Whichever is genuinely open and already has more than one real platform reading it, not the one a single platform is promoting as its own.