You pay for the enterprise tier. Use it.

Nobody has ever written a Function

Functions are the supported way to change how Shopify itself behaves: tiered pricing that kicks in at twelve units, a delivery option that disappears over 20kg, a payment method hidden from first-time customers, order routing that picks the right warehouse. They run server-side, they are written in JavaScript or Rust compiled to WebAssembly, and they are inside the fee you already pay.

Ask who wrote your last one. If the answer is nobody, then two things are true at once: you are paying a monthly app fee for at least one behaviour a Function does natively, and you have told a customer “Shopify can’t do that” about something Shopify can do.

Nothing about this is an AI problem, and it is worth saying so plainly on a page like this one. A Function that hides a payment method over a threshold should never consult a model — it is a compiled rule with a deployment pipeline and somebody who has written one before. The work here is engineering, not intelligence.

Flow is installed and running two workflows

Tag the high-value customers, email somebody when stock runs low. That is where most Flow accounts stop — while a person in ops holds flagged orders by hand, chases 3PL exceptions in a spreadsheet, re-tags orders so the fulfilment rules fire, and copies the same figures into a Monday report.

Flow is honest about its limits: it reaches Shopify’s own objects well and everything outside Shopify badly. So the real build is a split — Flow for what lives in the platform, and a self-hosted automation layer on a VPS in your own hosting account for the half of your operation that lives in the 3PL, the helpdesk, the ERP and the spreadsheet.

That second half is where the agents belong, and they are specific things rather than a capability: one that reads the Shopify order, the subscription record and the 3PL’s tracking event and answers “where is my box” from the actual fulfilment record instead of a template with a tracking link in it; a nightly job that compares Shopify against the 3PL against the ad platforms and reports only the rows that disagree; a draft purchase order — quantities, costs, dates — landing in Slack for a human to send. Each of them reaches its systems through a tool we wrote, scoped and logged. No shared admin login, no screen-scraping, no unbounded access.

The checkout migration stopped at parity

Checkout extensibility replaced checkout.liquid, the migration got scheduled against a deadline, and the brief was “make it look like it did before”. Which it does. Nothing was added, because adding things was not what the deadline was for.

The checkout is the highest-intent page you own and the least edited. Address validation that stops the bad address before it becomes a failed delivery and a WISMO ticket. A post-purchase offer after the payment is captured, where the AOV lift is 5.6% on average across more than 40,000 merchants (vendor-reported, ReConvert). A PO-number field for wholesale buyers. A subscription upsell at the moment somebody is already buying the thing.

The targeting is where this stops being an app install: which customer sees which offer, decided from the order they are in the middle of placing and the history behind it, rather than one offer shown to everybody because that was the setting available.

B2B and the extra storefronts are still in the box

Wholesale is running on email, a shared inbox and manual invoices, or on a second platform at a second monthly fee — while B2B on Plus sits unopened with company profiles, per-customer price lists and payment terms in it. Expansion storefronts are included in the plan and are usually either unused or spun up once and left to drift out of sync with the main catalogue.

Drift is the interesting half, because it is the one that never gets scheduled. Keeping several storefronts, markets and marketplace feeds true to one source of truth means generating the fields nobody has time to write — titles, attributes, specification tables, alt text — and writing them back through the Shopify Admin GraphQL API in batches you can roll back if the result is wrong. That is a job for a system. Doing it by hand is a job for a person who will eventually stop.

This is the pattern under all four: the capability was bought, the configuration was never done, and the gap gets filled by an app subscription or a person doing it by hand. Both of those are recurring costs sitting on top of a recurring cost.

The enterprise tier is not a status. It is a set of tools with a licence fee attached, and the fee is charged whether or not the tools are switched on.