Middleware, sync jobs, overselling apologies and a mapping table someone owns as a career: the integration tax is real, recurring, and optional. It disappears when the storefront and the ERP share one database.
There is a tax nobody itemizes when a distributor buys a storefront from one vendor and an ERP from another: the integration tax. It is paid in middleware subscriptions, in sync jobs that run every fifteen minutes, in a consultant on retainer for when they fail, and — most expensively — in the daily gap between what the website believes and what is actually true.
Consider what the storefront must know to take one honest order: the price this customer pays, the stock that is genuinely promisable, the customer's credit standing, and the substitutions if the item is short. In a two-system architecture, every one of those is a copy — extracted, transformed, transmitted, and aging from the moment it lands. The architecture's formal name is eventual consistency. A warehouse has a shorter name for it: overselling.
The failure cases are boringly predictable. A quantity of twelve syncs at 9:00; by 9:20 the counter sale took ten; at 9:25 the web sells eight of the two that remain. Now someone phones an apology, and operations 'fixes' it by publishing only half the real stock to the site — quietly refusing web revenue every day to paper over a sync interval. Price drift is worse, because customers assume malice: a storefront honoring last week's price list turns a contract renegotiation into a dispute about the difference.
Master data is the tax's compounding interest. Two systems means two product masters, two customer files, two definitions of unit of measure, and a mapping table between them that somebody maintains as a standing duty. Every new product is set up twice; every mismatch becomes a reconciliation meeting. Integration platforms soften the keystrokes, but they cannot repeal the underlying fact: the same fact now lives in two places, and one of them is always wrong.
The shared-database architecture dissolves the problem rather than managing it. When the storefront and the ERP are two faces of one system, the stock level the website promises is the same record the picking wave decrements — not a copy of it, the row itself. A receipt posted at the dock is promisable online in the same second. A price change is one change. There is no sync to monitor because there is nothing to sync.
Credit and compliance stop leaking too. A web order flows through the same credit check, the same tax logic and the same order workflow as an order keyed by a service rep — same allocation rules, same audit trail, one place to answer when an auditor asks who changed a price and when. In a synced architecture, every one of those rules must be reimplemented on the commerce side and then kept in agreement forever. Two implementations of a credit policy is two policies.
The standard counterargument is best-of-breed: a dedicated commerce platform will always out-feature an ERP's storefront. For consumer retail at scale, that can be true. But a distributor's storefront competes on exactly the things a copy cannot deliver — live contract pricing, honest availability, customer-specific catalogs, real order status, credit visibility. Those are ERP reads. A merely adequate storefront wired directly into the truth beats a beautiful one that is fifteen minutes behind it, every day, for years.
There is also the arithmetic. Two systems means two license lines, the middleware between them, the integration project — routinely quoted in months and delivered in more — and the standing reconciliation labor: a person, or a fraction of one, forever comparing reports. That spend produces no feature a customer sees. It exists to compensate for an architectural decision, and it recurs annually, which is the definition of a tax.
The test to run on any proposed stack is one question: when the storefront shows a quantity, is it reading the warehouse's own record, or a copy? If the answer involves the words 'near real time', you have found the tax — 'near' is where the apology phone calls live. One schema, one price, one stock level, one audit trail. Everything else in the demo is decoration.