Every restaurant back-of-house tool can read your invoices now. Almost none of them can run your purchasing. The difference is about to sort this market.
Watch the restaurant software market for a season and you will see the same product pitched under a dozen logos: snap or email-forward your supplier invoices, AI extracts the line items, recipe costs update, and a dashboard tells you your food cost percentage before the month closes. Some add stocktake apps. Some add task checklists. Some add an "AI insights" chat on top. The names rotate — the loop is identical.
There is a reason everyone converged here, and it is worth being precise about, because it tells you where this category goes next.
Invoice reading was a moat. Then it wasn't.
In 2019, parsing a supplier invoice was genuinely hard. Distributor invoices are a mess of catchweights, split lines, credits, and handwriting. Companies built real pipelines — trained extraction models, human review queues — and that capability was a defensible wedge. You could build a company on "we digitize your invoices," and several did.
Then large language models made document extraction a commodity. Today a general-purpose model reads a produce invoice with the accuracy that once took a specialized vendor years to reach, and the cost of that capability is measured in fractions of a cent. Every incumbent bolted it on. Every new entrant launches with it. When a capability becomes available to everyone at commodity prices, it stops differentiating anyone — which is exactly why you now see a flurry of near-identical tools. The wedge that used to separate products evaporated, and the category converged on the same feature list.
The tell is in the marketing. When products are structurally identical, positioning inflates: "insights" becomes "intelligence," dashboards become "copilots," and alert emails become "AI managers." The language escalates because the loop underneath cannot.
The rearview mirror problem
Here is the loop those tools actually run: something happens in your restaurant, paperwork is generated, the software reads the paperwork, and a report describes what happened. Invoice scanning, COGS dashboards, variance reports, menu engineering matrices — all of it is descriptive. All of it is the past.