Small manufacturers keep books with holes in them. Structero finds the holes.
An audit and intelligence platform that takes one accounting export, reconstructs the transactions that should exist, and shows every one that is missing, mispriced, or late.
This is a work in progress, not a finished product. The audit engine is built and validated. The application around it is designed but not built. This page is here to show the idea and the working prototype, nothing more.
A small manufacturer runs on QuickBooks and a lot of habit. Work orders get opened and never closed, so work-in-process sits on the balance sheet overstating inventory. Vendors bill for material nobody confirmed receiving. Invoices go out for goods that were never shipped, recognizing revenue that has not been earned.
None of this shows up as an error, because from the ledger's point of view nothing is wrong. A missing transaction leaves no trace. You cannot search for a row that was never written.
Their accountant finds some of it, once a year, by hand.
Every business transaction belongs to a chain. A purchase order should be followed by a receipt, then a vendor bill, then a payment. If you know the chain, you can generate the transactions that ought to exist from the ones that do, then match the two sets and read off the difference.
That distance is the product. The reaction we are designing for is "how did you get all of that out of the file I sent you?"
Gap: the bill posted with no receipt on record, so material was paid for that nobody confirmed arriving. The bill itself is also over the purchase-order price.
Gap: the work order never closed, so its work-in-process is stuck open and inventory is overstated until someone notices.
Gap: an invoice recognized revenue with no shipment behind it, which is a revenue-recognition risk the auditors will care about.
The audit arithmetic is deterministic SQL, so no figure is ever invented. The Claude application programming interface is used only to infer the dependency rules from a new client's data and to write the plain-language explanation of a variance. Numbers never come from a model.
| Gap | What it means | Why it costs money |
|---|---|---|
| Bill, no receipt | A vendor bill with no receiving transaction behind it | Paying for material nobody confirmed arriving |
| Work order never closed | Production started, no closing entry | Work-in-process stuck open, inventory overstated |
| Invoice, no shipment | Revenue recognized with nothing shipped | Revenue-recognition exposure at audit |
| Payment never deposited | Customer paid, cash never landed | Cash unaccounted for, reconciliation breaks |
Alongside these, the engine flags price variance (billed above the purchase-order rate), quantity variance (more material issued than the bill of materials called for), and timing variance (paid late against terms).
The four screens an analyst moves through, on an invented sample company. Every figure in it is dummy data chosen to make one coherent scenario legible, not a real client result.
Open the prototype →Controls marked "Rewire (roadmap)" are deliberately disabled. They mark where the next capability lands, and they are not built.
The person at the keyboard is an analyst working through a client's books, either inside the manufacturer or at the accounting firm that keeps them. The controller or general manager is who introduces the tool, and the finance lead is who has to believe the output.
That shapes the interface more than anything else. The first screen leads with the biggest findings in plain language rather than a table of records, because the first run has to be legible to someone who did not ask for it. Everything below that first screen is for the analyst who then has to work the list.
The comparison point is a consulting engagement that ends in a static slide deck. The intent here is something durable instead: re-runnable each period, drillable down to the individual transaction, with a record of who reviewed what.
Being straight about this, because it is the thing worth talking through. The core mechanic is proven and the platform around it is designed but unbuilt.
Honest one-line status: end of discovery, start of build. The next concrete step is the first vertical slice, upload through to report in a browser.
Ordered by decision, not by wishlist. Each stage assumes the one above it.