Transcript
The request
Monday, 9:00 AM. Keldmoor's external auditor asks for a list: every supplier invoice in the September AP sample. Supplier, invoice number and date, purchase order, payment terms, due date, currency and total – by Friday. No system holds that list in this shape. The answers are in the documents themselves. Invoices in English, Spanish and Czech – and a paper copy, in the morning scan from the mailroom. So the AP team builds the list straight from the documents. Three steps, no code.
Model Manager
Step one: Model Manager. Models define which fields are extracted from documents, and in what structure. Canonical models describe document types. Custom models are yours – shaped to your own systems, or to a single request. New model, Empty model. Name: Audit sample. Domain: accounting. Create model. Model studio opens on the new draft – ready for its fields.
Model studio
Add field. Supplier, a text. The description matters most – it goes to the AI that reads the documents. So say what the field means: the supplier's legal name, as printed on the invoice. Invoice date gets the Date format: however a document prints it, the column holds one format. Payment days is an integer: Net 45 becomes 45, and 30 días becomes 30. And due date may be empty – its description says: leave it empty if the invoice doesn't state one. Eight fields – the auditor's list, column by column.
Check and publish
Check looks over the schema before anything is saved: no errors, no suggestions. Save and publish walks through three steps: check, changes and impact, done. Eight fields added – none of them risky. Audit sample is published – ready for Reveal Manager.
File Manager
Step two: the documents. In File Manager's Explorer, filter by document type: invoices and freight bills. Eight documents – including the paper copy and the freight invoice split out of the mailroom scan. Select all. Reveal Group. Reveal Manager opens a new run with all eight files. Under Mapping models, add Audit sample. Parsing: use existing. Every file was read when it arrived – none is parsed again. The dataset is created on first write. Start run: eight files, one model.
Reveal Manager
The board follows every file: parse, then for each model – mapping, rules checks and Analytics. Parse is already done – the existing result is used. Mapping fills the eight fields from each document. This model has no rule groups yet, so checks are skipped. Each result is written to Analytics. A few minutes later: done. Eight written to Analytics. Open in Analytics.
Analytics
Step three: Analytics opens the new dataset in Data Explorer – the data layer of Audit sample. Pick the eight fields as columns – and run the query. Eight rows, one per document. American, Spanish and Czech dates – all in one format. Net 45, 30 días, 30 dní – now plain numbers. Dollars stay dollars, euros stay euros. Sort by invoice date, and INV-88163 shows up twice – the PDF and its paper copy from the mailroom scan. Caught before anyone pays it twice. The freight invoice states no due date – so that cell stays empty.
Export
Exports: Export CSV prepares the file from the same query – eight rows. Download – and the file is on your computer, the columns in the auditor's order. Behind the data sits an audit layer: for every row, the run, the model version and the time – plus the full extracted record. 9:19 AM: the reply goes out, the list attached.
Next time
The model stays. Next month, new invoices go through the same model – same columns, new rows. An AI assistant connected over MCP can query the same dataset – read only. And after document mapping, an integration can start on its own – the next step, automated. One request. Three managers. One table – and the next request starts from here.