DBT Invoice Extractor
3,456.00 is not 3,546.00.
A web app that reads invoices with AI, lets finance teams verify every field and the totals, then files them with a full audit trail.
Open on the old way failing: someone keys an invoice into a spreadsheet and transposes two digits, and the cell flashes red. Then follow one invoice through the product, from upload to a Balanced cross-check, an approval and an audit trail.
- 16:9
Invoice extraction tools all claim AI accuracy. For a finance team, the question is whether they can trust the numbers before approving them. The launch film had to show that trust being built, field by field and total by total, for an audience that's sceptical by profession.
Accounts payable and finance teams, and project accountants who file invoices against projects and vendors.
Scene by scene.
▶ 0:00The hookAn invoice keyed into a spreadsheet, two digits transposed, the total flashes red.
▶ 0:04The revealThe DBT Invoice Extractor lockup on navy: read, verify, approve and file.
▶ 0:07UploadThe invoice is dropped into the ingestion page and extraction starts.
▶ 0:11VerifyA scan passes over the document as each field fills in with its own confidence score.
▶ 0:17Cross-checkLine items plus tax are checked against the declared total. Balanced. Approve and file.
▶ 0:24AnalyticsSpend by project and vendor, and a payables horizon of what's due when.
▶ 0:30Audit trailNew rows slide into the immutable audit log.
How it was made.
- 01
Every screen is captured from the real app and rebuilt shot for shot in its own light slate UI.
- 02
The invoice, vendor and amounts are illustrative. No accuracy or speed figures are claimed, because the workspace was a test one.
- 03
A polished, precise tone on a navy stage, with the logo's orange used sparingly for highlights.
- 04
Cut to 120 BPM with cuts on the beat. No voiceover.
The client was stunned by the result, and more so that it was made entirely by AI.