Taco Bell
A receipt and invoice management platform that replaces paper. Every invoice and receipt is uploaded once and kept as data: read for you, checked where it’s unsure, and exported when finance needs it.
Overview
The platform is called AI DocProcess. It is where the team keeps its invoices and receipts, as records rather than paper: each one is uploaded once, its data is read and recorded, and from then on it can be found, checked and exported in the formats finance already uses: Excel, CSV, JSON or the ERP. Reading a document is the easy part; the platform also says how sure it is of every field, so a person only checks the few it doubts.
User goals
-
Accounts payable clerks
Check a document without retyping it.
-
Finance leads
Know which numbers were trusted and which were checked.
-
Platform admins
Get the month out to the ERP without babysitting it.
-
The finance team
Stop keying invoices by hand.
Problem
Invoices and receipts lived on paper.
Each one was filed, then keyed in by hand before finance could count it: slow, easy to get wrong, and hard to find again when someone asked.
Design goals
-
Paper becomes a record
Upload once; the data is kept, searchable, and never typed again.
-
Score every field
A number on each one, not a single score for the document.
-
Point at the doubtful ones
Flag what needs a human, and leave the rest alone.
-
Keep the source in view
The document beside the fields, so a check takes one glance.
-
Make the state readable
One status per document, the same word everywhere it appears.
-
Let the month run itself
Scheduled exports to the formats finance already uses.
Solution
Upload the paper once; keep it as data from then on.
Each document arrives through a three-step path (upload, processing, validate), so the wait has a shape. The engine returns every field with a confidence score, and the review screen puts the original beside what was read from it. Fields above the line sit quiet; the ones below turn amber, open for editing, and are counted in the header so the clerk knows the size of the job before starting. Once saved, the document carries one status through the list, and the export centre sends batches on by hand or on a schedule.
Some of the solutions
-
A score on every field
Not one number for the document. Eight fields, eight scores, so trust is specific.
-
Doubt, made obvious
Below the line a field turns amber, carries a warning mark and becomes editable. Everything else stays read-only.
-
The wait has a shape
Upload, processing and validate are numbered, so nobody wonders whether it worked.
-
Status and accuracy, side by side
In the list, what the document is doing and how well it was read sit in neighbouring columns.
-
Export without asking twice
Format, range, type and status in one row, with the record count under the button.
-
The month, on a timer
Daily, weekly and monthly jobs with their next run, each switchable without opening it.
Structure
Information architecture
AI DocProcessDesktop and mobile
- Dashboard
- Documents
- Upload files
- Review extracted data
- Export Center
- Generate export
- Export history
- Scheduled exports
- Integrations
The core flow: read it, doubt it, check it, send it
- AP clerk Upload the file → Processing
- OCR engine Read every field
- AP clerk Fix the flagged fields
- AP clerk Save and complete → Validated
- Finance lead Send the batch out → Exported
Final design
Clerks working through documents
Upload, wait, check only what the engine doubted.
Leads watching the whole queue
What came in, how it was read, and what is scheduled to leave.
© 2026 Sree Harini. All rights reserved.