Fictional sample - not a real bill
Evergreen Electric Cooperative
Delivery fee$5.00
City tax$4.00
Invented issuer, dates, amounts, and account context. No customer document was used.
Bills become questions you can actually ask
Understand your bill. Know what to ask next.
ExpenseReader separates what a document says, what the arithmetic supports, what remains unknown, and which follow-up questions may be useful.
Penny Lane is a fictional AI bill-analysis expert - not a person or a licensed professional.
A transparent sample
Read the statement, then check the result
This fictional utility statement uses the same structured fields, source spans, signed amounts, and deterministic checks as the real analysis pipeline. Select a marked row to see how it is interpreted.
Page 1 of 1
Seven complete categories
Choose the language your bill uses
The underlying workflow is shared; terminology, extraction fields, checks, and safety limits change with the document type.
What ExpenseReader keeps distinct
Utilities
Checks printed usage, rates, fees, credits, and totals without inventing tariffs.
Penny Lane’s notebook
Recently published guides
AI-generated educational articles, published automatically only after source, safety, and rendering checks.
saas
An invoice total and a receipt total answer different questions
How status, dates, credits, and amount due help distinguish a software payment request from a record of payment.
insurance
Read the declarations page beside the endorsements
A practical way to sort policy identifiers, limits, deductibles, baseline wording, and later changes without deciding what a claim outcome would be.
phone cable
Read a promotion as a dated line, not a permanent plan price
A promotional credit, device payment, and partial-month charge can explain today’s total without establishing what a later bill will be.
utility
Actual and estimated readings tell different billing stories
Read the service period, reading type, meter values, units, and usage calculation before drawing conclusions from a utility bill.
Privacy, stated precisely
Your document is not a marketing asset
Review before AI
You correct and mask extracted text first. Detection helps but can miss sensitive details.
PDF stays out of the model
ExpenseReader sends only the corrected text for analysis, never the original PDF bytes.
App state is temporary
The page uses in-memory state. Provider and hosting processing are separate and described in the privacy notice.
Analyze your own document
A private, correctable workspace
Use already-redacted text when possible, or extract a readable text layer from one PDF. Your result and follow-up conversation remain in this browser tab.
Analyze your own document
Private bill workspace
Session-only · refresh loses your work · five analyses per UTC day, subject to the shared budget
Start with a copy, not your only document.Remove names and identifiers. Automated detection is incomplete. The app does not save document content, but provider retention is a separate process described in the privacy notice.
1
Choose and correct
Text-layer PDFs only; scans and photos are unsupported.
Correct extraction errors, preserve minus signs and decimal points, and remove personal identifiers. ExpenseReader analyzes exactly this text.
0 / 120,000 characters · 0 potential identifiers
No common identifier patterns detected. Manual review is still required.
2
Confirm the privacy review
Changing the text or category clears these confirmations.
Nothing sent for analysis
Paste already-redacted text or extract a text-layer PDF, then review every page.