What Happens When You Upload Your Bank Statement: A Complete Walkthrough

Most budgeting apps show you a dashboard. They do not show you how they get there. Here is exactly what happens when you upload a bank statement PDF to Spend & Invest — every step, from the raw PDF to a complete spending breakdown.

Step 1: Upload — Drag, Drop, Done

The upload starts with a drag-and-drop zone. You drop in your PDF, or click to select it from your computer. That is all the input required. No bank username, no password, no two-factor authentication prompt. The file goes from your computer to our servers over an encrypted connection.

What gets sent: the PDF file itself. That is it. No metadata about your device, no browser fingerprinting, no request to access your camera or location. Just the PDF.

What does not get sent: your bank login credentials, your account numbers (beyond what appears on the statement face), or any information about your actual bank account. There is no connection to your bank. We cannot initiate transactions, check real-time balances, or see anything beyond the statement you uploaded.

File size limits: up to 60 pages per PDF, and up to 10MB. Most monthly bank statements are 3-10 pages and well under 5MB. Password-protected PDFs are supported — you enter the password during upload and it is used only to open the file, never stored.

Step 2: AI Parsing — Reading the Statement

Once the file is received, Claude (Anthropic's AI model) reads the PDF page by page. This is not simple text extraction — AI is needed because bank statement formats vary enormously across institutions, countries, and account types.

For each page, the AI extracts:

• Transaction date. The date the transaction posted to the account.
• Vendor name. The raw merchant name as it appears on the statement (e.g., "AMZN MKTP US*1X2Y3Z4").
• Amount. The transaction amount, with debit (money out) and credit (money in) properly distinguished.
• Statement summary fields. Opening balance, closing balance, total debits, total credits — used for reconciliation.

The parsing handles multi-page statements automatically. It also handles the quirks of different statement formats: some banks list credits as negative numbers, others use a separate column. Some use dates in DD/MM/YYYY format, others MM-DD-YY. The AI reads them all correctly.

Parsing typically takes 20-45 seconds for a standard monthly statement. Longer statements (3 months, annual) take proportionally longer.

Step 3: Categorization — The 4-Tier System

Once the transactions are extracted, each one gets categorized. This is not a simple keyword lookup — it uses a four-tier priority system that gets smarter every time you use it:

1. Your past corrections (highest priority). If you have ever manually corrected a category for a vendor, that preference is applied first. "TRADER JOE'S" goes to Groceries because you said so, and it will always go to Groceries from that point forward.
2. Your category list. The AI knows your custom categories and tries to assign each transaction to one of them specifically, not just a generic default.
3. Crowd data. For vendors that many users have interacted with, we use aggregate preference data (anonymized) to inform the guess.
4. AI inference (fallback). For unknown vendors, Claude analyzes the vendor name, amount, and context to make its best guess. A charge for "ADOBE SYSTEMS" at $54.99 goes to Subscriptions. "LYFT *RIDE" goes to Transportation.

Each categorization also receives a confidence score from 0 to 1. High-confidence assignments (above 0.7) are displayed with a green dot. Low-confidence assignments (below 0.7) get an amber dot and are highlighted for your review. The AI knows when it is uncertain and flags those cases rather than guessing quietly.

Step 4: Review — Teach the AI

The review screen shows you every transaction in the statement, sorted by confidence. Low-confidence items appear at the top with an amber highlight so you can review them first. High-confidence items are at the bottom — you can skim or ignore them entirely.

Correcting a category takes one click. Select the transaction, choose the correct category from the dropdown, and confirm. You will then be asked: "Apply this correction to all transactions from [Vendor] across all statements?" Say yes and the system learns your preference — every future statement will apply it automatically.

The reconciliation banner at the bottom of the review screen compares the AI's computed total against the statement's reported total. If they match, you get a green checkmark. If they do not match, the banner tells you the discrepancy amount — useful for catching parsing errors on unusual statement layouts.

When you are satisfied with the categorization, click Finalize. This locks the statement, adds the transactions to your historical data, and makes everything available on the dashboard.

Step 5: The Dashboard — Your Spending Picture

Once a statement is finalized, the dashboard updates with your full spending picture. What you can see:

• Category breakdown. A donut chart showing what percentage of your spending went to each category. Click any category to see the individual transactions behind it.
• Monthly trend chart. If you have uploaded multiple statements, a bar chart shows your total spending by month. Immediately visible if a month is unusually high or low.
• Top vendors. The five vendors you spent the most with, with exact totals and clickable drilldowns to see every transaction from that vendor.
• Month-over-month comparisons. Percentage changes for each category between the current month and the previous month. Categories that increased more than 20% are highlighted in amber as an early warning.
• Subscription tracking. After 2+ months of statements, the recurring charge scanner surfaces every subscription it detected, with estimated monthly cost and next expected charge date.

Step 6: Ask Questions — Natural Language Queries

The query bar at the top of every page lets you ask questions about your spending in plain English. The AI interprets your question, translates it into a database query, and returns a precise numerical answer with the supporting transactions.

Questions that work well:

• "How much did I spend on dining last month?"
• "What subscriptions am I paying for?"
• "How does this month compare to last month?"
• "What was my biggest single purchase in the last 3 months?"
• "How much did I spend at Amazon this year?"

The answers are computed using exact integer arithmetic — no floating-point rounding, no estimation. If you spent $847.23 on dining, that is the number you get.

A Note on Privacy — The PDF Is Discarded

Once your PDF is processed and the transactions are extracted, the raw file is not stored. We keep the structured transaction data (dates, vendors, amounts, categories) because that is what powers your dashboard and query history. But the original PDF — which contains your full name, account number, and address — is discarded after processing is complete.

This is intentional. The PDF contains more personal information than we need. We need the transaction rows. We do not need your address. Discarding the original file is the privacy-correct approach, and it also means we have no incentive to hold onto data we did not need in the first place.

To learn more about our privacy approach, read why we do not require a bank login or our overview of what Plaid does with your data.

The Whole Thing Takes About 5 Minutes

From download to dashboard, the entire process for a first-time user takes about 5 minutes:

1. Create a free account (30 seconds).
2. Set up your categories during onboarding — or use the defaults (1 minute).
3. Download your bank statement PDF (1 minute).
4. Upload the PDF and wait for parsing to complete (30-60 seconds).
5. Review any flagged transactions and finalize (1-2 minutes).

After the first upload, subsequent months take under 2 minutes — the AI already knows your vendor preferences and most transactions will be high-confidence.

Ready to try it? Get started free — no credit card, no bank login, no trial period.