Camelot is usually described as an open-source Python library for extracting tables from PDFs that runs on your own machine. If you want an alternative, the real question is whether you want to keep writing code or just want your statement rows in a usable shape. The choice usually comes down to how many statements you handle and how comfortable you are fixing a parse that goes wrong.
What are you actually trying to get out of the PDF?
Extracting a table is only a step. What most people want at the end is a list of dated debits and credits, sorted into categories, that adds up to the statement's own totals. Write down your end goal before judging any tool.
- A clean spreadsheet you can filter and pivot.
- Categorised spending so you can see rent, food and travel separately.
- A repeatable monthly routine that takes minutes, not an afternoon.
What does a code-based route involve?
A library-based route gives you control: you decide which page areas count as a table and how columns are split. The price is time. Every bank formats its statements differently, so a script that works for one layout may need adjusting for another, and a scanned PDF (a photo of paper) is a different problem from a PDF generated by a bank.
Whatever you use, check that the output ties back to the statement: add the debits and credits yourself and compare with the closing balance shown on the document.
How do I check any extraction is right?
| Check | What to do | What a failure looks like |
|---|---|---|
| Row count | Count transactions on the PDF page, compare with extracted rows | Missing or merged rows |
| Totals | Sum debits and credits, compare with the statement summary | Totals differ by one row's amount |
| Dates | Scan the date column for odd formats | Day and month swapped |
| Wrapped descriptions | Check long merchant names | One transaction split across two rows |
| Multi-page tables | Look at page breaks | Header row repeated as data |
Where does SpendixAI fit as an alternative?
SpendixAI accepts bank and card statements as PDF, CSV or Excel. You pick the account first, upload, and review the rows before anything is saved; a batch upload and an import history with status are included, and an Import failed notification tells you when something did not go through. Possible duplicate imports are flagged. For an unfamiliar layout, an optional feature can read the uploaded document, so only turn that on if you are comfortable with the file being processed.
It does not promise that every layout parses perfectly, which is exactly why the review step exists.
A quick arithmetic check you can do by hand
Say a hypothetical statement opens at ₹48,200, shows four debits of ₹1,250, ₹830, ₹14,000 and ₹2,420, and one credit of ₹60,000. Debits total 1,250 + 830 + 14,000 + 2,420 = ₹18,500. The closing balance should be 48,200 + 60,000 - 18,500 = ₹89,700. If the extracted rows give a different figure, a row is missing, duplicated or misread.
What if the statement is scanned or has an odd layout?
Two PDFs can look identical on screen and behave very differently underneath. A bank-generated PDF holds real text that extraction tools can read. A scanned or photographed one is just an image, which needs text recognition first and is much less reliable. Layouts with multi-line descriptions, separate debit and credit columns, or a running balance column also trip up simple extraction.
- Download the statement from the bank's own portal as a fresh PDF if you can, rather than a photo.
- If the bank offers CSV or Excel, try that first; it avoids table extraction altogether.
- Keep the original PDF in case you need to check a row.
- Never upload or share a document through a route you do not trust.
Where an upload-and-review tool is used, the review screen is your safety net: scan each row against the PDF before saving, especially around page breaks.
What should I decide before switching?
- How many statements a month you handle, and from how many accounts.
- Whether you want code you control or a review screen.
- Where the PDF will be processed: on your machine, or uploaded to a service.
- What you do with the result: analyse in a sheet, or see categorised reports.
Run one real statement through whichever route you choose, tick off the row count and totals, and only then move a full year across.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

