Evidently

Your Interactive Reconciliation Workspace

Inspiration

Reconciliation churns away large portions of time for bookkeepers, internal auditors and accountants at businesses of any size. But reconciliation is not something that's straightforward, you need to cross check bank statement, source invoice/receipt and general ledger, sometimes information is missing, sometimes the document isn't scanned properly. That's why we decided to leverage LLMs and OCR technologies to help tackle this problem and see if we can develop a rough prototype of a "Cursor" for these financial operators when it comes to reconciliation.

What it does

Automatically reconciles general ledger using bank statement and provided invoices/receipts, flags if any information is missing or unclear so human is kept in the loop. Once bank statement and GL is matched and approved by the operator, it will go ahead and try to fetch the document that supports this match.

How we built it

We figured out some initial edge cases of general ledger and bank statement mismatch in reconciliation and developed our prompt around that. We generated synthetic bank statement, GL and invoices to iterate on our prompt and approach, finally producing a prototype that combines automation with human-in-the-loop verification.

Challenges we ran into

LLMs are smart, but certain edge cases needed more hints in the prompt to be properly solved.
Thinking of an interactive UI, how can we provide a tool that looks like what finance professionals deal with everyday.
Scope. 2 days obviously isn't enough to build a full SaaS product, so we had to cut down on many ideas/features and just focus on the core functionality and delivering the best UX we could during the 2 days of Hackathon ## Accomplishments that we're proud of
We are able to successfully pass the test cases we came up with! ## What we learned We know there are many more edge cases out there in the real world, but this at least proves that LLMs are capable of doing reconciliation when given the right tools and context! ## What's next for Evidently We are planning to build it into our Excel extension for our startup Coalesc, and ship it to our current and potential future customers and see if this could capture the interest of small and medium businesses!