Accounting · Document processing
Helping finance teams spend less time entering documents
As part of the engineering team, I worked on document extraction and validation to reduce manual data entry for finance teams. Uncertain results were sent for review before posting to the accounts.
I contributed to this project as part of an engineering team. This case study focuses on the areas I worked on and what I learned along the way.
What needed to improve
Entering financial documents by hand was slowing down month-end accounting. Our team built a platform to read those documents and create accounting records. My work covered extraction, validation, and keeping each customer’s data and rules separate. The difficult part was deciding which results could be posted and which needed someone to check them.
What the system had to handle
- Documents arrived as scans, in different layouts, and sometimes with missing fields.
- An incorrect value could end up in the accounts, so extraction needed checks before posting.
- Customers shared the platform, but their data had to stay separate.
- Reports needed to update as new accounting records were added.
My contributions
Improve document-reading accuracy
Different document layouts and missing fields could lead to incorrect records. I investigated those errors and worked on improving the results, so finance teams had less information to correct by hand.
Check the results before they reach the accounts
Saving time on data entry would have limited value if incorrect figures reached the accounts. I worked on checks and review of uncertain results, so the system could reduce manual entry while keeping people involved where their judgment was needed.
Let customers use their own rules
Different customers had different accounting requirements. I helped support those differences so teams could use the product within their own processes.
What went wrong
The first version struggled with real documents
Documents used in practice were more varied than our early examples. I helped investigate issues with different layouts and missing fields, and improve how uncertain results were handled.
What I learned
I would build the checks and review path early. For this system, handling uncertain results was a large part of making document automation useful to the finance team.