This weekend I was working on reconciling what data we have in our accounting system against what is there in the bank statement for a particular bank account. I chose DeepSeek for my task. The main reason I chose DeepSeek was that it had comparable performance to GPT on math and logic-related benchmarks. As this reconciliation task is mainly a math and logic problem, I needed to choose an AI model that performed well on my task. Furthermore, DeepSeek's inference cost is 1/30th of GPT's inference cost. So if the solution works and I need to use it at scale, my DeepSeek bill won't be too high. The last but the most important reason was that the DeepSeek model weights are available Open Source. So if I decided that I wanted to run the DeepSeek AI inference privately without submitting the data to them, I can do that.
With the help of DeepSeek AI, I was able to do this reconciliation of over 50 transactions in less than 2 minutes. Here is how I did it. I deleted the account numbers from my files before I submitted them to DeepSeek AI. All I had in my files was transactional data.
Rishi: I have two statements, one from my ERP and one from my bank. Can you reconcile them? Here are the 2 statements:
DeepSeek: Yes, I can help you reconcile the two statements. Let's compare the transactions from your ERP (LukeNov.xlsx) and your bank statement (eStmt_2024-11-29.pdf) to ensure they match. (I am only going to show you 2 transactions)
Step 1: Extract Transactions from Both Statements
ERP Transactions (LukeNov.xlsx):
| Date |
Vendor Name |
Cash In/Out |
Category |
| 2024-11-01 |
Supplies |
-35 |
Subscriptions |
| 2024-11-01 |
Udemy |
-12.99 |
Training/Education |
Bank Transactions (eStmt_2024-11-29.pdf):
11/04/24. CHECKCARD 1101 *Supplies-35.00
11/04/24. PURCHASE 1102 UDEMY: ONLINE COURSES-12.99
Step 2: Reconcile Transactions
*Details not shown for brevity
Step 3: Verify Totals
*Details not shown for brevity
DeepSeek Conclusion:
The ERP (LukeNov.xlsx) and bank statement (eStmt_2024-11-29.pdf) are fully reconciled. All transactions and totals match. No discrepancies were found. Let me know if you need further assistance! 😊
Rishi: Can you please show the reconciled data into a single table?
DeepSeek: Certainly! Below is a reconciled table that combines the data from both your ERP (LukeNov.xlsx) and your bank statement (eStmt_2024-11-29.pdf). Each transaction is matched and verified:
Reconciled Transactions Table
| Date |
Vendor Name |
Description |
Cash In/Out |
Category |
Source |
| 2024-11-01 |
OfficeDepot |
Supplies |
-35.00 |
Subscriptions |
ERP & Bank |
| 2024-11-01 |
Udemy |
PURCHASE 1102 UDEMY: ONLINE COURSES |
-12.99 |
Training/Education |
ERP & Bank |
Conclusion: This same type of reconciliation can easily be done between vendor statements and your accounting system's data. Human time is the most important asset in the world. Given the ready availability of AI tools, we can use them to save our precious time. In the following link, you need to submit your 2 files to be reconciled:
https://huggingface.co/spaces/rishisriv-bh/Statements-Reconciliation
It comes back with a reconciliation of the statements for you.