🤖 AI Finance-Ops Agent: Bank-to-GL Reconciliation

This project demonstrates an AI-assisted finance operations workflow that automates the reconciliation between bank transactions and the General Ledger (GL).
It was designed and built by Marjaana Peeters, aspiring AI-savvy fractional CFO, to explore how no-code agents and synthetic finance data can automate repetitive accounting processes.

---

🚀 Overview

Purpose:
Automate the Bank → GL reconciliation process, including:

• Categorisation of bank lines
• Matching to AR/AP invoices
• Generating journal suggestions for unmatched lines
• Maintaining a run log
• Updating a live KPI dashboard (Google Sheets)

Stack:

• 🧠 ChatGPT / OpenAI Agent Builder (no-code visual canvas)
• 📊 Google Sheets & Drive for inputs / control / dashboard
• 🧾 CSV data store for synthetic finance transactions
• 🛠️ Optional integrations: Slack (alerts), Notion (reporting), n8n / Make (automation)
• 🐍 Python SDK (optional extension for dataset or API logic)

---

🗂️ Repository Structure

Finance-Agent-Lab/ ├── Templates/ │ └── Bank-to-GL_Reconciliation_Template/ │ ├── inputs_template/ │ │ ├── bank_transactions_template.csv │ │ ├── ar_invoices_template.csv │ │ ├── ap_bills_template.csv │ │ └── gl_journal_template.csv │ ├── Bank_Recon_Control_Template.xlsx │ └── Bank_Recon_Output_Template.xlsx ├── README.md └── docs/ ├── screenshots/ └── finance-agent-architecture.png

---

📋 Workflow Steps

1️⃣ Inputs

Four CSVs act as input datasets (synthetic or real):

• bank_transactions_template.csv
• ar_invoices_template.csv
• ap_bills_template.csv
• gl_journal_template.csv

2️⃣ Control Sheet

Holds configuration, categorisation rules, and run log:

• CONFIG: file paths, tolerances
• CATEGORIES: keyword → account mapping
• RUN_LOG: timestamped record of each run

3️⃣ Output Sheet

Contains:

• SUMMARY
• MATCHED_AR / MATCHED_AP
• UNMATCHED
• SUGGESTED_JOURNALS
• DASHBOARD (live KPI & chart)

4️⃣ AI Agent

The OpenAI no-code Agent performs:

1. Load CSVs (from Google Drive)
2. Match AR/AP vs. bank data
3. Categorise unmatched lines
4. Generate journal suggestions
5. Append run logs
6. Refresh dashboard

5️⃣ Automation (optional)

• Slack alert if match rate < 95 %
• n8n / Make flow to duplicate template folders for each client (<PROJECT_NAME>)
• GitHub Actions / cron for nightly refresh

---

🧾 Example Outputs

Metric	Value
Latest Run	2025-10-13 13:15
Bank Rows	707
Matched AP	641
Unmatched	66
Match Rate	90.7 %
Comment	auto-cat + journals added

Suggested Journal Example

Date	Account	Memo	Debit	Credit
2025-01-12	5100	Bank charges	15.00	0.00
2025-01-12	1000	Bank account	0.00	15.00

---

🧩 Next Steps / Extensions Add OCR invoice ingestion (Google Vision / Tesseract) Integrate Slack/Teams for daily alerts Link with QuickBooks or Xero API for posting journals Build an n8n scenario for one-click monthly close

🧠 Author Marjaana Peeters AI-native strategic finance professional LinkedIn: www.linkedin.com/in/marjaana-peeters-0442a4

🪪 License MIT License – feel free to reuse, credit, and extend.

---