Introduction

In large manufacturing enterprises operating on SAP HANA/ERP System, vendor reconciliation is one of the most critical yet time-consuming month-end activities. Discrepancies between vendor-submitted statements and ledger data can lead to incorrect liabilities, GST mismatches, and delayed payments — all of which directly impact audit quality and working capital management.

This use case presents a full-stack AI-powered Vendor Reconciliation Portal (VRP) developed and deployed at the plant finance level, demonstrating how a Chartered Accountant can independently architect and implement an enterprise-grade automation solution using modern AI tools.

Shape

Problem Statement

Finance teams at organisation face the following reconciliation challenges:

Vendors submit ledger statements in inconsistent formats — Excel, PDF, Word, or image-based scans — making standardized processing impossible

Manual matching of vendor ledger lines against SAP open items is prone to errors and consumes 3–5 working days per month-end cycle

There is no centralized audit trail of reconciliation communications — emails are scattered, versions are untracked

Escalation and vendor follow-up depend entirely on individual team members rather than a system-driven workflow

Sensitive financial data is shared via insecure channels without any access control

Shape

Technology Used

Component / Technology

Purpose in This Portal

Node.js built-in HTTP server

Runs the local portal server, vendor submission server, and OpenAI reconciliation webhook.

JSON file storage

Stores portal state, users, vendors, audit logs, reports, submissions, and reconciliation data in JSON files under the data folder.

OpenAI Responses API

AI engine used for vendor reconciliation, ledger comparison, document extraction, risk scoring, and structured reconciliation output.

HTML / CSS / Vanilla JavaScript

Frontend portal for admin login, vendor upload, reconciliation dashboard, reports, settings, and email triggering.

SheetJS / XLSX

Reads Excel files such as .xlsx and .xls uploaded into the portal.

jsPDF

Generates downloadable PDF reports from the portal.

Google Apps Script

Provides the public vendor submission page, status sync, and email webhook integration.

GmailApp via Google Apps Script

Sends vendor reconciliation request emails from the configured Google account.

Google Drive link / Google-hosted vendor flow

Used as the public-facing vendor submission and file-sharing route through Apps Script / Drive configuration.

CSV / Excel file upload

Source for company-side AP/vendor ledger data.

Shape

Solution — How the VRP Works

1. Vendor Onboarding & Secure Upload Each vendor receives a unique login to the portal. They upload their ledger statement in any format — PDF, Excel, CSV, or image. The portal accepts all without requiring any standardization from the vendor's side.

2. AI-Powered Ledger Parsing The API acts as the intelligence layer. Regardless of whether the vendor sends a scanned PDF or a multi-column Excel, AI reads, understands the document structure, and extracts line-item data — invoice number, date, amount, debit/credit — into a standardized JSON format.

3. Automated Reconciliation Engine The extracted vendor data is matched against the open items pulled for that vendor. The system applies a three-level matching logic:

Company/AP ledger is the base and vendor ledger is compared against it.

Closing balance must match within Rs 1; otherwise reconciliation is not closed.

Invoices match mainly by normalized invoice number, then gross value, date, GST, and TDS.

Score 82%+ is matched, 45%-81% is partial, and below 45% is unmatched.

Payments, credit notes, debit notes, duplicates, and missing entries are separately checked and flagged.

4. Discrepancy Report Generation A structured reconciliation report is auto-generated, showing matched items, unmatched items (with reasons), and the net reconciling difference. This is stored in Local Folder with a full audit trail.

5. Automated Email Communication via Google Script The system dispatches a personalized email to the vendor with the reconciliation report attached — without any manual intervention from the finance team. Escalation reminders are also triggered if the vendor does not respond within a defined SLA.

Shape

Benefits Delivered

Metric

Before VRP

After VRP

Time per vendor reconciliation

45–60 minutes

3–5 minutes

Vendor ledger formats supported

Only standard Excel

Any format (AI-parsed)

Audit trail availability

Nil (email-based)

Complete (SQLite logs)

Month-end reconciliation cycle

4–5 days

Same day

Human error risk

High

Near zero

Shape

Innovation & Differentiators

Format-agnostic AI parsing: Unlike conventional reconciliation tools that require templated inputs, VRP accepts any vendor ledger format — a true enterprise-grade capability

Fully self-built by a CA: Designed, developed, and deployed by a Chartered Accountant — demonstrating the expanding role of CAs as finance technology architects

SAP-integrated workflow: Directly aligned with SAP HANA data without any middleware cost

Data privacy: Runs on a locally hosted or company-controlled server — no vendor data is shared with third-party SaaS platforms

Scalable across plants: The same portal can be extended to cover all type of organisation codes and vendor bases with minimal configuration

Local Machine Learning: The local machine also learns from the AI results leading improvement in the solution provided

Shape

Conclusion

The AI-based Vendor Reconciliation Portal demonstrates how finance professionals can leverage cutting-edge AI tools to solve real, high-stakes operational problems — without dependence on expensive ERP add-ons. By combining the AI API with a robust Node.js backend, local database, and Gmail automation, this solution transforms a process that once took days into one that completes in minutes.

This use case reflects ICAI's vision of Chartered Accountants as digital transformation leaders — capable of not just using AI, but architecting AI-native solutions for the Indian finance ecosystem.