When a trading company scales to hundreds of vendor transactions a month, the cracks in manual back-office processes stop being inconveniences and start being operational liabilities. Every hour a staff member spends copying data between systems is an hour not spent on anything that moves revenue.

• ⏱ Staff spending 3+ hours every morning compiling a daily trade summary from 4 separate Excel sheets — before any actual work could begin.
• 📧 Vendor invoices arriving by email, WhatsApp, and SMS — each requiring manual data entry into their accounting system. One staff member's only job was copying invoice numbers.
• 📦 Stock reconciliation done by physically walking warehouses and updating a shared Google Sheet — 2 hours daily, error rate around 12%.
• 🔔 No alerting system — low-stock situations were discovered only when a customer order failed to fulfil, causing rush restocking at premium prices.
• 📊 MD required a daily briefing PDF that took 3 hours to compile — by which point decisions being made were based on yesterday's data, not this morning's.

KLYX designed a 3-layer automation stack. Layer 1: an email/WhatsApp ingestion pipeline that captured all vendor communications and fed them into a normalisation queue. Layer 2: an AI-assisted invoice parser (Python + GPT-4o) that extracted vendor name, invoice number, line items, amounts, and GST breakdowns from PDFs and images with 94% accuracy. Layer 3: an n8n orchestration layer that connected the parsed data to their accounting software, updated stock levels via API, triggered WhatsApp alerts when stock fell below threshold, and assembled the morning brief automatically at 7:30 AM.

1. Mapped every manual task to its trigger, data source, destination, and frequency — built a workflow dependency graph before writing a single line of code.
2. Set up n8n self-hosted instance on a ₹800/month Hetzner VPS — vendor lock-in avoided, full control.
3. Built an email webhook listener that classified incoming vendor emails by type (invoice / PO / inquiry) using a lightweight NLP classifier.
4. Integrated WhatsApp Business API to capture vendor invoices sent over WhatsApp — the single most common channel for their top 20 vendors.
5. Trained a Python+GPT-4o invoice extraction pipeline on 200 historical invoices — handling scanned PDFs, photos of printed invoices, and digital PDFs in Hindi and English.
6. Built the stock reconciliation module: warehouse staff scan QR codes on stock, data hits an API, n8n reconciles against POs and flags discrepancies.
7. Built the morning brief generator — n8n pulls 6 KPIs, formats a Slack message with colour-coded status indicators, posts at 7:30 AM daily without human input.
8. Deployed with monitoring — all automation nodes emit structured logs, failures trigger instant WhatsApp alerts to the ops manager.

• Zero missed low-stock alerts in the first 90 days of operation — previously averaging 3 missed restock events per month.
• Invoice processing backlog eliminated — from a 48-hour backlog to same-day posting.
• Payback period: KLYX's build cost recovered within 58 days of deployment.

• Map before you build — every hour spent mapping the workflow dependency graph saved three hours of rework. Automation that isn't built on a precise understanding of the manual process will automate the wrong thing.
• Self-hosted n8n beats SaaS for high-volume Indian businesses — per-task pricing on Zapier/Make becomes prohibitive at 500+ automations/day. A ₹800/month VPS running n8n handles unlimited tasks.
• AI parsing + human review is the right starting point — 94% accuracy means the AI handles 470 out of every 500 invoices perfectly; 30 go to a human review queue. That's still a 94% time saving, without the risk of fully automated errors in financial records.