One of our most interesting recent projects was building a complete multi-level approval workflow that operates entirely through WhatsApp. Here's how we designed, built, and deployed it.
The Problem
Our client, a manufacturing company, needed an approval system for purchase requests. Their existing process involved:
- Employees submitting requests via email
- Managers checking email (when they remembered)
- Finance teams losing track of pending approvals
- CEO approvals getting delayed by days
The result: a 5-7 day average approval cycle that was costing the company real money in delayed purchases and frustrated employees.
The Solution: WhatsApp-Native Workflows
We built an AI-powered approval system that operates entirely within WhatsApp. Here's the architecture:
System Architecture
The system consists of three core components:
- WhatsApp Business API Integration — Connected via the Meta Cloud API for reliable message delivery
- AI Intent Router — A natural language processor that understands commands like "approve request #42" or "what's pending?"
- Workflow Engine — A state machine that manages the approval chain: Department Head → Finance → CEO
The Approval Chain
When an employee submits a request via WhatsApp:
- The AI validates the request format and confirms with the employee
- The request is sent to the Department Head with full details
- Upon approval, it automatically forwards to Finance
- Finance approval triggers the final CEO review
- At each step, all parties receive real-time status updates
Technical Implementation
We built the system using:
- Python + FastAPI for the backend API
- SQLite for request persistence (lightweight, zero-config)
- Meta WhatsApp Business API for messaging
- Intent-based routing to distinguish between commands and regular chat
The intent router was crucial. We needed the system to understand that "yes approve it" and "approved" and "looks good, go ahead" all mean the same thing, while "what's the status of request 5?" is a completely different action.
Key Design Decisions
1. Confirmation Step for Every Action
We added a mandatory confirmation step before any approval or rejection. This prevents accidental triggers — a manager casually saying "sure" in a group chat won't accidentally approve a $50,000 purchase request.
2. Sequential, Not Parallel
We chose sequential approval (one approver at a time) over parallel approval. This ensures each approver has full context of previous approvals and can make informed decisions.
3. Real-Time Notifications
Every stakeholder receives updates at every step. The requestor knows exactly where their request is in the pipeline at all times. No more "let me check on that for you."
The Results
After deploying the WhatsApp approval system:
- Average approval time dropped from 5 days to 4 hours
- 100% of requests are tracked — nothing falls through the cracks
- Zero training required — everyone already knows how to use WhatsApp
- Mobile-first by design — approvals happen from anywhere, anytime
Lessons Learned
- Keep the interface conversational — Don't make users learn commands; let them type naturally
- Always confirm destructive actions — One wrong approval can be expensive
- Build for offline resilience — WhatsApp messages queue when users are offline; your system should handle delayed responses gracefully
- Log everything — Complete audit trails are non-negotiable in enterprise workflows
What's Next
We're now extending this pattern to other workflows: leave requests, expense reports, and vendor onboarding. The fundamental architecture — WhatsApp + AI intent routing + state machine — is remarkably flexible.
If your business still runs approvals through email chains and spreadsheets, it's time for an upgrade. The tools exist, the technology is mature, and the ROI is immediate.