## CRM Automation Suite — A sales pipeline that follows up on its own.

- URL: https://www.naveentehrpariya.dev/projects/crm-automation
- Industry: B2B SaaS
- Client type: Agency
- Country: Australia
- Role: Full Stack Product Engineer
- Duration: 8 months
- Status: Live
- Year: 2024

A CRM built around one idea: the follow-up should happen whether or not a salesperson remembers. Leads arrive from every channel, get scored, and move through sequences that pause the moment a human replies.

### Problem

The agency's leads landed in four places: a web form, an inbox, a phone log and a spreadsheet. Reps worked whichever list was in front of them, follow-ups depended on memory, and management had no view of the pipeline until the month closed.

### Objectives

- Bring every lead source into one queue
- Automate follow-up sequences without spamming replying leads
- Score leads so reps work the right ones first
- Give management live pipeline visibility

### Solution

Laravel handles the domain and background queues; a Next.js dashboard handles the interface. Every lead source normalises into one lead record, and automation runs as queued jobs against a state machine, so a reply, a booking or a manual note immediately halts the sequence. Reporting reads from pre-aggregated tables instead of scanning the activity log.

### Features

- Unified lead inbox across web, email and phone
- Visual pipeline with drag-to-stage
- Email and WhatsApp sequences that stop on reply
- Rule-based lead scoring
- Role-based access for reps, managers and admins
- Live pipeline and conversion reporting

### Stack

  - Frontend: Next.js, React, Tailwind CSS
  - Backend: Laravel, PHP, Horizon
  - Database: MySQL, Redis
  - Cloud: AWS, Docker
  - Integrations: Twilio, SendGrid, Google Calendar, Zapier

### Engineering challenges

### Automation that kept emailing people who had already replied

Problem: Sequence steps were scheduled up front, so a reply on day two did not stop the email queued for day three.

Approach: Each step re-checks lead state at execution time rather than at scheduling time, so any human touch cancels everything downstream.

### Reports that got slower every month

Problem: Dashboards aggregated over the full activity table, which grew with every interaction the system recorded.

Approach: Aggregates are rolled up on write into summary tables, so dashboard queries stay flat regardless of history size.

### Outcome

- Four lead sources consolidated into a single queue with no manual re-entry
- Follow-up stopped depending on individual memory
- Management moved from month-end reporting to live pipeline visibility

### What it taught

Schedule the intent, not the action. Re-evaluating state at execution time is what separates automation people trust from automation people switch off.

Case study: https://www.naveentehrpariya.dev/projects/crm-automation
More work: https://www.naveentehrpariya.dev/llms.txt