Skip to content

README.md

Latest commit

 

History

History
169 lines (119 loc) · 6.25 KB

README.md

File metadata and controls

169 lines (119 loc) · 6.25 KB

Healthcare Clinic Optimization — AI Strategy Dashboard

Report 1 End-to-end AI strategy simulation for a 5-clinic outpatient chain — no-show reduction, smart overbooking, patient segmentation, and automated reminders.

Overview

This project implements the full business strategy from Report 1: Healthcare Clinic Optimization as an interactive Streamlit dashboard. The chain currently loses ₹10.92 lakh/month to a 30% no-show rate — above India's outpatient benchmark of 15–20%. Every recommendation targets that gap without adding a single doctor, clinic, or price change.

The dashboard runs six strategy modules end-to-end, from current-state revenue analysis through a live ML-powered appointment risk scorer, and projects a 12-month path from ₹25.5L to ₹32–34L/month in recovered revenue.

The Business Problem

Parameter Value
Clinics 5
Scheduled appointments/day 200
No-show rate 30% (benchmark: 15–20%)
Actual patients seen/day 140
Revenue per visit ₹700
Monthly revenue lost ₹10.92L
Technology cost to fix it ₹8,000–18,000/month
Estimated ROI 50–100×

The chain is not capacity-constrained. It is engagement-constrained.

Modules

Module 1 — Current State Analysis

Calculates actual vs. theoretical revenue, slot utilisation, and the cost of inaction across four no-show rate scenarios (30% → 22% → 14% → 0%).

Module 2 — No-Show Prediction Model

GradientBoosting classifier trained on 6,000 synthetic appointment records. Outputs a 0–1 risk score per appointment and routes it to a three-tier reminder protocol:

Risk Band Score Reminders Channel
Low < 0.30 1 WhatsApp
Medium 0.30–0.60 2 WhatsApp + SMS
High > 0.60 3 WhatsApp + SMS + Call

Includes a live risk scorer — enter patient details and get an instant band and protocol recommendation.

Module 3 — Smart Overbooking Engine

Calculates conservative (10% cap) overbooking buffers per time band based on historical no-show rates. Runs a Monte-Carlo waitlist fill simulation across 26 working days.

Module 4 — Patient Segmentation (RFM)

Recency–Frequency–Monetary segmentation across the patient base into four behavioural cohorts — Champions, At-Risk, High No-Show, and New Patients — each with a differentiated WhatsApp engagement strategy.

Module 5 — Automated Reminder Scheduler

Generates a full daily 200-slot reminder schedule, computes per-channel costs (WhatsApp / SMS / Call), and shows the band distribution and cascade logic.

Module 6 — KPI Dashboard

12-month dual-axis projection (revenue + no-show rate), revenue waterfall broken down by intervention, and a full phase-target scorecard (Baseline → 3-Month → 12-Month).

Projected Outcomes

Baseline 3 Months 12 Months
No-show rate 30% ≤ 22% ≤ 14%
Patients/day 140 155 170–176
Monthly revenue ₹25.5L ₹29–30L ₹32–34L
Slot utilisation 70% 77% 85–88%
Revenue uplift — +₹3.5–4.5L +₹6.5–8.5L

Project Structure

├── healthcare.py                  # Streamlit dashboard (all 6 modules)
├── Healthcare Report.ipynb      # Core strategy implementation (Matplotlib version)
├── README.md
└── Healthcare Clinic Optimization_Report_1.pdf   # Full business report

healthcare.py is the primary deliverable — a self-contained interactive dashboard. Healthcare Report.ipynb contains the same six modules as a standalone script that generates a static PNG master dashboard.

Setup

Requirements

Python 3.9+
streamlit
plotly
scikit-learn
pandas
numpy

Install and run

git clone https://github.com/<your-username>/healthcare-clinic-optimization.git
cd healthcare-clinic-optimization

pip install -r requirements.txt

streamlit run healthcare.py

Or run the static report generator

python Healthcare Report.ipynb
# Outputs: clinic_optimization_dashboard.png

requirements.txt:

streamlit>=1.35.0
plotly>=5.20.0
scikit-learn>=1.4.0
pandas>=2.0.0
numpy>=1.26.0

Dashboard Features

  • Sidebar scenario controls — adjust no-show rate, revenue per visit, waitlist fill rate, and overbooking cap live; all charts update instantly
  • Live ML risk scorer — enter 8 appointment features and receive a risk score + reminder protocol in real time
  • Plotly interactive charts — hover, zoom, and export on all 10+ visualisations
  • Raw data toggle — expose underlying DataFrames for any module
  • Dark clinical theme — Syne + Mulish typography, teal/navy palette

Key Design Decisions

Why GradientBoosting over Logistic Regression? The non-linear interaction between past no-show count, lead time, and distance is better captured by an ensemble. ROC-AUC consistently lands at 0.74–0.78 on the synthetic dataset, sufficient for risk-band routing.

Why synthetic data? The model architecture, feature set, and risk-band thresholds are production-ready. Real deployment requires substituting the synthetic DataFrame with a clinic's historical appointment export (6–12 months minimum).

Why a 10% overbooking cap? Aggressive overbooking risks wait-time complaints and patient attrition — the single highest-cost outcome for a price-sensitive patient base. The cap is parameterised and can be raised incrementally as validation data accumulates.

Recommended Tool Stack (Production)

Layer Tool Cost/Month
WhatsApp automation Wati / Interakt / AiSensy ₹2,000–5,000
SMS fallback MSG91 / Textlocal Pay-per-message
Voice reminders Exotel / Sarvam AI IVR ₹3,000–8,000
ML prediction Google Colab / DataRobot Free
CRM & segments Zoho CRM / Airtable Free–₹1,500
Scheduling Practo / Sheets + Zapier Free
Analytics Looker Studio Free
Total ₹8,000–18,000

Author

Abhishek Sharma Strategy & Analytics Division · May 2026

All projections are model estimates based on stated operating parameters and India outpatient healthcare benchmarks. Simulation uses synthetic data representative of clinic appointment patterns.