Michael Belli

Insurance Marketing Analytics Decision Engine

← Back to Home

---

This is a self-directed demonstration project. It walks an end-to-end analytics workflow for a B2C insurance company — from exploratory analysis through predictive modeling to budget optimization and business-impact measurement — to show how I approach the work.

Note on the data: Everything here runs on a synthetic dataset I generated to mirror realistic insurance-marketing dynamics. The figures throughout are outputs of that simulation, included to illustrate the methodology end to end — not results from a client engagement.

View Repository →

---

The Business Problem

In insurance, growth without risk discipline destroys value. Marketing teams optimize for lead volume and cost-per-lead, but cheap leads often become unprofitable policies. This project works through three questions:

1. Which marketing channels drive profitable growth—not just volume?
2. How do we model the diminishing returns of marketing spend?
3. Where should we reallocate budget to maximize lifetime value?

---

1. Exploratory Data Analysis

Eight exploratory analyses reveal the key dynamics of insurance marketing: Credit score strongly predicts both conversion and claims risk Channel quality varies: cheaper leads have higher loss ratios Cross-sell customers have 2x higher lifetime value Geographic variation requires state-level pricing adjustments The EDA surfaces the core insight: email has the highest ROI despite the lowest lead quality, because its acquisition cost ($8/lead) is dramatically lower than paid search ($45/lead). View Full EDA →

Click for details

---

2. Marketing Mix Model

The Marketing Mix Model (MMM) quantifies the relationship between spend and conversions using Hill saturation curves: Response curves capture diminishing returns at higher spend levels Unit economics (avg profit per conversion) vary by channel ROI-saturation constraint ensures high-ROI channels aren’t mistakenly labeled as “saturated” The model reveals that email is operating at just 8% of its half-saturation point—significant room to scale—while search and social are both near 70%. View Full Model Documentation →

Click for details

---

3. Budget Optimization

Using the fitted response curves, we simulate two scenarios: Period 1 (Current): Historical budget allocation Period 2 (Optimal): Budget reallocated to equalize marginal ROI across channels With the same total marketing spend, the optimal allocation shifts ~$4,800/month out of search and social into under-saturated email, yielding: +12.8% more conversions +5.9% higher profit 11% lower customer acquisition cost View Optimization Details →

Click to enlarge

---

4. Business Impact

Within the simulation, reallocating the same modeled $308K annual marketing budget produces:

Metric	Before	After	Change
New Customers	957	1,079	+12.8%
Lifetime Revenue	$18.4M	$20.8M	+$2.4M
Policy Profit	$3.49M	$3.69M	+$205K
CAC	$322	$286	-11%

(Profit rises far less than conversions because the reallocation leans into email — high ROI on acquisition cost, but lower-margin policies. Surfacing that trade-off is exactly the point of the model.)

---

Technical Implementation

Data Generation: Python script creating synthetic but realistic insurance marketing data with:

• 3 years of lead data across 3 products (Health, Life, Property/Casualty)
• 3 marketing channels with quality/cost trade-offs, where daily spend varies widely (budget regimes and experiments) and leads follow a saturating response to spend
• Full sales funnel with claims simulation, plus cross-sell dynamics (bundled customers convert better and retain longer)

Modeling: Hill saturation functions with an ROI-based constraint, fit on monthly-aggregated data using nonlinear least squares.

Optimization: Profit-maximizing budget allocation using scipy.optimize.

---

← Back to Home