Home Services Portfolio About Get Your Free Analysis →
Portfolio

Real work.
Real results.

Three projects showing exactly what Luma Analytics delivers — the insight, the numbers, and what action was taken because of them.

Project 1 — Meta Ads ROAS (Live)
Nonprofit · Live Project · Real Data

University Donor Intelligence Dashboard

Analyzed 34,508 university donor records using an end-to-end machine learning pipeline. Three predictive models — churn risk, lifetime value, and next-gift forecasting — delivered through a live Metabase dashboard with automated donor prioritisation.

Python Random Forest pandas PostgreSQL Metabase
Dataset34,508 donor records
Models built3 (churn, LTV, next gift)
Revenue at risk$287K identified
53% of donors flagged as high churn risk
The churn model scored every donor by lapse probability — surfacing $287K in immediate revenue at risk that the team had no visibility into before.
$102M in predicted lifetime donor value
A lifetime value model estimated the long-term worth of every donor — so the team could prioritise outreach based on who matters most, not just who gave most recently.
$7.3M projected next-year revenue
A third model forecasted expected giving for the coming fiscal year per donor — giving the fundraising team a data-backed revenue projection for the first time.
Automated outreach prioritisation table
Every donor gets a recommended action — Re-engagement Email, Personal Call, Upgrade Ask — ranked by churn probability and revenue at stake, updated live.
Live Dashboard
Donor Summary — 34,508 donors, $287K revenue at risk, $7.3M projected Donor Risk Distribution and High Risk Donor Table Donor Scoring Matrix — Churn vs LTV Churn Score Distribution and 5 Year Giving Trends
Small Business · Live Project · Real Data

Meta Ads ROAS Prediction Dashboard

A live end-to-end analytics project built on real Meta/Facebook ad data. A Random Forest model predicts return on ad spend per campaign — actual vs predicted ROAS shown side by side in a live Metabase dashboard connected to PostgreSQL.

Python scikit-learn Random Forest pandas PostgreSQL psycopg2 Metabase
Data sourceMeta Ads Manager
PredictsAd return on spend
OutputLive dashboard
Real Live Meta ad data
29 Campaign features modelled
Live Metabase dashboard

What was built

Real ad data, not samples
Started with a raw Meta Ads Manager export — messy, real-world data — and cleaned it into something actually usable.
Predicted which campaigns would win
Built a model that looks at how a campaign is performing now and forecasts its return — so you know where to put budget before it's too late.
Everything updates automatically
Predictions feed directly into a live database — no manual exports, no spreadsheets. The dashboard always reflects the latest numbers.
A dashboard a team can actually use
Actual vs predicted return shown side by side — built so a marketing team could open it every morning and know exactly where to focus.
Live Dashboard
AI Marketing Analytics Dashboard — Campaign Overview AI Predictions — Actual vs Predicted ROAS

Want this for your business?

I build these for free for nonprofits and small businesses. All I need is your data.

Get Your Free Analysis →