DONOR BEHAVIOR ANALYSIS & FORECASTING

DRIVING IMPACT WITH DATA: INSIGHTS FOR A NOT-FOR-PROFIT ORGANIZATION

BY KAGNA EM

​

© Kagna Em 2025. All rights reserved.

PROJECT OVERVIEW

METHODOLOGY

DATA FLOW/PIPELINE (ELT)

DATA FLOW/PIPELINE (ELT)

• Below is the process used to create the Data Pipeline with Microsoft Fabric (ELT approach) in the Microsoft 365 environment.
• The process began in Power BI Service (Online) within an environment where Microsoft Fabric was enabled

Loading data to Power BI for Data Modeling and Visualization

Power BI Desktop was used to load data from the created pipeline for data modeling, developing advanced DAX measures and table calculations, and building interactive visuals and dashboards.

LARGE DATASET TRANSFORMATION

This is how the cleaned and structured data tables appear. After filtering, over 800,000 records were successfully transformed in minimal time and are ready for use.

REPORT DASHBOARD

Donor Profile – Overview of Donor Characteristics

Donation Trend – Overview of Donor Contributions

DONOR BEHAVIOR ANALYSIS

​

• Spanning five fiscal years, the bar chart and line chart illustrate the quarterly total number of donors categorized by their donation behaviors: new, increasing, decreasing, retained, or lapsed.

DONOR BEHAVIOR DETAIL

• Viewing detailed information about donors by behavior within each fiscal quarter adds further insight. Clicking on a bar in the chart reveals donor details below, while the slicer in the bottom left corner allows filtering by specific donation behavior.

DONOR MAPPING

• The final product is a combined dataset integrating donor information with external data sources, including boundary maps from LINZ and property data from OneRoof.
• This interactive map displays over 1000 suburb boundary layers across New Zealand, visualizing donor information alongside the average property price in each donor’s location.
• A slicer in the top-right corner allows users to filter the map by property price segments.

RFM ANALYSIS

RFM analysis segments donors based on their giving patterns—Recency (how recently they donated), Frequency (how often they donate), and Monetary (how much they give). This helps organizations focus efforts on the most engaged and valuable supporters, making it a powerful tool for optimizing outreach and increasing donor loyalty

FORCASTING

• This line chart is used for displaying actual donations over the past five fiscal years compared to forecasted donations for the next four quarters. See the next figure for a comparison among 3 different forecasting methods: Average, Linear Regression and Phophet

COMPARATIVE ANALYSIS OF THREE FORECAST MODELS

Prophet Model

Forecasting future donations using the Prophet model proved most effective given the data's complexity. This method is often considered effective—especially for business time series—because:

• Automatic Handling of Seasonality and Holidays – Captures recurring patterns and allows custom holiday inputs, making it ideal for business and donation trends.
• Robust and Flexible – Works well with missing data, outliers, and supports sudden trend changes using changepoints.
• Easy to Use and Interpretable – Requires minimal tuning, is open-source, and produces clear visual components for trend, seasonality, and forecasts.

Using Python Script

• To use this model in Power BI, a Python script must be embedded. The script uses Prophet as the core forecasting algorithm and Matplotlib for data visualization. The process involves preparing the data in Power BI, passing it to Python for forecasting, and then returning the forecasted results back to Power BI for display in interactive visuals.

Using DAX

This analysis would not have been possible without the use of DAX for creating calculated tables and measures.

FORECAST AT THE DONOR/SEGMENT LEVEL

• The slicer is a valuable tool for enabling drill-down functionality—either by donor group (the left figure) or by individual donor account (the right figure).

SUMMARY

THANK YOU