Plan Canada acquires new monthly donors through a range of marketing channels and external vendors. Each source comes with different costs—and the donors acquired behave differently over time:

• Varying monthly donation amounts
• Different long-term churn rates
• Distinct patterns of ongoing engagement

Comparing the ROI of acquisition channels was complicated by a key challenge: You can't directly measure a donor’s lifetime value (LTV) until they churn—which could take years or even decades.

Without a way to estimate LTV early, Plan risked over- or under-investing in acquisition strategies, reducing the efficiency of its fundraising operations.

CVM partnered with Plan Canada to design and deploy an AI-powered predictive model that estimates donor lifetime value based on early behavioral signals—enabling faster, data-driven acquisition decisions.

Key aspects of the solution included:

• Neural Network Modeling with PyTorch
  Built a custom deep learning model using PyTorch to estimate LTV. The model captured complex, non-linear relationships across time-series signals such as initial donation behavior, engagement frequency, acquisition source, and demographics.

• Early Signal Extraction
  Engineered features from the first 60–90 days of donor behavior to enable high-confidence LTV predictions long before actual churn occurred.

• Integrated Data Pipeline
  Combined vendor acquisition data with internal donor records to create a unified dataset for training, evaluation, and ongoing prediction.

• Transparent Model Governance
  Incorporated stakeholder feedback into model development and emphasized explainability (e.g., feature importance scoring) to ensure alignment and build trust across fundraising teams.

• Production Integration
  The model was deployed within Plan’s donor management ecosystem using Python, allowing for automated scoring of new donors and seamless updates as new data became available.

• Smarter Acquisition Decisions
  Plan could now evaluate the projected ROI of each channel or vendor shortly after donor signup—rather than waiting years for lifetime data.

• Improved Fundraising Efficiency
  Budgets were shifted toward higher-LTV acquisition sources, maximizing long-term returns.

• Faster Time-to-Insight
  Early predictions enabled marketing and fundraising teams to act on data within months instead of years.

• Increased Trust in AI
  A transparent, collaborative development process helped drive adoption and confidence among non-technical stakeholders.

• AI Can Predict the Future, Responsibly: Neural networks are powerful tools for forecasting lifetime behavior—even in uncertain or incomplete data environments.
• Transparency Drives Adoption: Explainable AI techniques and stakeholder engagement are critical to building trust in predictive models.
• Iterative AI Pays Off: As more donor lifecycle data becomes available, models are retrained and improved to increase accuracy and relevance over time.