Efficient learning of donor retention strategies for the American Red Cross

We present a new sequential decision model for adaptively allocating a fundraising campaign budget for a non-profit organization such as the American Red Cross. The campaign outcome is related to a set of design features using linear regression. We derive the first simulation allocation procedure for simultaneously learning unknown regression parameters and unknown sampling noise. The large number of alternatives in this problem makes it difficult to evaluate the value of information. We apply convex approximation with a quantization procedure and derive a semidefinite programming relaxation to reduce the computational complexity. Simulation experiments based on historical data demonstrate the efficient performance of the approximation.