Identify the potential donors and predict ask amount
Problem:
A lot of Manpower invested in the solicitation.
Mail cost per piece is 30-50 cents.
The response is very low.
No clue on what amount to ask for.
Data Science Challenges:
Messy data had to lot of work to understand it.
Organization insistence on including external ratings that they bought from third-party firms reduced the predictive performance of models.
Solution/Approach:
We ranked all the constituents by their probability of giving. Now, gift officers who will solicit whom, which campaign they will be part of and whom to ignore.
Predicted the ask amount for each. The accuracy of the best machine learning model was 83%. Now when the offers go out, there will be no longer random ask amounts. They will be so precise that donors are more likely to donate.
Put forward a broader strategy for fundraisers and gift officers and teach them how to use our one-click machine learning software.
Results :
Donations went up by $10,000 on average for every campaign.
Fewer call centre reps hired since fewer people needed to solicit now.
$4000 quarterly savings in mailing cost.
