Machine Learning for Selecting Letter Recipients
Smarter mailings with machine learning
Score your donor list before you print. We train a model on your past mailings and donations, so you can send fewer, better-targeted letters and save postage, paper and CO₂ — guided by a clear ranking.
Your benefits at a glance
Optimise the number of letters
Send the letters most likely to be worthwhile, not everyone by default. That means less postage, paper and CO₂ — and more surplus for your work.
Use scoring to tailor outreach
Every recipient gets a clear score. Use the ranking to build priority groups, tailor letters and vary your ask where it counts.
Review every recommendation
See why each person ranks where they do, where the model is confident, and where your judgement should still decide.
Learn which signals matter
See which patterns in your data drive stronger returns — insight that outlasts any single campaign.
Why is it worthwhile?
Fundraising letters still work, but printing and posting them to everyone is expensive — financially and environmentally. A model helps you optimise how many letters to send for each campaign by scoring potential recipients for that specific mailing.
That score is useful beyond a simple yes/no list. You can create priority groups and use the ranking to tailor the message, format or ask. The model does not replace your fundraising judgement; it gives you a clearer basis for using it.
Our approach
We connect to your fundraising data, look at past mailings and donations, and produce a scored recipient list for the next campaign. You stay in control of mailing size, segmentation and final export. After the mailing, the results flow back in so the model and the strategy improve over time.
1
Connect the data — donation history, previous mailings, response data and the fields you already use.
2
Score each recipient — the model ranks each potential recipient for this specific mailing.
3
Review and segment — inspect the reasons, choose your mailing size, and export the groups you want to contact differently.
4
Send and learn — run the mailing, import the results, and see which signals mattered most.
+6.3%
more surplus for one client’s letter campaign. The model is retrained continuously, so the organisation mails more precisely and profitably.
See what a model could save your next campaign
Get a quick sense of whether your mailing data is ready, what we would need, and where the savings are likely to be.
Possible objections
Privacy?
For mailing selections, your supporter data never goes to ChatGPT, AWS or other big-tech clouds. We use open-source tools on encrypted machines in our own office.
Black box?
You get more than a list of names. We show why people ranked higher or lower, which fields lifted the score, and where the model is unsure — so your team can check the output and learn from it.
Systems we’ve already connected
We have already connected these fundraising and communication systems for NGO work. If your data lives somewhere else, we can usually build a focused connector too. We import only the fields needed for the analysis — not names, addresses or bank details unless explicitly agreed.
German article
In September 2025 the German Fundraising-Magazin published our approach. (Sorry, German only.)
Among the organisations we work with on this
