CphML An applied AI and Machine Learning Company
Direct Mail Model · Dansk Røde Kors · Live v01 / 2026

Contact the
people who will actually
say yes

We score every contact in your next campaign with a gradient-boosted model - so your direct mail and outreach budget only goes to the audience most likely to respond. Same revenue, a fraction of the spend.

RANKED LEADS · OUTPUT n = 148,200
# ACCOUNT tx interval segment score contact
01 d_8841 12d a7 0.94 1
02 d_3019 31d a4 0.88 1
03 d_9903 18d b3 0.81 1
04 d_2287 44d a9 0.74 1
05 d_5612 62d b1 0.68 1
t = 0.62 · cut-off
Threshold T is chosen to maximize expected ROI.
Expected return at T = 0.62
kr 2.4m
Projected return
contact12,400 skip135,800
peak · max expected ROI projected return kr 2.4m
contact 12,400 skip 135,800
Gradient-boosted propensity models
XGBoost · cross-validated · SHAP-explainable
Tele marketing / Web / Postal / Direct sales
Any channel where you control who receives.
Live model
Currently running with Dansk Røde Kors donor campaigns.
CPH
Based in the Innovation District in Copenhagen. GDPR compliant · EU data residency
◆ 01 - Process

A score, a threshold,
a smaller print run

1 DATA
contact_id recency freq monetary channel d_8841 12 6 1.240 post d_3019 240 1 200 post d_9903 31 14 3.400 email d_2287 410 2 450 post

We ingest your data.

Existing customer or donor records - purchase history, engagement signals, demographics - cleaned and assembled into a feature set we can train on. SFTP, CSV, or warehouse export.

zero engineering on your side
2 MODEL

We train a classifier.

An XGBoost model fits on your historical conversions. Output: a probability score for every contact in your next campaign audience. We share feature importance so the model isn't a black box.

XGBoost · cross-validated · explainable
3 TARGETING
τ

Target the top-scored contacts

Pick a threshold based on your CPM and expected conversion. Mail or serve ads only above the cut-off. Same fundraising or revenue, fraction of the print, postage, or media spend.

Decide the slice that fits your economics
◆ 02 - Who it's for

Anyone running volume
campaigns where every contact
has a cost

If you're paying per recipient - postage, print, ad impressions, calls - and most of them don't convert, scoring before you spend pays for itself in the first send.

NGO fundraising

Donor mail-outs, newsletters, recurring-gift upgrades. Cut volume without cutting yield.

DONOR APPEALS · LOTTERY

Retail & loyalty

Catalogues, win-back coupons, churn campaigns. Predict the customers worth reactivating.

CATALOGUE · WIN-BACK · CRM

Political campaigns

Door-knock priorities, persuadable-voter mail, GOTV cohorts. Same field budget, more contacts that matter.

DIRECT MAIL · CANVASSING · GOTV

Subscriptions

Renewal letters, retention offers, free-trial conversions. Spend on the segment that's actually on the fence.

RENEWAL · RETENTION · TRIAL
◆ 03 - Live pilot

First in production
a humanitarian one

PILOT ONGOING · 2026

Optimising donor campaigns for Dansk Røde Kors

By scoring their donor base before each send, we identify which contacts are unlikely to donate in a given cycle — reducing print and postage costs without cutting fundraising yield.

Dansk Røde Kors
National humanitarian org · est. 1876

Direct maildonor appeals
Scorebefore send
Outcomevolume ↓ yield =
◆ 04 - Get started

Ready to stop paying for contacts who won't respond

Send us your campaign setup and we'll assess fit within 1–10 business days.

Get in touch