The question banks are actually asking
Every large Dutch bank runs a central decisioning engine — for several of them it is Pega’s Customer Decision Hub, deciding in milliseconds which message a customer sees in the app, the email, the branch conversation. The technology was built to answer “what can we sell this person?” But that is no longer the question Dutch banks are funded, regulated, and publicly committed to answer.
The question now is: “what does this person need?” The AFM’s duty of care (zorgplicht) expects banks to act in the customer’s interest. Bank mission statements promise financial health. And commercially, a customer whose overdraft you quietly prevented stays for decades; a customer who got a fifth credit-card offer during a difficult month does not.
Here is the part I find genuinely exciting as a decisioning specialist: the machinery doesn’t have to change — the values plugged into it do. And there are four specific places where that happens.
1. Arbitration is where your values become math
CDH ranks every candidate action with one formula: propensity × context weight × value × business levers. Most estates fill “value” with expected revenue — which means a savings-buffer nudge can never beat a loan offer on points, no matter who is looking at the screen.
A duty-of-care estate prices avoided harm into that same field: prevented overdraft fees, prevented arrears trajectory, long-term retention. Suddenly “start an emergency buffer” can win arbitration for the customer living paycheck to paycheck — not because a rule forced it, but because the economics finally include the customer’s side. The formula is unchanged. The bank’s definition of value is what got an upgrade.
2. Propensity tells you who will say yes — not who you helped
The standard model behind every action predicts acceptance. But for financial-health actions, acceptance is the wrong trophy. Some customers would have opened that savings account anyway; nudging them wastes attention. The customers worth finding are the ones who act because you asked — and identifying them is a different discipline called uplift modeling, measured with treatment-versus-control comparisons rather than raw response rates.
For sales, propensity and uplift often point the same way. For care, the gap between them is the whole point: helping means changing outcomes, not predicting them. A decisioning team that can articulate this — and design champion/challenger tests that measure genuine impact — is doing duty of care with evidence instead of intentions.
3. Calibration is a consumer-protection feature
A less obvious one, and my favorite, because it’s pure mathematics with regulatory consequences. Adaptive models output propensities that get multiplied with value and levers. If a model is overconfident — saying 12% when reality is 6% — it silently inflates the priority of its own actions and distorts every ranking behind it. Nobody notices, because the dashboard metric everyone watches (discrimination, AUC) can look excellent while calibration is broken.
In a sales estate, miscalibration costs money. In a duty-of-care estate, it means the helpful actions systematically lose arbitration to overconfident sales models — the exact failure mode the regulator would ask about. Reliability curves belong on the same dashboard as AUC. (For the technically curious: I demonstrate this class of model-health analysis on a public dataset in my decisioning health check.)
4. Contact policies are the quiet heroes
The least glamorous configuration in CDH — suppression rules, contact limits, “stop after three ignores” — is where customer respect is actually enforced. Interaction History summaries make it precise: patterns like ignored this action three times in two weeks become hard suppressions, protecting customers from the drip of irrelevant messages that erodes trust one notification at a time.
I’d argue contact policies deserve the same governance ceremony banks give credit models: owned, reviewed, tested. Over-contact is not a tuning issue; under zorgplicht it’s a conduct issue.
The team this requires
None of the above is a product you can buy; all of it is judgment applied to a platform banks already own. It needs people who are bilingual — fluent in the platform (arbitration, adaptive models, engagement policies, Interaction History) and in the mathematics underneath (uplift, calibration, controlled experiments), with the governance instinct to know which decisions must stay explainable and reviewable.
That intersection is precisely where I work. If your decisioning team is wrestling with how to make “customer interest first” true in the arbitration math and not just the annual report — I’d genuinely enjoy that conversation.