The Pricing Algorithm That Punished Loyalty: A Case Study in Perverse Incentives

Opening Scene

A subscription insurance company rolls out a new pricing engine designed to optimize renewal pricing for every individual policyholder. Within two renewal cycles, an internal analyst runs a routine check and finds something uncomfortable: customers who had stayed loyal for five, six, seven years were, on average, being quoted noticeably higher renewal prices than brand-new customers with otherwise identical risk profiles. The model hadn’t been told to punish loyalty. It had simply learned, correctly, that long-tenured customers were less likely to shop around and price-check, and priced accordingly.

In Plain English

This composite case is a textbook example of a perverse incentive: an algorithm optimizing a legitimate business objective — maximizing revenue per customer — finds a technically effective but ethically corrosive way to do it, in this instance by exploiting price insensitivity rather than earning loyalty through better service or value. The model is behaving rationally by the only measure it was given; the ethical failure is in a business objective that was never checked against a simple question: is it acceptable to charge people more specifically because they trust us?

The Old Way

Before granular, individualized pricing algorithms became common:

Pricing was typically set through simpler, more visible tiers, which made large individual discrepancies more likely to be noticed and questioned internally.
There was little practice of scrutinizing whether a legitimate optimization objective could produce a specific outcome — like penalizing loyalty — that most people would find unfair if described plainly.
Customers themselves had almost no visibility into whether their price reflected their actual risk or their measured unlikeliness to switch providers.

Asking whether an optimization objective can produce a plainly unfair specific outcome is exactly the check this kind of case is meant to institutionalize.

What’s Changing (and Why AI Is the Reason)

Regulators and journalists are paying closer, more systematic attention to algorithmic price discrimination, particularly where it correlates with loyalty, age, or other protected or quasi-protected characteristics.
This connects to the responsible AI principles covered in this content library’s dedicated series, specifically the discipline of stress-testing a business objective for unintended consequences before deployment, not just after a complaint.
AI-driven personalization makes individualized pricing dramatically easier and cheaper to implement at scale than it ever was with manual, tiered pricing, which widens the gap this kind of algorithm can quietly exploit.

The Metaphor, Fully Extended

The Case File	The Perverse Incentive Concept
A crime that technically follows the letter of an old law	An outcome that technically follows a legitimate business objective
The pattern only visible once someone compares many cases	The loyalty penalty only visible once prices are compared across tenure
A motive that made sense but still didn’t excuse the outcome	A revenue objective that made sense but still produced an unfair result
The detective asking “would this hold up described plainly?”	The practitioner asking “would this hold up if we described it to customers directly?”

For Beginners: What to Actually Do

Practice asking, of any personalized price or offer you receive, whether it’s based on your actual need or your measured unlikeliness to compare alternatives.
Learn to recognize “optimizing revenue per customer” as a goal that needs an explicit fairness check, not an inherently neutral one.
Get comfortable asking simple, plain-language questions of complex systems: “would this be acceptable if we explained exactly how it works?”

For Practitioners and Leaders: The Deeper Layer

Stress-test any pricing or personalization objective against specific plausible outcomes — including loyalty penalties — before launch, not just aggregate revenue impact.
Apply the objective-scrutiny discipline from this content library’s dedicated responsible AI principles series to commercial optimization models, not only to overtly high-stakes systems.
Build a standing practice of comparing outcomes across customer tenure and other non-risk-based segments, specifically looking for unintended penalty patterns.

Quick Recap

A model optimizing a legitimate objective can still produce a plainly unfair specific outcome, like penalizing loyalty.
The algorithm behaved rationally; the unchecked business objective was the actual ethical failure point.
Individualized AI-driven pricing makes this kind of quiet exploitation dramatically easier to implement at scale.
Explicit fairness stress-testing of business objectives, not just revenue impact, is the practical safeguard.

Where This Fits in the Series

Article 6 traced a breach back to consent that was technically valid but not meaningfully informed; this article traces an unfair price back to an objective that was technically legitimate but ethically unchecked. Article 8 turns to a different domain entirely, where the stakes rise again: a health AI tool that quietly underserved a population its training data had never adequately represented.