HBR Mapped the Loyalty Blind Spot in 1995. Most Banks Still Can’t See It.

In my last article, I argued that customer sentiment and customer behavior are two different forms of evidence.

What customers say matters. What they do with their financial lives matters too. The mistake is treating one as proof of the other.

That distinction is not new.

In fact, Harvard Business Review mapped the operational problem thirty years ago.

Banking is still trying to manage it with one axis.

The model HBR published in 1995

In 1995, Thomas Jones and Earl Sasser published “Why Satisfied Customers Defect.” Their argument was more precise than the headline is sometimes remembered: merely satisfying customers was not enough to produce loyalty, and apparent loyalty was not always evidence of genuine preference.

When they looked at satisfaction and loyalty together, four customer behaviors appeared:

• Loyalists were highly satisfied and highly loyal. The most vocal became “apostles.”
• Mercenaries could be highly satisfied and still show little loyalty.
• Hostages remained despite being dissatisfied, often because alternatives or switching were difficult.
• Defectors were low on both dimensions. The most vocal critics became “terrorists.”

The language belongs to another era. The management problem does not.

Read the model as a banker.

A mercenary can rate the mobile app nine out of ten while keeping savings, investments, and most daily activity somewhere else.

A hostage can have a fifteen-year relationship, a mortgage, direct deposit, and recurring bills—and still be one frictionless switching offer away from leaving.

The same observed “loyalty” can represent preference, inertia, constraint, convenience, or habit. The same satisfaction score can sit on top of radically different economics.

HBR identified the blind spot in 1995. What most institutions lacked was a practical way to connect customer-reported experience with individual-level financial behavior at scale.

That limitation is far smaller now.

The evidence already exists across survey, complaint, service, transaction, payment, product, and digital systems. The modern problem is not the absence of data. It is the absence of shared meaning—and the discipline to keep unlike signals separate.

The operational adaptation: two signals, not one super-score

The framework I have been developing through my work in banking uses two independent signals:

Relationship Health describes the experience the customer appears to be having with the institution. Its evidence can include trust, satisfaction or advocacy, effort, complaints, friction, and resolution quality.

Behavioral Primacy describes how central the institution appears to be in the financial activity it can observe. Its evidence can include recurring inbound flows, everyday transactions, bill payments, active product use, and the persistence of those behaviors over time.

PFI is the familiar banking shorthand. But observed behavioral primacy is the more honest construct.

A bank can see activity inside its own walls. It cannot automatically see every external balance, competing relationship, or reason money moved. Internal data is evidence—not omniscience.

The two signals should not be blended into one score. And the same behavior should not quietly carry weight on both axes. Digital logins, for example, may provide context, but they should not be allowed to make a relationship look emotionally healthy and behaviorally primary at the same time without a clear evidentiary reason.

Every signal needs a contract: what it means, which axis it informs, how fresh it is, and where it is incomplete.

Thresholds also need local validation. A consumer checking relationship, a mortgage-only household, and a small business do not express primacy in the same way. The framework should classify the unit the bank can actually manage—and show insufficient evidence when the data does not justify a high-or-low label.

Once those rules are explicit, plotting Relationship Health against Behavioral Primacy produces four useful relationship states.

Four states—and four different first questions

These are not permanent customer identities. They are operating states based on the best available evidence. Customers can move between them, and the first management question changes with the state.

Healthy & Primary

This is the modern echo of the Loyalist: strong relationship evidence and strong observed primacy.

The temptation is to celebrate and cross-sell. The better first question is:

What must we protect?

These relationships are valuable precisely because both signals agree. Unnecessary pressure, unresolved friction, or a poorly timed offer can turn strength into drift. The objective is not to extract more from every customer. It is to preserve the conditions that made the relationship central.

Healthy, Not Primary

This resembles the Mercenary, but the label should not be taken literally. The customer may genuinely like the bank. The relationship simply is not central in the behavior the institution can see.

A satisfaction dashboard records success. A primacy lens exposes the unanswered question:

What is preventing consolidation?

The barrier may be product fit, pricing, missing capabilities, habit, an external employer relationship, or the simple fact that nobody has made a relevant case for moving more of the financial life here.

This segment is not an automatic sales list. It is a diagnosis queue.

Primary, Unhealthy

This is the modern echo of the Hostage: the behavior is central, but the relationship evidence is weak.

The dangerous interpretation is “loyal customer.” The better question is:

What friction must be resolved before we ask for anything else?

Structural loyalty can look durable until a competitor removes the switching cost. A simpler onboarding journey, a better rate, or a clean refinance process can expose how little emotional commitment was underneath years of activity.

Growth outreach here can be tone-deaf. Repair comes first.

Weak & Peripheral

This resembles the Defector state: weak relationship evidence and weak observed primacy.

It is tempting to call every customer in this quadrant “at risk.” That overstates what the model knows. Some may be drifting. Others may be dormant, narrowly served, misclassified, or simply low-opportunity for the institution.

The first question is:

Is further attention warranted—and what evidence would justify it?

Not every weak relationship should trigger an expensive retention intervention. The framework should improve judgment, not automate panic.

The wallet is already telling us the old proxies are failing

Banking relationships rarely end with a dramatic account closure. They fragment.

J.D. Power reported in 2026 that the average retail checking customer holds three deposit accounts across different institutions. It also found that 20% of customers had moved money away from their primary bank during the previous three months.

Cornerstone Advisors found that only 47% of U.S. consumers—and 35% of Gen Z—cited direct deposit as a reason they considered an account primary.

The direct deposit can remain. The account can remain open. The survey score can remain green.

The relationship can still be becoming less central.

That is why “PFI” cannot be a permanent flag attached at onboarding. Primacy is a behavioral condition that can strengthen, weaken, and migrate without announcing itself.

What this framework does—and does not—solve

The value of the two-axis model is not that it produces a smarter label.

It makes disagreement visible.

When relationship health and behavioral primacy agree, the management posture is clearer. When they diverge, the bank has learned something a single score could not reveal—and earned the right to ask a better question.

But classification is not causation.

The framework does not know why a customer sits in a quadrant. It does not prove that an offer will change behavior. It does not calculate full share of wallet from inside-the-bank data. And it should not automatically route a customer into a campaign before the institution has validated the signal, the opportunity, and the appropriate treatment.

That is the limitation most segmentation decks skip.

Knowing what appears to be true does not make Marketing, Product, Sales, Service, and frontline teams agree on what happens next. A model can reveal a Healthy, Not Primary relationship. It cannot, by itself, decide whether the next move is a product fix, a service intervention, a banker conversation, or no action at all.

I have seen that coordination gap undo good segmentation work more often than bad math ever does.

That is the subject of the third and final article in this series: the decision and coordination layer that must sit on top of the model—or the four states become one more sophisticated dashboard nobody acts on.

A better model does not tell a bank what to do.

It tells the bank which question it has earned the right to ask.