Your Customer Experience Is Your Org Chart—Rendered in a Browser

In 1968—long before websites, mobile apps, or customer-experience teams—a computer scientist explained why so many digital experiences now feel like organizational diagrams.

Melvin Conway called his paper “How Do Committees Invent?” Its central observation became Conway’s Law:

“Organizations which design systems … are constrained to produce designs which are copies of the communication structures of these organizations.”

Not the strategy.

Not the values statement.

The communication structure: who must coordinate, who is allowed to decide, and where the handoffs break.

Conway was writing about system design. He used the term in a deliberately broad sense.

The banking inference is mine, but it follows directly:

A customer experience is a system too.

When a banking experience is divided into product-owned navigation, separate authentication, disconnected service records, and conflicting outreach, the seams often correspond to internal ownership boundaries.

The customer does not see the org chart.

They experience it.

The model can be right—and the response can still be wrong

In the second article in this series, I described four operating states that appear when Relationship Health and Behavioral Primacy are plotted independently:

• Healthy & Primary
• Healthy, Not Primary
• Primary, Unhealthy
• Weak & Peripheral

Those states make disagreement visible. They do not explain it, and they do not choose the next action.

Consider one customer classified as Primary, Unhealthy: the bank appears central to their financial activity, but the available relationship evidence is weak.

Marketing sees a retention audience and prepares an offer.

Product sees recurring friction and adds a backlog item.

A banker sees a relationship that deserves a phone call.

Service sees the next complaint without knowing any of the other work exists.

Every response can be reasonable inside its own function.

Together, they can be incoherent.

The customer may receive an incentive before the friction is understood, a phone call that cannot resolve it, and another request to explain a problem already reported twice. Meanwhile, the product issue waits for next quarter.

The data did its job. The organization reproduced itself in the intervention.

That is the point where most segmentation programs stop being analytical problems and become operating-model problems.

A customer state is not a campaign trigger.

It is permission to ask a specific question.

The structural fix: a Relationship Action Protocol

Conway’s Law is not a motivational poster about collaborating more.

If fragmented communication produces a fragmented experience, the response has to change the decision structure—not merely add another shared dashboard.

The operating protocol I am proposing has five coordination responsibilities:

1. See the evidence

Before anyone acts, the institution must know what the state is built from.

Every signal needs a contract: the customer unit being classified, its source and lineage, the axis it informs, its freshness, its evidence-sufficiency rule, and its known blind spots.

Relationship Health cannot quietly borrow behavioral activity to make itself look stronger. Behavioral Primacy cannot pretend internal activity equals the customer’s complete financial life.

Unknown evidence stays visible.

2. Ask the earned question

The state selects the first diagnostic question—not the answer.

• Healthy & Primary: What must we protect?
• Healthy, Not Primary: What keeps the relationship secondary?
• Primary, Unhealthy: What is weakening the relationship?
• Weak & Peripheral: Is further attention warranted?

The question is intentionally narrower than “What should we sell?” It prevents the label from outrunning the evidence.

3. Own the inquiry and decision

One accountable owner receives the question. The required collaborators are named. The next review point is explicit.

That owner does not need to personally perform every action. They are responsible for ensuring that Marketing, Product, Sales, Service, Data, Risk, and other necessary functions do not create competing responses around the same relationship state.

Shared visibility is not shared accountability.

4. Act within permission

For each state, the institution defines what the model permits—and what it does not.

A Primary, Unhealthy state may permit service recovery and human review while prohibiting automated growth outreach. Healthy, Not Primary may permit diagnosis of unmet needs without automatically becoming a cross-sell list.

The model should constrain premature action, not merely accelerate it.

5. Learn from what follows

The action, customer response, operational outcome, and unresolved uncertainty return to the evidence record.

That feedback can challenge the classification, expose a missing signal, or reveal that a response worked in one context and failed in another. One successful interaction does not prove causation. It becomes another observation the institution can use to recalibrate the model.

This protocol is not an industry standard or a universal formula. It is an operating design: a way to prevent a useful relationship state from becoming an unexamined next-best-action machine.

The most important output is sometimes “hold”

Most decision systems are built to produce an answer.

A responsible one must also know when it has not earned one.

If the evidence is insufficient, stale, contradictory, or built on disputed lineage, the system should not force the customer into a high-or-low classification. It should show insufficient evidence and not trigger action from the model.

That does not prohibit human judgment. It prevents the model from laundering uncertainty into authority.

The same principle applies when Relationship Health and Behavioral Primacy diverge. Divergence is valuable because it opens a better question. It is dangerous when the institution treats the quadrant as the explanation and launches an intervention before investigating why the signals disagree.

Sometimes the most intelligent action is not an offer, a call, or a journey.

It is a pause with an owner.

What organizational design looks like when the journey wins

There are banking examples of institutions changing the structure around the customer rather than asking the customer to navigate the structure.

USAA has described organizing work around customer “episodes”—the complete set of steps required to meet a member’s need—rather than around products alone, with leaders assigned to manage and improve those episodes.

DBS created a “Managing Through Journeys” program that expanded to more than sixty customer journeys, each led by a senior leader. Its platform model also paired business and technology leadership rather than treating technology as a downstream implementer.

Neither example proves that one operating model belongs in every bank. They demonstrate the design principle: end-to-end customer work needs end-to-end ownership.

Research on group goals adds a warning. A meta-analysis found that specific group goals were associated with better group performance. In the smaller subset examining interdependent teams, individual goals focused on maximizing one member’s performance were associated with worse group outcomes, while goals focused on contributing to the group result were associated with better outcomes.

The practical lesson is not that functional metrics should disappear.

It is that every function can hit its target while the shared customer system fails.

The three articles, in one operating argument

The first article challenged the assumption that expressed sentiment can stand in for an entire relationship.

The second put Relationship Health and Behavioral Primacy side by side, revealing four states that a single score hides.

This third article adds the missing coordination layer: the state opens an owned question, permissions govern the response, and new evidence returns to the system.

The argument can be reduced to three lines:

Feelings are not behavior.

Behavior is not explanation.

Classification is not coordination.

Banks do not need another dashboard that makes fragmented work easier to observe.

They need a decision structure that makes fragmented action harder to perform.

A customer relationship is cross-functional by nature. If the decision system is not, the experience will expose it.

Conway was right: your customer experience is your org chart.

The only question is whether you designed it.