In 2007, researchers in Tangier, Morocco were studying take-up of piped water connections among households without access. The Moroccan government had made subsidized loans available to make the connection affordable. The economic case was crystal clear. Households spent significant time and money getting water from public taps and private vendors. A piped connection would save them both. The loan made it affordable. By any standard economic analysis, the program should have produced near-universal take-up. It did not. Fewer than ten percent of eligible households took the loan. The model was correct on the merits. The economic analysis was sound. Yet the take-up rate was a tenth of what the model predicted. This was not a flaw in the causal reasoning. It was a flaw in the implementation. The model said piped water would come. The organization could not deliver it, because no one had accounted for what actually happens when a household tries to complete the application.

The Tangier study is not an isolated case. It exemplifies a consistent pattern across policy interventions. Most do not fail because the underlying model is wrong. They fail because the implementation is poor. The model says 'do X.' The organization does something that looks like X from the outside, but is structurally different on the inside. The result is not what the model predicted. The causal reasoning could be impeccable. The economic logic could be sound. But if the implementation does not actually execute what the model describes, the intervention will not produce predicted effects. What fills the distance between correct causal reasoning and successful organizational change is the work of translating intention into practice. This requires something the modeling apparatus has not prepared practitioners for: understanding of how real organizations work, where friction points are, and how to solve them when they do not show up in any equation.

Households in Tangier without piped water faced a real cost. They spent significant time and money obtaining water from public taps and private vendors. This was not a minor inconvenience. It was a daily burden consuming household resources and time. The Moroccan government, recognizing this, made subsidized loans available. The loans made connections affordable. The economics were straightforward. A household could obtain piped connection, eliminate recurring costs of purchasing water, and save both time and money in perpetuity. The causal logic was sound. Lower the price of something valuable through subsidy. Enable the person to make the rational economic choice. Predict near-universal adoption. This is basic applied economics, tested across thousands of interventions. It should have worked. The question became: why did it not?

In the early phase of the program, fewer than ten percent of eligible households took the loan. The model said this should not happen. The economic incentives were clear. The households faced a genuine daily burden. The loan removed the financial barrier. There was no rational reason for near-universal take-up not to occur. Yet it did not. This was not a marginal miss. It was not a model ninety percent right. It was a program that produced a tenth of the expected result. The question the researchers had to face was difficult: either the model was fundamentally wrong about human behavior, or something was happening between the loan being available and the household actually taking it. Something in the implementation was clogging the path from eligibility to enrollment. To find out what, the researchers did something economists often do not do. They went into the field and watched what households actually did.

The researchers went into the field and watched what households actually did when faced with the loan offer. The answer had nothing to do with the loan's terms or the economic value of the connection. It had to do with the application process. To get the loan, a household had to obtain a copy of an identity document, then photocopy it, then fill out a form, then travel to the water utility office during specific hours. Each step was small. No single step was impossible. But the cumulative effect was a barrier. Households had the intention to apply. They had the incentive. They understood the economic case. But when they ran into the actual process, they encountered one step that required travel, time, or resources they did not have readily available. They started the application, got stuck, and did not come back. This is not a failure of the economics. It is a failure of the installation. The model was correct. The implementation was clogged.

The researchers ran an experiment. They divided eligible households into two groups. The control group had access to the loan under the normal application process. The treatment group had the same loan offer, but with one addition: the research team would help with procedural steps. When a household in the treatment group expressed interest, the researchers brought the form, helped fill it out, photocopied the documents if the household did not have copies, and submitted the application on the household's behalf. The intervention added nothing to the loan's economic terms. It did not increase the subsidy. It did not improve the value of the piped connection. It only removed the administrative friction blocking households from actually taking the loan. This is a crucial intervention, because it tests whether the barrier was truly administrative or whether something deeper about household preferences was driving low take-up.

Take-up in the treatment group was sixty-nine percent. The same households, facing the same subsidized loan, with the exact same economic logic, increased take-up by a factor of seven simply because someone else handled the paperwork. This is not a marginal improvement. The model's prediction was completely insensitive to what turned out to be the dominant barrier to adoption. The model said the loan would produce piped water connections. The model was correct on the causal structure. The loan does cause connections in households that take it. But the path between loan availability and loan acceptance was clogged with administrative steps that no economic model would have predicted to matter at the scale they did. The model had established the causal link. The implementation had broken that link at the administrative layer. The plumbing was the issue.

In her 2017 presidential address to the American Economic Association, Esther Duflo offered a metaphor that propagated through development economics. The economist has historically aspired to be a scientist, identifying universal laws governing human and economic behavior. Some economists, particularly those working on mechanism design, have functioned as engineers, taking the scientist's laws and building specific systems applying those laws to real problems—auction designs, school-choice procedures, kidney-exchange algorithms. But Duflo's argument was that neither role captures what is actually needed when an intervention has to be made to work in reality. The scientist establishes causal knowledge. The engineer designs an intervention informed by that knowledge. But the intervention then has to be installed in a real organization, with its real people, its real procedures, its real bureaucracies, and its real history. That installation is the work of a plumber.

The plumber takes the engineer's design and figures out how to fit it into the existing system of pipes—what couplings to use, where pipes are likely to leak, what to do about awkward bends no one anticipated. Plumbers do not have the prestige of scientists or the elegance of engineers. They get their hands dirty. They tinker. They make decisions without complete knowledge of the system they are working in, because no one has complete knowledge. They understand that the design looks good on paper but will encounter friction in reality. They monitor for leaks after installation, because installations always leak somewhere. The leaks are not design failures. They are signals telling the plumber where the system differs from the blueprint. They tell the plumber where the next intervention has to go. This is the work a policy intervention requires when installed in a real organization. The Tangier researchers were plumbers. The economists who designed the loan were scientists and engineers. But someone had to watch what happened, find the barrier, remove it, and monitor whether the new installation worked.

The plumber metaphor has resonated because it names something practitioners have felt for decades and lacked vocabulary for. Most policy interventions, most strategic initiatives, most organizational changes do not fail because the underlying analysis was wrong. They fail because the implementation was poor. The model said 'do X.' The organization did something that looked like X from the outside and was structurally different on the inside. The result was not what the model predicted. The model might have been impeccable. The economist might have identified the causal link correctly. But when the organization tried to actually execute the model, something broke. In Tangier, the causal link was there. The loan does cause connections in households that take it. But no household took it because the path to taking it was blocked by administrative friction no model had captured. The plumber's work is making the implementation actually be what the model says it is, in the messy organization where it has to live. This is not extra work happening after science is done. It is the essential work determining whether science produces any results at all.

The objection the plumber raises against causal modeling is not that the modeling is wrong. It is that modeling, taken on its own, is incomplete. A correct causal estimate tells the decision-maker what the causal link is. It tells them that if X happens, Y will follow. But it does not tell them how to make X actually happen in an organization that has never made X happen before, in the way required to produce Y. It does not tell them where friction points will be. It does not tell them what steps will get stuck, which ones will create barriers, which parts of the real bureaucracy will resist. This is why a Living Model has to be more than an estimate. It has to be a framework for watching implementation, finding barriers, removing them, and iterating on the real constraints standing between correct analysis and effective organizational change. The plumber's objection is not a challenge to science. It is a statement that science is insufficient.