The Economist as Engineer

A soft, human, strategic field already turned itself into an engineering discipline. Go-to-market never noticed.

Consider a problem that sounds like it should have no clean solution. A person needs a kidney. They have a friend or a spouse willing to donate one, which is an act of startling generosity, and then the medical tests come back and the willing donor is incompatible, wrong blood type or wrong tissue match, and the gift cannot be given. This happens constantly. For a long time it was simply a tragedy with no structure to it, two people stuck, a healthy kidney that could save a life and no way to move it to the person who needed it. The donor wanted to give, the patient needed to receive, and the match was wrong, and that was the end of the story.

Except that somewhere there is another pair in exactly the same position, a second patient with a second willing, incompatible donor, and it sometimes happens that the first donor is compatible with the second patient and the second donor is compatible with the first. If you could find those two pairs and bring them together, you could swap the donors, and two people who were going to wait and possibly die would instead receive kidneys, from strangers, in a single coordinated exchange. The kidneys do not have to be bought or sold, which is illegal and would be monstrous anyway. They only have to be matched, and matching is a problem with a structure, and a structure is a thing you can design.

That is roughly what a group of economists did. They took the tragedy of incompatible donor pairs and treated it as a design problem, building systems that find these chains of compatibility across large pools of patients and donors, sometimes linking many pairs at once into long sequences where each donor gives to the next patient down the line. Some of the longest chains begin with a single altruistic donor, a person who simply wants to give a kidney to a stranger, whose gift starts a cascade in which each recipient’s incompatible donor passes a kidney forward to the next patient down the line, so that one act of generosity, routed through a well-designed system, can free a whole sequence of kidneys that were otherwise stuck. The chain is not a fact of nature. It is an artifact, the product of a system someone built to find and order those matches, and without the design the same generous donor and the same willing pairs would have produced almost nothing, because the compatibility was there the whole time and only the mechanism to use it was missing.

The systems are real, they run, and they have saved a very large number of lives that the older, structureless arrangement would have lost. The people who built them were not doing medicine and they were not doing pure theory. They were doing something closer to engineering, and the fact that the material they were engineering was a market made of human beings rather than a bridge made of steel did not make it any less an act of design.

I start here because it is the cleanest example I know of a soft, human, strategic domain becoming an engineering discipline, and because go-to-market, which is also a soft, human, strategic domain, has somehow never noticed that this is possible, let alone that the tools to do it already exist.

A field that engineered itself

The kidney exchange is one instance of a broader thing that happened in economics over the last several decades, quietly enough that most people outside the field never registered it. Economics had always thought of itself as a science, in the business of describing how markets and people behave, discovering the laws of supply and demand the way a physicist discovers the laws of motion. Then a part of the field turned toward a different question. Instead of only asking how existing markets behave, it began asking how to build markets that work, how to design the actual rules by which buyers and sellers, or patients and donors, or students and schools, are brought together, so that the outcome is good rather than merely whatever happens.

This is the field that came to be called market design, and it has a real record. When governments wanted to allocate the radio spectrum that mobile phones depend on, economists designed the auctions that did it, carefully, because a naive auction produces terrible outcomes and a well-designed one produces good ones, and the difference is in the rules. When cities wanted to assign children to public schools without the process collapsing into chaos and gaming, economists redesigned the matching systems that do it, in Boston and New York and elsewhere. When newly graduated doctors needed to be matched to the hospitals where they would train, a process that had broken down repeatedly over the decades, economists rebuilt the matching mechanism that now runs it every year. Each of these is a market, in the broad sense, and each of them works as well as it does because someone designed the mechanism underneath it rather than leaving it to form on its own.

Each of these stories has the texture of engineering rather than discovery, full of specific failures that had to be diagnosed and designed around. The medical match had a long history of breaking down, with hospitals making offers earlier and earlier each year to beat their rivals, until students were being asked to commit to training positions long before they had finished the stages that were supposed to come first, a slow-motion unraveling that no amount of exhortation could stop and that only a redesigned mechanism could halt. The school systems had close to the opposite problem, mechanisms that quietly punished honesty, where a parent who ranked schools by true preference could end up worse off than one who gamed the form, so that the system was effectively teaching families to misrepresent themselves and then delivering worse outcomes to the ones who refused. In each case the remedy was not a better attitude or a clearer set of values. It was a different mechanism, designed by people who had studied exactly how the old one failed and built the new one to remove the failure.

The economist most associated with naming what was happening is Alvin Roth, who called it, in so many words, the economist as engineer. His argument, which he made carefully and which earned a Nobel Prize for the body of work it described, was that this design activity is a different kind of intellectual work from the science economics had mostly done before. The scientist economist seeks to understand what is. The engineer economist seeks to build what should be, a working institution that produces good outcomes in the messy particular conditions of the real world, and that second activity, Roth argued, requires its own kind of knowledge, not reducible to the elegant theory it draws on.

Why it counts as engineering

It is worth being precise about why this deserves the word engineering, because the parallel to everything go-to-market needs is exact.

A pure theory of markets can tell you, under clean assumptions, what an idealized mechanism would do. It can prove that a certain kind of auction has a certain property, that a certain matching procedure cannot be gamed, that an equilibrium exists. This is the theoretical-tools layer, and it is real and necessary, and it is also nowhere near enough to build a working market, for the same reason that knowing the equations of fluid dynamics is nowhere near enough to build an aircraft. The theory holds under assumptions that the world violates. Real bidders collude and make mistakes and run out of patience. Real hospitals have couples who want to be placed in the same city. Real donors drop out at the last moment, which can break a long exchange chain, so the systems have to be designed to fail gracefully when that happens. None of these complications appears in the clean theory, and all of them have to be solved before the thing works, and solving them is design under constraint, which is the heart of engineering.

The history is full of cautionary tales that make the point. When New Zealand auctioned radio spectrum in 1990 using a design that looked elegant on paper, the results were embarrassing enough to become a standing lesson, because a bidder who valued a license very highly could end up paying a trivial amount for it once the pricing rule met the thin field of actual competition, and the gap between the winning value and the price collected was politically indefensible. The theory underneath the design had been satisfied. The design itself had not been engineered against the conditions it would actually face. The two are not the same, and the distance between them is measured in outcomes that range from excellent to absurd depending entirely on details the clean theory treats as beneath its notice, which is precisely the territory where engineering lives.

Roth made this point directly, and it could have been written by an aircraft engineer. The theory, he argued, gives you a frame, and then the actual work of design is in the details the theory abstracts away, the institutional particulars, the computational realities, the behaviors of real participants, all of which have to be studied empirically and accommodated in the design. The engineer economist learns by building working markets and watching them run, by experiment and computation, by the careful study of where a mechanism failed and why, which is the same way the aircraft designers learned what made an airplane flyable. The knowledge that results is engineering knowledge, made by people building things, organized around making the thing work in the world rather than around understanding the world in the abstract.

This is worth dwelling on, because it is the precise move go-to-market has never made. The market designers did not abandon theory, and they did not worship it either. They held it in its proper place, as the frame that tells you roughly where to look, and then they did the engineering, which is everything the theory leaves out. They ran experiments, sometimes in the laboratory with paid subjects and sometimes in the field with real institutions, to see how people actually behaved rather than how the model assumed they would. They used computation to handle the combinatorial reality of matching thousands of participants at once, which no closed-form result could touch. And they treated each deployed mechanism as an instrument that generated knowledge, watching it run, finding the cases where it strained, and feeding what they learned back into the next design. That loop, from theory to design to deployment to study and back again, is the engineering method, and it is how a soft domain accumulates the hard knowledge that turns it into a discipline.

What strikes me about this is not only that it happened, but that it happened in a domain everyone would have called soft. Markets are made of people, with all their irrationality and private motives and capacity to game any system you put in front of them. If any field had an excuse to say its material was too human and too messy to engineer, it was this one. It engineered anyway, and the engineering works, and lives are saved and spectrum is allocated and doctors are matched because someone refused to accept that a human, strategic domain was beyond the reach of deliberate design.

The same problem, one layer out

Here is the connection that I think the go-to-market world has missed entirely. Market design builds the marketplace, the venue and the rules by which buyers and sellers find each other and transact. Go-to-market builds something adjacent and arguably harder, the apparatus by which a market for a particular product comes into being in the first place, by which attention is gathered, trust is established, and demand is created and then captured. These two are different tasks of the same kind, the deliberate design of a system that allocates something scarce among strategic human participants, and the second one has every bit as much structure as the first.

Look at go-to-market through the lens market design uses, and it stops being a fog of campaigns and starts looking like a mechanism. There are participants, the potential customers, each with private information the designer cannot see directly, their real needs and their actual willingness to pay, which they reveal only partially and sometimes strategically. There is something scarce being allocated, attention and trust on the buyer’s side, time and money on the seller’s. There are incentives that determine how participants behave, and there is a set of rules, mostly implicit and undesigned today, that determines who ends up matched with the product and on what terms. Market design has a precise vocabulary for all of this. A mechanism is the set of rules that maps what participants do into an outcome. A mechanism is well designed when it produces good outcomes even though participants act in their own interest and hold information you cannot see. The central results of the field are about exactly this, how to design the rules so that the outcome is good despite the participants being strategic and informed in ways you are not.

A concrete case makes the parallel less abstract. Consider the ordinary decision of how to structure a free tier, which nearly every software company now faces. From a distance it looks like a marketing tactic, a question of how generous to be and what it does to conversion. Seen as a mechanism, it is an allocation rule that sorts strategic participants by their private information. A free offer is taken up by different people for different reasons, some who will never pay and are drawn precisely because it costs nothing, some who will pay once they have felt the value, and some who will take exactly what they need from the free tier and shape their usage to never cross the line into paying. The boundary you draw, what sits inside the free tier and what sits outside it, is a mechanism that elicits and separates these types, and where you draw it determines who self-selects into which group. A market designer would recognize this on sight as a screening problem, would reason explicitly about how each type responds to the boundary, and would set the line to separate the types in a way that serves the business. Go-to-market usually draws the line by intuition and benchmark, sets the mechanism running, and then studies the conversion rate afterward, which is measuring the output of a mechanism it never consciously designed.

Go-to-market has all of these elements and almost none of the discipline. It has participants with private information, and instead of a designed mechanism for eliciting it, it has a folk practice of guessing. It has something scarce being allocated, and instead of an allocation rule, it has a pile of tactics. It has strategic behavior everywhere, customers who learn to ignore the outbound, who game the free tier, who say one thing in the survey and do another, and instead of designing for that strategic behavior the way an auction designer designs against collusion, it mostly acts surprised by it. The raw material of a designed mechanism is all present. The design is missing. And the field that could teach go-to-market how to do this has been sitting one discipline over the whole time, having already proven that a soft strategic domain can be engineered, having already built the theoretical tools and the design methods and the hard-won practical knowledge of how real strategic participants actually behave.

What the borrowing would look like

I am not suggesting that go-to-market can lift the mathematics of auction theory off the shelf and apply it unchanged, because that would be the applied-science mistake in a new costume, the belief that you can take a clean theory and drop it onto a messy problem and have it work. The lesson of market design is the opposite of that. The theory is the starting frame, and the real work is the engineering, the patient accommodation of everything the theory leaves out.

What go-to-market can borrow is the posture and the method. The posture is that a market for a product is a thing you design, not a thing that merely happens to you, and that the quality of the design determines the outcome more than the quality of any individual campaign. The method is the one market design actually uses, which is to model the participants and their incentives explicitly, to reason about how strategic people will respond to the rules you set, to build the mechanism and then watch it run and study where it fails, to use experiment and computation rather than intuition and anecdote, and to treat each working system as a source of accumulating knowledge rather than a one-off. This is engineering knowledge in Roth’s sense, made by building things and learning from them, and it is exactly the kind of knowledge go-to-market has never set out to build on purpose.

There is even a warning embedded in the parallel, which is that mechanisms produce strategic responses, and a mechanism designed carelessly produces bad ones. A school-choice system designed without care teaches parents to game it, and a spectrum auction designed without care teaches bidders to collude, and the designers of those systems learned to anticipate the gaming and design against it. Go-to-market sets mechanisms in motion constantly, every pricing scheme and referral program and free tier is a mechanism that creates incentives, and it tends to set them without thinking about the strategic response at all, and then it is surprised when customers behave strategically, as people always do when you put a mechanism in front of them. A discipline that took the market-design lesson seriously would design for the strategic response from the start, because it would understand that it is always building a mechanism whether it means to or not.

Designing for the response is a different craft from running the tactic. A referral program designed for the strategic response anticipates that some people will refer low-quality sign-ups to collect the reward, and shapes the reward rule so that the incentive points at the behavior the business actually wants rather than the one that is easiest to fake. A usage-based price designed for the strategic response anticipates that customers will reshape their usage to sit just under the next threshold, and either makes peace with that or moves the threshold to where the reshaping does no harm. The discipline is not in having the program or the price. It is in reasoning, before launch, about how a strategic person will move once the rule is in place, and then choosing the rule whose resulting behavior you can live with. That reasoning is routine in market design and rare in go-to-market, and the distance between them is the distance between a mechanism that is engineered and one that is merely set loose.

The opening

The thing I find almost strange is how available all of this is. The intellectual work of showing that a soft, human, strategic domain can be turned into an engineering discipline has been done, decisively, by a field next door, and rewarded with the highest honors that field can give. The tools exist. The methods exist. The proof of possibility exists, sitting in every kidney exchange chain and every spectrum auction and every year’s medical match. And go-to-market, which has the same raw materials and an arguably larger prize, has gone on treating itself as a craft that depends on talent and luck, reinventing folk versions of ideas that the discipline next door formalized decades ago.

I do not fully understand why the borrowing has not happened, except that disciplines are oddly bad at noticing one another, and that the people who do go-to-market and the people who do market design move in worlds that almost never touch. The marketers do not read the economics, and the economists, with a few exceptions, have aimed their design tools at public institutions and large platforms rather than at the ordinary problem of bringing a product to a market. The result is a strange waste, a solved meta-problem on one side and an unsolved practical problem on the other, separated by little more than the fact that nobody whose job it is to bring products to market was looking over the fence.

None of this hands go-to-market its discipline ready-made, because the borrowing still has to be done, and doing it is real work, the work of figuring out which tools transfer and how, and of building the practical knowledge of how strategic participants behave in this particular domain rather than in an auction. But it changes what kind of problem this is. It is no longer a question of whether a field like this can be engineered, because that question has been answered, by people who took a domain at least as soft and at least as human as go-to-market and built working machinery on top of it. The question is only whether anyone will do the same here, starting where the market designers started, with a clear account of the participants, the scarce thing being allocated, and the mechanism that moves one to the other. Which returns, as it keeps returning, to the object itself, the thing that go-to-market builds and has never quite been able to name, because you cannot design a mechanism until you can say what it is supposed to produce.

---

References

• Alvin E. Roth, “The Economist as Engineer: Game Theory, Experimentation, and Computation as Tools for Design Economics,” Econometrica 70, no. 4 (2002).
• On market design in practice: the redesign of the National Resident Matching Program, school-choice mechanisms in Boston and New York, FCC spectrum auctions, and kidney exchange; see the body of work recognized by the 2012 Nobel Memorial Prize in Economic Sciences (Alvin Roth and Lloyd Shapley).
• David Gale and Lloyd Shapley, “College Admissions and the Stability of Marriage,” American Mathematical Monthly 69 (1962), on the deferred-acceptance algorithm underlying many matching markets.
• Foundational mechanism-design theory: Leonid Hurwicz, Eric Maskin, and Roger Myerson (2007 Nobel Memorial Prize).