Introducing Model-based Problem Solving
STRUCTURING BUSINESS PROBLEMS:
Model-based problem solving is a discipline nurtured by decades of scientific thinking and practice that is essential to the modern solver of business problems.
what they do in generic terms, many exWHEN ASKED TO DESCRIBE ecutives or entrepreneurs will likely say that they ‘solve problems’. Problem solving is what brings ‘busyness’ to business. However, this phrase misleads the mind, because business problems are not the clean, tidy, well-defined puzzles students everywhere grow up with on exams and problem sets. Problems like: • Economists calculating the profit-maximizing price and
quantity conditional on market conditions; • An engineer calculating the transient response of an electri
cal circuit to a sudden surge in input current; or • A company’s operations managers identifying optimal de
Though they look quite different, these problems have some common characteristics.
You know the starting point and you THEY ARE WELL DEFINED. have a conception of what the solution might look like. You do not need a map or other representation to make sense of what the problem is saying.
The method by which you atTHEY ARE WELL STRUCTURED. tempt to solve them will not alter the problem itself, and that method lays out a sequence of steps (iterated dominance reasoning, linear programming, or some other standard tool of the discipline) that will take you from the problem to a solution, which is usually unique.
They are THEY ARE USUALLY PERFORMED BY A SINGLE MIND. meant to test individual prowess — effort times ability — and are therefore solvable by a single person working alone.
Business problems, on the other hand, should not even be called ‘problems’ because they are so different from what we are used to when we hear the word. They should instead be called situations, predicaments or challenges — and they look more like this:
• Design, finance, test build and deploy an electric car provisioning-and-charging infrastructure for the province of Ontario within three years; • Increase the productivity of a 1000-person research and development group in an $8 billion per year gross revenue pharmaceuticals company by 20 per cent over the next six quarters; or • Address what the CEO describes as ‘a sudden breakdown of accountability’ at the level of the top management team of a 70,000-person organization.
These are the types of problems that our executives and entrepreneurs are being called upon to solve. These are the problems of business, and what makes them so different from run-of-themill problems is that they are:
You need to come up with measures for successILL-DEFINED. ful solutions and with the variables that you will focus on. For instance, productivity in an R&D group can be alternately defined as ‘patents granted per dollar spent, ‘number of new products incorporating innovations emerging from the group per headcount’, ‘total gross or net revenue generated by products and services incorporating the group’s innovations’ — and so on. And accountability can be thought of in terms of ‘optimal alignment of the objectives of the individuals on the team with those of the business’ or as ‘the optimization of the reliable flow of accurate and timely information to those that need it most’, and so on.
The very process of getting together a synILL-STRUCTURED. dicate to finance new infrastructure can trigger political and economic actions aimed at derailing the project; or, the very act of making inquiries into the way in which executives pass, distort or withhold sensitive information from one another can cause them to strategically appear more open and trustworthy than they normally are;
None THEY REQUIRE COLLABORATION AND COLLECTIVE ACTION. of these problems can be resolved by a single mind working in a cubicle, office or garage — and as a result, resolving the tensions, clashes and subversions that come to light when humans work together becomes part of every problem you are trying to solve.
To define such problems, you need models or ‘maps’ of the situ- ation that represent what you see and sense onto what you can think with — like variables and the relations between them. To structure these problems, you need to model not only the situation itself, but also the way in which you are going to solve it. And to collaborate productively, you need to be able to communicate, relate and co-reason with others through the objectives, data and models you use to make sense of the situation.
These ‘maps’ can be precise quantitative statements of a set of relationships, or they can be mental maps of the interactions between actors and/or objects. The bottom line is, to arrive at replicable and actionable solutions, we need structural thinking. In short, we need Model-based Problem Solving.
The Elements of Model-based Problem Solving
Model-based Problem Solving is the discipline of defining and structuring a situation or predicament, using the formal and informal language systems that researchers and scientists have developed. It is a distinctive approach to business problem solving that combines the rigor of formal (analytical or mental) modeling with the relevance and actionability of applied problem solving. Like all disciplines, it has several components.
To make sense of a predicament, you need to MODEL BUILDING. map it out. Just as a map is a simplification of a geographical domain, a model is a simplification of a situation: it captures the variables that matter and the relationships between them.
Models from different disciplines can be used to make sense of a predicament and will yield different problem statements. In terms of understanding the role of interpersonal relations, take our ‘accountability crisis’ example. We could use network models to represent the information flows between executives — via formal and informal ties and interactions — and then seek to optimize the speed, reliability and accuracy of those flows to increase the degree to which everyone is informed of every promise made, kept or broken at the top management level.
Or, we could use models of economic agency to represent the authority (i.e. decision rights) and the incentives of each of the executives involved, and then seek to re-allocate decision rights and incentives so as to maximize the degree to which each person has the right to make the decisions she is best equipped to make, and the degree to which her incentives are aligned with those of the organization as a whole.
Alternatively, we could model the process by which the team makes decisions as ‘a set of causal interactions between the brain states of executive team members’, and seek to optimize the
emotional and visceral landscape of meetings by changing their length, format, and physiological cost (by paying attention to the effects of low blood sugar, for instance) for each team member.
Just as you test a map by using it to navigate a MODEL TESTING. known terrain before you venture off into unknown hinterlands, you test models by specifying the relationships you would expect to hold if the model were true and testing them against data you already have.
That is most easily understood in the context of trying to identify the causal relations between variables in one’s map — say, prices and quantities. But we can also explore the role of organizational structures, such as ‘the relationship between the decentralization of decision rights and top management team performance’ by testing the relationship between performance and decision-right centralization in different organizations in the same industry, or in different industries; or, by looking at changes in performance induced by changes in decision-right allocations in top management decision processes in the same firm, or across several firms where such changes can be documented.
Likewise, we could test models of team functioning based on physiological optimization of the milieu in which decisions are made (‘meetings’) by examining the performance of various teams using different decision-making protocols under different physiological conditions; or by looking at changes in the performance of a single team after making changes to the way they interact that impact the neurophysiological conditions of each team member.
Just as maps have free parameters (like scale MODEL CALIBRATION. and texture symbols) that must be interpreted and adapted to the specific landscape you are navigating, models have free parameters that must be tailored to the specific situation you are attempting to act upon. Where structural relations can be quantified and historical data are available, a range of data-driven tools are available for structural identification. But we can do more to understand the interpersonal dimensions of a business problem.
Take our research and development productivity challenge. Focus on a specific developer (‘Tom’) and suppose that we have built a set of models for understanding the variables that make a difference to his productivity, and tested these models to recover a set of relationships we believe to be valid for people in similar positions. We now have ‘loose maps’ ( Figure One) for how Tom’s effort level responds to incentives (i.e. how ‘motivatable’ he is); how his output varies with his effort (i.e. how com- petent he is); how his motivation varies with payoffs to other developers (i.e. how envious he is); how likely he is to tell the truth when he is perversely incentivized to distort or hide it — and so forth.
We now need to calibrate the general form of these relationships to the specific situation at hand — to the R&D developers that we are dealing with here and now. For this, we have to collect data on the specific relationships between incentives and effort levels, effort and output, perverse incentives and probability of distortion that will describe Tom’s behaviour — and that of other key players in the R&D group.
As in the case of quantitative model building and testing, the applied business problem solver can benefit greatly from the experimental art of the social scientist, who must design both unobtrusive measures for the variables she is interested in, and valid tests that will help her ‘tighten’ her map of the domain.
When we work on business problems, we are inMODEL FUSION. terested in predicting and acting as well as in replicability. As a result, we may wish to combine predictions and forecasts from different models — just as we would combine topographical, morphological and geological maps in order to reconstruct—as realistically as possible — the likely pitfalls, waterfalls and landfalls of a mapped terrain.
Every map has a ‘domain of relevance’ and a ‘radius of validity’ — and so does every model. We can have, for instance, different models of ‘the effects of prices on sales volume’ or ‘the efficacy of incentives on managerial integrity’, that have been tested in different data sets in different industries, and therefore have very different functional forms. Should we average across them to come up with an ‘uber-model’? How do we resolve the almost inevitable model clash? That depends on the specific situations in which the models were tested: how similar or different they were; how much of a difference to the dependent measures these differences make; and, in the case of models involving choices humans make, the range of options they had at their disposal in each situation in which the model was tested. Precision in all things—in matters of model specification and testing — is paramount in the delicate game of model fusion.
Alas, fusing models to generate better predictions — even if carried out with care and precision — works most reliably only when the models use the same basic language. For instance, when they are all economic models, or physiological models, or function models of network function and performance.
However, science — as well as metaphysics and ‘lay thinking’— turns up a panoply of models that are often vastly different. Some are relational — like economic models of markets; some are structural, like geological models; some are causal, like physiological or psychological models of behaviour; some are teleological, like economic models of choice; and some — like sociological models of class warfare and ideological conflict — are functional, in that they explain phenomena in terms of functions.
When models are derived from different language systems, fusion often falters; in which case we need something that goes beyond it — something that integrates rather than just aggregates.
A Foundation for Integrative Reasoning
Business problem solving often involves the synthesization of past experience towards a new (and better) solution. Genuine innovations, however — which expand the realm of possibilities — are based on creativity and ingenuity directed towards the design of a new product, process or experience.
famously characterized the ‘designer of solutions’ as someone who ‘devises courses of action to change existing situations into preferred ones’. The act of generating options involves empathetic listening, the playful generation of ideas (e.g. prototyping) and a judicious interweaving of
analysis and intuition. But while the generation of ideas may well be unstructured, their implementation and realization requires a (conscious or unconscious) map.
Furthermore, business decisions do not occur in the vacuum of one’s mind, but in the caldron of the boardroom and the office. It is a process fraught with political and interpersonal conflict and tension, but which relies on collaboration and cooperation in spite of the emotional landscape on which it unfolds. No single mind can behold the right solution at a glance, nor can any single mind even behold all of the promising solutions. As a result, disagreement needs to be harnessed rather than camouflaged in order to uncover all of the sources of value that each contributor brings to the table. This often requires the deployment of special skills and personal ‘technologies’ to turn disagreement into productive and generative tension.
These are powerful instruments, but they rely on a level of specificity about what we disagree on — and a level of agreement on what we do agree on. Such sessions are more likely described by combinations of disagreements on what we want and how the world works that go hand in hand with one another. For example, • One may have a normative, ideological commitment to egalitarian systems, coupled with a base of facts and models that show egalitarian compensation systems exhibit better performance over time; and face • One who has an equally powerful libertarian commitment to individualistic values and radical responsibility, coupled with a fact base that shows how making compensation variable and dependent on individual output, making individ-
Disagreement — as Howard Stevenson has pointed out — can be understood as having two components: disagreements about what we want — i.e. divergences in our goals and objectives; and disagreements about how the world works — i.e. divergences in the facts, data, models and inferences we draw from them. Over time, humans have evolved highly specialized and conflict-specific tools and techniques for resolving ‘pure’ types of disagreement: • Model validation, fusion and selection techniques to help us figure out which models are valid in view of what we perceive to be true, to resolve (within the limits of uncertainty) disagreements about how the world works. • Multi-objective optimization, cooperative and competitive bargaining, and competitive game theory to help us resolve — or at least think our ways to the best outcome of — disagreements about what we want. ual output measurable, and smoothing out kinks in a pay for performance system increases the performance of the organizations in one’s sample.
The normative (what we should do) and the descriptive ( what is) go hand in hand in these cases, and what we see is only the tip of the iceberg — the prescriptive (or, what we should do given what we think is): ‘Reward individual effort without regard to average compensation’; or ‘Make individual compensation responsive to the average compensation of the team’ — which hides all of the normative and descriptive components of each person’s ‘model of the world’ and brings the disagreement to the boil of a real conflict.
This ‘prescriptive dimension’ is the battleground for business problem solvers everywhere. While scientists often use models to refer to diagrams of cause-and-effect among variables — which they take great pains to define in objectively measurable ways — the mind of the business person most often hears ‘models’ to be short for ‘business models’ — which are heavily prescriptive in nature and indicate, ‘Here is what we should do to appropriate value, given what we currently know.’
Because business models are prescriptive and are stated as statements about what one should do, their assumptions must be laid bare before we can use any of our sophisticated instruments for resolving disagreement to bring forth better alternatives. To do this work, we need another family of tools. Hence, the final requirement for solving business problems in collaborative
settings is a set of tools that turn disagreement into generative tension. How? Recognizing that disagreement has separate-butcoupled components (normative and descriptive), it seeks to first ‘unpack’ any prescriptions — including business models — into their normative and descriptive components. • Normative: ‘What do we want to happen or believe should
happen?’; and • Descriptive: ‘What do we think or believe to be true?’ After validating the descriptive components of each prescriptive model (‘what would be true if this belief were valid?’) it proceeds to search for a better model by selectively recombining the descriptive and prescriptive components of each of the prescriptive models that ‘clash’.
If, for example: Prescriptive Model A is, ‘increase accountability by optimizing the physiological milieu of executive meetings’ (and relies on normative assumptions about the physiological determinism of action — how to manipulate brains to get bodies to act); and Prescriptive Model B is, ‘increase accountability by instilling a culture of radical responsibility and vesting decision rights in those who have the most accurate information and the most relevant knowledge’ (and relies on ‘tested’ models of consumer and strategic behaviour, and a normative commitment to individual responsibility), then – The integrative mindset will - uncover and separate out the normative and descriptive
components of each model; - seek to retain only those descriptive components that
have been validated; and then - selectively recombine descriptive and normative components from each model to get to a solution that will be neither optimal nor ideal, but better than the ‘pure’ alternatives.
Such a solution most often emerges as an answer to questions raised by the productive tension that was generated by the clash of the models, such as: • What are the areas and domains in which rationality fails and
falters — and we need to focus on minds rather than brains? • What are the means by which the self-control and self-command of executives can be bolstered in ways that increase their ability to make and deliver on promises? • What are the optimal conditions for meetings and gatherings that make executive team members heed reasons and arguments, as opposed to visceral sensations and hot emotions? Attending to the above will serve to generate models of interaction and executive team development that increase accountability writ large.
Model-based Problem Solving is a discipline nurtured by decades of scientific thinking and practice which is essential to the modern solver of business problems. As indicated herein, it provides a foundation for a comprehensive and integrative approach to defining and structuring business problems: the discipline to base decisions on replicable and (whenever possible) validated structural rationales, the facility to translate innovative and intrinsically non-testable ideas into reality, and a systematic way to resolve interpersonal tensions and opposing, dissenting views.
It is indeed a foundation for the broader craft of analyzing and resolving predicaments of any kind, which are bound to turn up wherever big challenges are tackled. Like all disciplines, Model-based Problem Solving is transferrable through practice and feedback, and not just teaching and ‘imprinting’; and, like most disciplines, it builds skill through repetition and constant adaptation.