The future of a decision

When working with foresight projects it quickly becomes apparent that what we focus on matters. Just focusing on the future in general is sometimes exciting, because it allows us to explore the broad spectrum of possibilities that tomorrow brings - but it also means that we sacrifice specific insights for broad brush perspectives. 

An alternative is to focus on something specific - like the way a budget will change, or how the origin of tourists to a city might change. Such foresight projects are more focused and can often provide more actionable results. One variation on this approach is to look more closely at a decision and a decision maker.

In fact, you could argue that if you understand the decisions that will be made about a company in the future, and if you know how they will be made - with some greater precision than before - then you have a pretty awesome competitive advantage. 

Decision modelling has a long and contested history. The orthodox lineage runs through Ronald Howard's decision analysis school at Stanford in the 1960s, the multi-attribute utility theory of Keeney and Raiffa, and the later "decision quality" movement of Spetzler and Matheson — all of which assume that a decision can be usefully decomposed into weighted attributes and evaluated against expected utility. This is the tradition our frigate example sits within. But the tradition has been challenged from several directions, and the challenges are worth taking seriously. 

Herbert Simon argued that decision makers satisfice rather than optimize, bounded by cognitive limits and the cost of search. James March and his collaborators went further, describing organizational decisions as garbage-can processes where problems, solutions, and decision makers collide somewhat randomly in temporal streams. Gary Klein's work on recognition-primed decision making showed that experts in high-stakes domains — firefighters, military commanders, ICU nurses — rarely weigh attributes at all; they pattern-match against accumulated experience and run mental simulations of the first workable option. Gerd Gigerenzer's fast-and-frugal heuristics tradition has shown, perhaps inconveniently, that simple lexicographic rules ("take the best") often outperform weighted models in real environments. 

And then there is the political-science critique: Allison's three models of the Cuban missile crisis, Baumgartner and Jones on punctuated equilibrium, the whole tradition reminding us that decisions are made by coalitions in venues, and that venue migration often matters more than weight reshuffling. So why model at all? Because — and this is the move worth making explicit — the critiques themselves are modellable. A satisficing decision is a decision model with an aspiration threshold and a stopping rule. A recognition-primed decision is a model with a single dominant pattern-match coefficient and near-zero weights elsewhere. A garbage-can decision is a stochastic model over a coupling matrix of streams. A coalition decision is a weighted model where the weights are themselves negotiated outputs of a prior bargaining model. The point of decision modelling, in foresight at least, is not to claim that real decisions look like tidy MAUT scorecards — they usually don't — but to make our assumptions about how the decision is being made explicit enough to be challenged, reweighted, and forecast forward. The orthodox model is one starting point; the heterodox critiques give us others. What matters is that we pick a model, name it, and then ask the genuinely interesting question: how will the model itself change?

Once we have the decision modelled we can start working out how we think the decision model might change: will some considerations become more important? Will others fade? Are new factors entering the equation? And who will be making the decisions in the future? Let’s look at what it could look like for our boat manufacturer in a simple toy AI-tool using the standard decision model framework.