Context: How to Reduce the Risk of Missing What Matters to a Decision
Context can never be captured completely, so decision governance should guide the search for it rather than try to inventory it. This text identifies five places where decision makers miss context, plus time, and proposes a counterfactual test for which context is decisive.
Which contextual information matters to a decision, and how can a decision maker, or a group deciding together, reduce the risk of missing it? This text proposes a way to conceptualize context for that purpose. It draws on research on context in philosophy of language, artificial intelligence, human-computer interaction, linguistic anthropology, psychology, and organization theory.
**Context and information**
Context is not the same as information. Information is what decision makers consider when they deliberate: data, reports, forecasts, opinions. Context is what determines which information is treated as relevant, and how it is interpreted. A missing fact is an information problem. A context problem is when a fact was available to someone involved, but was not seen as relevant to the decision, or was read under assumptions that did not hold.
The research on context is spread across fields, and its definitions differ. Some treat context as a given set of features of a situation, such as who is speaking, when, and where (Kaplan, 1989), or any information that characterizes the situation of an entity (Dey, 2001). Others treat it as something participants construct as they interact (Sperber & Wilson, 1986; Duranti & Goodwin, 1992; Dourish, 2004).
One point is common to nearly all of them: context is always the context of something. Brézillon and Pomerol (1999) define it relative to a focus, as what constrains that focus without being part of it. For decision governance, the focus is the decision. What matters for one decision is background for another. Relevance therefore has to be judged against the specific decision, not against the situation in general.
**Context cannot be captured completely**
McCarthy (1993), proposing a formal treatment of context for artificial intelligence, observed that a context can never be fully specified. It is always possible to move to a broader context in which further considerations apply. Dourish (2004) reached a related conclusion from a different direction. He argued that context is not a stable set of attributes that surrounds an activity and can be recorded separately from it. It is a relation that arises from the activity and is relevant to it.
If this is right, decision governance that aims to capture all relevant context will fail, and the attempt will be costly. Checklists grow, required fields multiply, and the effort spent filling them in is attention not spent on the decision.
The alternative is to govern the search for context rather than its inventory. That means two things: defining where the decision makers look, and applying a test of which context is decisive. The next section proposes where to look. The one after it proposes the test.
**Five places where context gets missed**
Read together, the research traditions describe different ways in which decision makers miss context. Each points to a specific failure and to something a designer of decision governance can do about it.
**The situation**
The first place is the situation itself: constraints, stakeholders, timing, resources, regulation. This is what most decision processes already check, and it is the easiest to turn into a checklist or into required fields in a decision record.
The failure is that a checklist reflects past decisions. It omits whatever no earlier decision needed. Simon (1990) described rational behaviour as shaped by two blades of a pair of scissors: the structure of the task environment and the computational capabilities of the actor. The question that follows is not which features of the situation are present, but which features the outcome of this decision depends on.
The governance answer is to keep the checklist, and add a required step asking what it does not cover for this decision.
**The causal nature of the situation**
The Cynefin framework distinguishes situations by how cause and effect relate in them: clear, complicated, complex, or chaotic (Kurtz & Snowden, 2003; Snowden & Boone, 2007). In a complicated situation, the relevant context can be found through analysis and expertise. In a complex one, it shows up only in response to action, through small experiments.
The failure is treating a complex situation as complicated. Decision makers then commission analysis when the context they need does not yet exist to be analysed.
The governance answer is to require decision makers to state which kind of situation they believe they are in, and what observation would show that belief to be wrong.
**The shared ground**
Stalnaker (1978) described the context of a conversation as its common ground, the propositions that participants take for granted. Lewis (1979) described how that ground changes. When someone speaks as if something is settled, others tend to adjust to it, and the claim enters the common ground without being examined. He called this accommodation.
Groups assume they share more presuppositions than they do. The failure is an assumption that mattered but was never stated, and therefore never challenged.
The governance answer is a step in which the group writes down what it is assuming, and reads it back before the decision is made. Disagreement about assumptions is easier to see on paper than in discussion.
**Perspectives**
Each participant holds a partial, local model of the situation (Ghidini & Giunchiglia, 2001). Each also selects, from what they know, what seems relevant to them (Sperber & Wilson, 1986). Research on memory adds that what people recall depends on the setting they are in when they try to recall it (Tulving & Thomson, 1973; Godden & Baddeley, 1975). A meeting room brings back different facts than the place where the decision will be carried out.
The failure is that the person who holds the decisive context is not involved, or is involved but does not see it as relevant to the discussion as it unfolds.
The governance answer has two parts. Collect each participant's view independently before group discussion, so that early speakers do not set what others consider relevant. And involve people from where the decision will play out, or hold part of the deliberation there.
**The frame**
Goffman (1974) described frames as the schemes people use to answer the question of what is going on. Tversky and Kahneman (1981) showed that how options are described changes which one people choose, even when the options are the same. The formulation of a decision does the same to context. “Should we renew the vendor contract?” brings up different considerations than “How should we meet this need over the next five years?”
The failure is a frame that excludes context before any search for it begins.
The governance answer is to require at least two alternative formulations of the decision, and to check whether they bring up considerations the first one did not.
**Time**
Time cuts across the other five. Weick (1995) argued that people partly create the environment they then make sense of, and that much sensemaking is retrospective. A decision changes its own context, and part of what mattered becomes visible only afterwards.
The governance answer is to move some of that hindsight forward. A pre-mortem asks participants to imagine that the decision has failed and to explain why (Klein, 2007). The decision record should also state the conditions under which the decision is to be revisited.
**Which context is decisive**
Since context cannot be captured completely, decision makers need a way to tell which context deserves checking. The most useful single test is counterfactual: what, if it were different, would change the choice? Context that passes this test needs to be verified. Context that does not is background.
Two questions apply the test from different directions. The first is: what would have to be true for the option being rejected to be the right one? It directs the search toward context that would overturn the choice, rather than context that confirms it. The second is: who would disagree with this decision, and what do they know that the decision makers do not? It is a short route to the perspectives that the group does not already hold.
Both questions can be standard prompts in a decision record. The answers then document why some context was treated as decisive and the rest as background.
**What this costs, and who it serves**
Each of these practices costs time and attention. Collecting views independently delays the first discussion. Stating assumptions and alternative formulations adds steps. A pre-mortem can produce implausible scenarios that take attention away from likely ones. Repeated often enough, an assumption list becomes a ritual that is filled in and not read.
The cost is not worth paying for every decision. It is hard to justify for decisions that are easy to reverse, low in stakes, or made in a situation where cause and effect are clear and a checklist is enough. It is easier to justify when a decision is hard to reverse, when its outcome depends on factors outside any existing checklist, or when participants come from different units and so hold different local models of the situation. Governance of the search for context can therefore be proportional: a checklist for every decision, and the further steps only for decisions that meet stated criteria.
These practices also redistribute influence, and not everyone involved benefits from that. Whoever formulates the decision chooses the frame, and requiring alternative formulations reduces that advantage. Making assumptions explicit can expose disagreements that some participants preferred to leave implicit. Collecting views independently gives weight to people who would otherwise not be heard in a group discussion, and less to those who would speak first. Whether that is an improvement depends on whose interests the governance is meant to serve.
**Implications for the design of decision governance**
The five places and time fall into three groups, and each is governed differently.
The situation can be governed through content: checklists and required fields, with a prompt for what they omit. The causal nature of the situation, the shared ground, perspectives, and the frame cannot be captured as fields. They are governed through process: steps that require a stated assumption about the kind of situation, a written list of assumptions, independently collected views, and alternative formulations of the decision. Time is governed through commitments made at the moment of deciding: a pre-mortem and stated conditions for revisiting.
Most decision processes rely on the first group. Yet that is where they are least likely to be missing something, because it is the only one they already check.
A decision record designed on this basis would hold more than the context the decision makers considered. It would also record which searches for context were made: the assumed kind of situation, the stated assumptions, the alternative formulations, whose views were collected, and the conditions for revisiting. This matters when the outcome of the decision later needs to be explained. The record then shows not only what was known, but where the decision makers looked, and where they did not.
**References**
- Brézillon, P., & Pomerol, J.-Ch. (1999). Contextual knowledge sharing and cooperation in intelligent assistant systems. Le Travail Humain, 62(3), 223–246. - Dey, A. K. (2001). Understanding and using context. Personal and Ubiquitous Computing, 5(1), 4–7. - Dourish, P. (2004). What we talk about when we talk about context. Personal and Ubiquitous Computing, 8(1), 19–30. - Duranti, A., & Goodwin, C. (Eds.). (1992). Rethinking context: Language as an interactive phenomenon. Cambridge University Press. - Ghidini, C., & Giunchiglia, F. (2001). Local models semantics, or contextual reasoning = locality + compatibility. Artificial Intelligence, 127(2), 221–259. - Godden, D. R., & Baddeley, A. D. (1975). Context-dependent memory in two natural environments: On land and underwater. British Journal of Psychology, 66(3), 325–331. - Goffman, E. (1974). Frame analysis: An essay on the organization of experience. Harper & Row. - Kaplan, D. (1989). Demonstratives. In J. Almog, J. Perry, & H. Wettstein (Eds.), Themes from Kaplan (pp. 481–563). Oxford University Press. - Klein, G. (2007). Performing a project premortem. Harvard Business Review, 85(9), 18–19. - Kurtz, C. F., & Snowden, D. J. (2003). The new dynamics of strategy: Sense-making in a complex and complicated world. IBM Systems Journal, 42(3), 462–483. - Lewis, D. (1979). Scorekeeping in a language game. Journal of Philosophical Logic, 8(1), 339–359. - McCarthy, J. (1993). Notes on formalizing context. In Proceedings of the 13th International Joint Conference on Artificial Intelligence (pp. 555–560). - Simon, H. A. (1990). Invariants of human behavior. Annual Review of Psychology, 41, 1–19. - Snowden, D. J., & Boone, M. E. (2007). A leader's framework for decision making. Harvard Business Review, 85(11), 68–76. - Sperber, D., & Wilson, D. (1986). Relevance: Communication and cognition. Blackwell. - Stalnaker, R. C. (1978). Assertion. In P. Cole (Ed.), Syntax and semantics 9: Pragmatics (pp. 315–332). Academic Press. - Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80(5), 352–373. - Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), 453–458. - Weick, K. E. (1995). Sensemaking in organizations. Sage.