Decision Analysis

Goal

Decision analysis produces a defensible choice among several options, evaluated against explicit criteria. Without it, the choice falls to gut judgement, to emotion or to the loudest voice in the room. The term covers a family of techniques that share a single spine: frame the problem, define the options, evaluate them against agreed criteria, choose, then implement the choice. The techniques in the family differ in how they conduct the third step, the evaluation, where the choice is decided.

Its strength lies in making a decision debatable. Faced with a complex or uncertain situation, it forces stakeholders under pressure to compare options against criteria, it forces an honest weighting of what matters and it puts financial and non-financial criteria on one scale. The central deliverable is the traceable justification attached to the chosen option: given these options, these criteria and this weighting, here is why this one wins. That preserved reasoning can be challenged, corrected and replayed later.

Usage

When to use it

• Contested or high-stakes choice: several options exist and the decision must be defensible to others.
• Disagreement over what matters: stakeholders diverge on the value of outcomes and need a common basis for comparison.
• Decision to be traced: governance, regulation or sign-off on a spend calls for a justification that holds up later.
• Financial and non-financial criteria mixed: heterogeneous dimensions have to be compared on one scale.

When not to use it

• Decision needed on the spot: decide on judgement and record the reasoning briefly, with no formal procedure.
• Evaluation information unavailable in time: run elicitation or collection first, otherwise accept a judgement call rather than an analysis.
• Trivial, reversible decision of little value: even the lightest structured analysis costs more than it returns, decide directly with no formal step.

Choosing the right technique

The choice of technique reads off a few discriminants, from the most decisive to the finest: how many options are in play, whether the criteria carry weights and whether those weights are contested, whether the question is a go or no-go on a single change rather than a comparison among rivals, what rigour the decision must be able to withstand and whether the stakes justify the effort asked. One last axis separates techniques at comparable effort: some aim at buy-in and dialogue, others at analytical precision.

The first fork isolates a case apart. When the question is "should we make this change?" and not "which of these options?", force field analysis is the tool: it frames a single proposed change into driving forces and restraining forces, without ever setting options against one another. If you find yourself comparing three variants, you are already in the territory of the decision matrix or of multi-criteria analysis.

When several options are in competition, the scale runs from informal to formal. For a binary choice or a single option, low-stakes and legible to anyone, pro-versus-con analysis is enough: two columns, no criteria, no weighting, no score, it is the fastest and the cheapest. Adding weighted criteria to it would oversize the exercise.

As soon as several options are evaluated against several criteria, and those criteria can carry different weights, the decision matrix is the default choice, the one most decisions fall to: a transparent score with no heavy machinery. When the weight of the criteria itself becomes the object of disagreement, when it cannot be set by simple inspection or when the decision must withstand close scrutiny, the analytic hierarchy process (AHP) derives those weights by pairwise comparison and tests their consistency, a rigour a flat matrix does not provide. When the criteria are many and conflicting, the value at stake high and several deciders around the table, multi-criteria decision analysis (MCDA) offers the most complete apparatus for trade-offs, at the cost of the most demanding method in the family.

AI considerations

AI helps mainly upstream and at the margins of the reasoning. From a problem statement, a language model proposes candidate options and a first list of criteria, which widens the field before humans prune it and corrects the tendency to freeze too early on a set of options that is too narrow. Once the matrix or the model is filled in, AI excels at the mechanical recomputation that analysts readily skip: reviewing the weights through sensitivity analysis, checking whether the winning option flips when a given weight moves, flagging dominated options, computing and explaining a consistency index. It also writes up the narrative of the justification decently from the finished model.

Human judgement stays whole where the decision is really made. The values and the weights themselves are a matter of arbitration by the organisation and the stakeholders, the very thing the technique forces into the open: a weight supplied by an AI dresses an unowned arbitration in false objectivity. The meaning of the criteria and the political or cultural nuance behind them escape the model, which does not know which "cost" the organisation fears. The decision itself belongs to the person accountable for it, all the more since the numeric result already tends to look more certain than it is, an overconfidence that AI amplifies. Finally the option and criteria data are often sensitive, supplier bids, salaries, amounts in CHF: they are not poured into a tool without the authorisation that permits their processing.

Sources

• IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.16 Decision Analysis.
• Saaty, T. L., The Analytic Hierarchy Process, McGraw-Hill, 1980.
• Belton, V. and Stewart, T. J., Multiple Criteria Decision Analysis: An Integrated Approach, Springer, 2002.
• Lewin, K., Field Theory in Social Science, Harper & Row, 1951.