Decision-Making Under Deep Uncertainty

Choosing and sequencing actions when probabilities cannot be assigned.

One robust path kept open across many futures Six faint paths fan out from a single origin across a wide field. One firm path runs through the middle of them, staying within the spread throughout. Two markers sit on that path at future moments, and from each a short dashed branch shows where the path would be taken instead if conditions changed.

At a glance

Produces
Options that perform acceptably across many futures, sequenced with signposts for revision.
Depends on
Agreement that the disagreement about models will not be resolved before a decision is needed.
Fails when
It is run as a search for the optimal option under a single probability distribution.

Deep uncertainty describes situations where the parties involved do not know, or cannot agree on, the relevant models, variables or probability distributions. Decision-making under deep uncertainty shifts the question from "what is most likely?" to "what performs acceptably across a wide range of plausible futures, and how should it adapt?"

How we approach it

Where uncertainty cannot be reduced, the decision itself has to change shape. That is usually a design problem rather than an analytical one.

  • We work with plural futures rather than a single distribution, so that disagreement about models does not have to be resolved before a decision can be made.
  • Options are assessed for robustness rather than optimality, and for how they fail rather than only for how they score.
  • Sequencing is treated as a decision variable. Deferring, staging, hedging and keeping options open are legitimate choices, not admissions of indecision.
  • Adaptive plans are designed with signposts and triggers, so the decision can be revised deliberately rather than under pressure.