Control · C112

Probability the plan holds: committed conditions, bounds and a defensible scope

Question addressed

What probability does the plan have of meeting its target, when it rests on several conditions each of which looks likely?

Direct answer

A plan is never more likely than its weakest condition, and two bounds follow without assuming anything; multiplying the probabilities assumes an independence the assumption register does not establish, and a plan whose bounds say nothing is committed to too many conditions.

Decision supported

01Cut the number of committed conditions, or lift the weakest, before defending a figure in committee.

02The rule, the evidence and the next review date stay visible.

How to use this

Move from the question to the deliverable.

01Starting point

What probability does the plan have of meeting its target, when it rests on several conditions each of which looks likely?

02Decision prepared

Cut the number of committed conditions, or lift the weakest, before defending a figure in committee.

03Deliverable to use

Register of committed conditions

Calcul de bornes · SQ-112

Bound the plan’s probability, then reduce what it commits to

A plan built on six conditions each judged likely is not a likely plan. Two bounds can be computed without assuming anything, and they decide: when they say nothing, the plan carries too many conditions, and the remedy is not to estimate better but to commit to less.

Working context

Illustrative case: the annual marketing plan of a B2B software vendor, a revenue target committed in committee, six conditions recorded in the assumption register with their confidence level, and no dependence recorded between them.

  1. 01List

    Take from the assumption register only those conditions the committed target depends on, with their probability and the consequence of their failure. An assumption that does not move the target is not a condition.

  2. 02Cap

    Take the smallest probability: the plan cannot be more likely than that. This ceiling only rises by treating that one condition.

  3. 03Bound

    Compute the lower bound as the sum of the probabilities less the number of conditions minus one, floored at zero. The gap between the two bounds measures what the plan does not know about its own dependencies.

  4. 04Release

    Move the conditions that can be reclassified as options out of the committed scope, until the two bounds close enough for a figure to be defensible.

Original element

Bounds on the probability of the plan

The calculation takes the committed conditions and their probability, derives the plan’s ceiling and lower bound, then compares the full scope with a scope reduced by two conditions reclassified as options.

Calculation methodUpper bound = smallest committed probability. Lower bound = max(0; sum of the probabilities − (number of conditions − 1)). The product of the probabilities, 30.5% then 58.1%, is only one point in the interval and assumes independence.
Committed conditionProbabilityConsequence if it failsCan become an optionPlan on six conditionsPlan on four conditions
Acquisition lead hired in Q10.85Capacity cut by a quarterNoCommittedCommitted
Conversion held at 2.4%0.80Pipeline falls shortNoCommittedCommitted
Site rebuild delivered in June0.70Autumn campaigns postponedYesCommittedOption
Media budget not frozen in H20.90Paid acquisition stopsNoCommittedCommitted
Integration partner available0.75Go-live pushed backYesCommittedOption
Price held without further discount0.95Margin below thresholdNoCommittedCommitted
Probability the plan holds———0% to 70%50% to 80%
CSV · conditions · probabilities · bounds · scopesDownload the probability bounds calculation
Decision interpretation

01The ceiling is set by the weakest condition, and by it alone: 70% on the full scope, 80% once the site rebuild is taken out. Strengthening any other condition does not raise it by a single point. A conjunction is never more probable than one of its terms, and that rule admits no exception.

02On six conditions the plan says nothing about itself: the interval runs from 0% to 70% and does not even rule out certain failure. This is not a flaw in the calculation but in the instrumentation, because the assumption register records one probability per line and no dependence between lines, while it is dependence that places the plan inside the interval.

03Taking two conditions out of the committed scope brings the interval between 50% and 80%, and a figure becomes defensible again. The decision is therefore not to estimate better but to commit to less: the two conditions stay in the plan, they simply stop carrying it.

04The error this calculation corrects has been measured, and among professionals. Tversky and Kahneman observed it on one hundred and fifteen forecasters gathered at the Second International Congress on Forecasting: they judged a conjunction more probable than one of its own terms, 0.47% against 0.14%. The authors name our very case, a sequence of plausible steps that looks safer than its failure, which can happen in a thousand unlikely ways. Detailing the scenario deepens the illusion instead of reducing it.

Illustrative case. Both bounds are exact and assume nothing beyond the probabilities entered in the register. The product, by contrast, assumes the conditions are independent, which the plan establishes nowhere. The starting probabilities remain judgements: this page bounds how they combine, it does not validate their value.

Coherence controls

Minimum conditions before validation.

01Versioned assumptions

The status, the evidence and the owner must be available for review.

02Quantified exposure

The status, the evidence and the owner must be available for review.

03Observable triggers

The status, the evidence and the owner must be available for review.

04Fundable responses

The status, the evidence and the owner must be available for review.

Usable deliverable

Output produced

Register of committed conditions

Explore the decision further with an interactive control, or start from a structured template to produce your deliverable.