Skip to content

Reasoning

Reasoning is the capability to use context and Organizational Memory to evaluate situations, compare options, identify implications, and prepare decisions.

Reasoning transforms memory and context into decision support.

Role in the Organizational Intelligence Cycle

Reasoning supports the Reason stage of the Organizational Intelligence Cycle.

It helps the organization answer:

  • What is happening?
  • What has happened before?
  • What options are available?
  • What risks or constraints apply?
  • What evidence supports each option?
  • What decision should be considered?
  • What confidence or uncertainty is present?

Reasoning should improve future decisions by making prior experience usable.

Responsibilities

The reasoning capability should:

  • retrieve and apply relevant Organizational Memory
  • use current context to evaluate the situation
  • apply policies, rules, constraints, and goals
  • compare possible actions
  • identify risks, tradeoffs, assumptions, and uncertainty
  • produce recommendations or decision support
  • preserve reasoning traces where appropriate
  • support human review for high-impact decisions

Inputs

Possible inputs include:

  • current context
  • Organizational Memory
  • prior decisions and outcomes
  • policies and rules
  • business goals
  • constraints
  • risk signals
  • human expertise
  • model outputs
  • confidence indicators

Outputs

Outputs may include:

  • explanations
  • options
  • recommendations
  • risk assessments
  • confidence scores or qualitative confidence
  • assumptions
  • decision rationale
  • escalation suggestions
  • review notes

Controls

Reasoning must be governed because it influences decisions and actions.

Controls should address:

  • decision authority
  • human review requirements
  • explainability expectations
  • model and rule versioning
  • policy compliance
  • bias and fairness considerations where relevant
  • auditability of reasoning traces
  • separation of recommendation from final decision
  • limits on automated reasoning

Quality Measures

Reasoning quality can be assessed through:

  • relevance of retrieved memory
  • consistency with policy
  • explainability
  • treatment of uncertainty
  • accuracy of risk identification
  • usefulness to decision makers
  • alignment with organizational goals
  • outcome quality over time
  • frequency of overrides or corrections

High reasoning quality means the organization can explain how context and memory informed a recommendation or decision path.

Anti-patterns

Common anti-patterns include:

  • generating recommendations without source evidence
  • ignoring prior outcomes
  • treating AI output as final authority
  • hiding assumptions in prompts, rules, or spreadsheets
  • applying policies inconsistently across teams
  • optimizing local metrics while harming organizational outcomes
  • producing explanations that cannot be audited
  • using stale memory or outdated policy

Implementation-neutral Examples

Examples of reasoning include:

  • comparing a current claim with similar prior claims, policy terms, evidence, and outcomes before recommending approval or escalation
  • evaluating a transaction against risk indicators, prior investigations, party relationships, and jurisdiction-specific constraints
  • recommending a customer service action based on current sentiment, account history, known issues, and previous resolution outcomes
  • identifying that a document exception should be reviewed because similar past exceptions led to compliance findings
  • suggesting workflow changes after repeated failures in a process step

Reasoning may use rules, analytics, AI, human judgment, or combinations of these. The framework requires explainable and governed reasoning, not a specific technology.

Unless otherwise noted, this document is licensed under CC BY 4.0.