Decision Intelligence state of the art¶

Candidate v0.9.0-rc.1 · Focused assessment, not a systematic review

Executive conclusion¶

Decision Intelligence (DI) is treated here as an emerging integrative field of practice: it brings mature disciplines and practices together around the full decision cycle, but the review does not support presenting it as a consolidated scientific discipline with universally accepted boundaries, methods and a corpus of its own.

This is a dated, falsifiable candidate assessment. Absence of evidence in this corpus is not evidence that no such work exists.

1. What is DI and where does it come from?¶

DI shifts attention from producing information or models to designing, executing and learning from decisions. Its lineage does not begin with one technology:

Layer	Contribution to DI	Boundary
Decision theory and analysis	Alternatives, uncertainty, preferences, value and process quality	Do not cover continuous operations or conversation by themselves
Operations research and DSS	Models, optimization and computational support	An optimum does not guarantee adoption, control or learning
Data and information quality	Fitness of evidence for use	Data quality is not decision quality
Modeling and automation	Explicit decisions, rules and reproducible execution	Automation does not resolve legitimacy, accountability or causality
Organizational learning	Feedback, exploration and exploitation	Recording decisions does not prove that an organization learns
Human–AI Interaction	Collaboration, deliberation and control design	Effects depend on task, interface, user and comparator

2. How does it differ from adjacent fields?¶

Field	Primary question	Relationship to DI
Business Intelligence	What happened and how can it be observed?	Provides visibility; DI connects evidence to decision and action
Analytics and Data Science	What pattern, forecast or effect can be estimated?	Provide models; DI adds authority, alternatives, guardrails and follow-through
Decision Analysis	How should a decision under uncertainty be structured?	A direct foundation, not a predecessor that DI replaces
Decision Support Systems	How can systems support a decision?	Provides a sociotechnical and operational tradition
EDM and DMN	How can repeatable decisions be modeled, managed and automated?	Mature evolving practices in the map, not synonyms for DI
Conversational Analytics	How can data be explored with natural language?	Conversation is an interface; it does not guarantee a governed decision
AI agents	What tasks can a system plan or execute?	Agency increases the need for rights, stop conditions and traceability

3. What is established?¶

“Established” does not mean infallible or universal. It means that the component has its own lineage and verifiable primary or official sources:

decision analysis and structuring under uncertainty;
operations research and decision support systems;
use-dependent data and information quality;
outcome bias and separation of process from result;
scoring rules for repeated probabilistic forecasts;
decision modeling through the formal DMN specification;
AI risk management through public frameworks such as NIST AI RMF;
organizational learning as a feedback and adaptation problem.

4. What remains emerging or contested?¶

what boundary separates DI from a well-governed integration of existing disciplines;
when conversational interfaces improve deliberation or merely increase fluency and confidence;
how to compare human, human+AI and AI alone without mistaking acceptance for quality;
what meaningful human control means when an agent can act;
how to measure outcome contribution without attributing causality to the interface;
whether DI has its own cumulative corpus and under which criteria.

The human–AI evidence reviewed is context-dependent. An exploratory deliberation study does not authorize a universal effect; an evaluation in another domain may find no improvement. In high-stakes decisions, model interpretability may be a preferable design property to a post-hoc explanation of a black box.

5. What does PULSE integrate?¶

PULSE organizes its own synthesis around:

PDAMR: Priority, Decision, Action, Metric and Risk;
Decision Circle: reality, decision, action, result and learning;
five verbs: frame, understand, decide, act and learn;
Human-in-Control: explicit authority, escalation and accountability;
Decision Experience: evidence and context turned into an action-oriented experience.

Constitutional sources govern this identity. The author's texts explain its lineage and purpose. Neither source type proves by itself that PULSE produces better outcomes than an alternative.

6. What has not been validated?¶

It has not been established that:

CDI is a mature scientific discipline or that the project coined the term;
PULSE is superior, universal or causally effective;
the L0–L7 architecture is approved doctrine;
interoception, allostasis or a “decision organism” are literal organizational mechanisms;
five levels of analytics form a universal maturity scale;
conversation, copilots or agents always improve a decision;
the B2B case represents an executed intervention or attributable outcome.

7. What evidence would change the conclusion?¶

Claim strength should increase or decrease in response to:

a reproducible systematic review of the DI corpus;
multi-context comparisons of human, human+AI and AI alone;
published replications and null or adverse results;
preregistered PULSE pilots with baseline, comparator, guardrails and follow-up;
external evidence about the validity, usefulness and cost of its artifacts;
new official versions of registered standards and frameworks;
independent academic and professional review of the map.

Decision boundary: this map supports decisions about what to investigate and how to communicate precisely. It does not authorize saying that CDI or PULSE has already been scientifically validated.

Selected primary sources¶

Continue with the PULSE evidence map or the research agenda.