Inteq's Agentic AI Q&A Series
Question: How do you formally model SME tacit knowledge and decision logic for AI agent design?
Answer: Externalizing tacit SME decision logic into agent-usable artifacts is the single most consequential business analysis activity in any Agentic AI initiative. The work that experienced subject matter experts perform at each decision point – such as the conditions they evaluate, the patterns they recognize, the thresholds they apply, the judgment they exercise - is rarely written down. It exists as accumulated expertise in the operator's mind. Agents cannot inherit this expertise by observation; it must be deliberately decomposed, structured, validated, and translated into explicit artifacts that the agent can execute and the governance team can audit.
A model-driven analysis approach uses a small set of integrated visual artifacts that work together. The first is the decision table, which structures decision logic as a matrix of conditions and outcomes. For each decision point, the analyst captures the input conditions the SME evaluates, the threshold values that trigger different outcomes, and the resulting actions. Decision tables surface gaps and contradictions immediately. When SMEs review their own decision logic in tabular form, they often discover that the rules they thought were consistent actually vary based on context, or that two SMEs handling the same decision apply different thresholds. The decision table is also directly translatable into agent decision rules with minimal reinterpretation.
The second is the decision tree or decision-flow diagram, which maps the sequence of decisions and the conditions that determine which decision is made next. Where the decision table captures the logic of a single decision, the decision tree captures how decisions chain together to produce a process outcome. This artifact is essential for agent design because agents do not make isolated decisions, they reason through sequences of decisions where each decision's outcome conditions the next. The decision tree externalizes that sequencing logic.
The third is the state transition diagram, which captures how the state of a process instance changes as decisions are made. State transition modeling is particularly valuable for agent design because it forces explicit definition of every state the work item can occupy, every transition between states, and the conditions that govern each transition. Agents operate by reasoning about state and selecting appropriate transitions; without explicit state modeling, the agent's behavior becomes ambiguous.
The fourth is the information requirements model, which specifies what data each decision requires, where that data resides, and what confidence level the SME currently has in that data. This artifact directly informs the agent's data integration architecture and the data confidence assessments that determine agent autonomy boundaries. SMEs often assume access to data they actually obtain through informal channels or compensate for through experience; the information requirements model surfaces these dependencies explicitly.
• Walkthrough validation: Brings multiple SMEs together to review the same decision logic. Disagreements among SMEs are not problems to suppress but signals that the decision logic was never as consistent as assumed.
• Historical replay validation: Applies the externalized logic to past decisions and compares the predicted outcomes to what actually happened. Significant divergence indicates that the logic captured does not match the logic actually applied.
• Edge case validation: Deliberately tests the externalized logic against unusual situations—the cases where SME judgment most differs from rule-following. The agent will encounter these cases in production, and the logic must handle them or escalate appropriately.
The output of this work is an integrated set of artifacts: decision tables, decision trees, state transition diagrams, information requirements models that together specify the decision logic at sufficient detail to inform agent design and at sufficient transparency to support governance review.
* * *
Related Posts:
The Agentic AI Ontology Question
Data, Meaning, Reasoning and Agentic AI
The PR/FAQ Is a Scoping Document - Not a Specification
Spec-Driven Development Starts with Model-Driven Analysis
Related Consulting Services:
Agentic AI Readiness & Strategy Analysis
AI Agent Opportunity & Portfolio Design
Business Process Mapping
Process Improvement & Reengineering
Related Training Courses:
Discovering Agentic AI Opportunities
Analyzing and Specifying AI Agent Business Requirements
* * *
Visit our Insights Hub
Visit my YouTube Channel | Connect with me on LinkedIn
Check out our business analysis Training Courses and Consulting Services
Contact us at info@inteqgroup.com
