01 Research program

Operational Decision Systems

What makes an analytical system operationally valuable?

Research begins where recurring operational questions resist one-off solutions. The objective is not simply to improve predictive performance, but to determine whether analytical capabilities produce better consequential decisions.

Models are therefore evaluated in relation to decision evidence, operational constraints, execution conditions, organizational processes, and human judgment.

One program, three connected questions
Question 01

When does an analytical capability create operational value?

Planning Decisions & Operational Value
Question 02

What evidence makes an AI-enabled operational decision trustworthy and deployable?

Decision Evidence, Traceability & Deployability
Question 03

How should partial information and uncertainty change operational action?

Decision Support Under Partial Information

Operational

Decision Systems

Models evaluated in relation to execution, evidence, and operational action.

02 Origins of the program

Where recurring operational problems became research questions.

The program developed gradually through repeated encounters with operational systems.

Engineering provided a language for constraints, tradeoffs, and interconnected decisions. Analytics expanded the methods available for making those relationships more legible. Operational practice then revealed a persistent gap: analytical performance alone does not determine whether a decision can be executed, reviewed, or sustained.

Engineering
Systems, constraints, and tradeoffs.
Analytics
Methods for uncertainty and choice.
Operational practice
Execution, reviewability, and sustained use.
Research
Questions examined beyond one setting.
See how recurring operational tensions shaped these questions

03 Research journey

From infrastructure research to operational decision systems.

  1. 01

    Early Applied Research

    Transportation infrastructure asset management introduced research as a way to study public-sector operational decisions.

    Mathematical modeling and asset-management frameworks connected analysis to maintenance, resource allocation, and infrastructure planning.

  2. 02

    Graduate Analytical Expansion

    Graduate study expanded the methodological toolkit and strengthened the connection among models, organizational context, and implementation.

  3. 03

    Operational Questions in Practice

    Professional practice repeatedly exposed the gap between analytical performance and executable, reviewable, and sustainable operational decisions.

  4. 04

    Operational Decision Systems

    Current work unifies questions of operational value, trustworthy decision evidence, deployability, and action under partial information into one long-term research program.

04 Research streams

Three connected research questions.

Three consistently defined streams examine value, trustworthy decision evidence, and operational action under partial information.

Question 01

When does an analytical capability create operational value?

Stream
Planning Decisions & Operational Value
Motivation
Model-level improvement does not necessarily produce a better executed decision. Operational value depends on the scope a capability can support, the changes it produces in planning, and the burden it places on real operating systems.
Representative problems
Forecast-to-plan translation · Planning capability attribution · Executed-plan change · Planning stability · Inventory and capacity constraints

Representative research

  • Capability-Level Operational Value in AI Planning
  • Planning Under Capacity and Stability Constraints
  • Semiconductor Supply-Chain Decision Systems

Question 02

What evidence makes an AI-enabled operational decision trustworthy and deployable?

Stream
Decision Evidence, Traceability & Deployability
Motivation
An operational decision cannot be trusted merely because a model produced it. Decision-time evidence must be valid, traceable, reviewable, and sufficient for the context in which the decision will be deployed.
Representative problems
Decision-time evidence · Temporal validity · Provenance and decision lineage · Deployment gates · Workflow reviewability · Lifecycle controls

Representative research

  • Decision-Ready Trace Construction
  • Evidence Gates for AI Deployability
  • Traceability and Governance in Operational AI

Question 03

How should partial information and uncertainty change operational action?

Stream
Decision Support Under Partial Information
Motivation
Incomplete, uncertain, and unstructured information should change which actions are taken, where additional evidence is acquired, and how limited resources are allocated.
Representative problems
Semantic signals · Condition uncertainty · Inspection targeting · Reverse logistics · Selective information acquisition · Human oversight

05 Selected active research

Current questions moving through the research pipeline.

Status is shown conservatively and changes only when the underlying scholarly record changes.

Operational decision research system A sequence from operational signals through evidence and uncertainty to a decision rule and operational action. Signals Evidence Uncertainty Decision rule Operational action
Across the portfolio, analytical evidence is carried forward to the decision rather than treated as an endpoint.
Accepted · Revision in ProgressLead authorIEOM Irvine 2026

When Does an Added Planning Capability Create Operational Value? An Observable Evidence Standard

What observable evidence shows that an added analytical capability changes executable plans and creates value under operational constraints?

Yihua Xu, Yan Lyu, Jize Li, Jiani He, You Lin, Yidan Hu, Rubing Qiang

Public preprint available ↗
Submitted · Under ReviewLead authorICOAI 2026

Policy-Bounded, Provenance-Aware Deployment Assurance for LLM Request Routing

When is the evidence behind an LLM routing decision sufficiently valid, attributable, and policy-compliant for deployment?

Yihua Xu, Jiani He, Ishita Chirag Talati, Yitian Qian, Youting Wang, Zhenyu Xu, Dingyan Shang

Public preprint available on SSRN ↗
RequestEvidence gateRoutePolicy
Manuscript in PreparationLead authorTarget: NeurIPS 2026 ML×OR

When the Nearest Analog Is Not the Best Evidence: Cold-Start Replenishment under Analog Uncertainty

How should a replenishment decision preserve uncertainty when several historical analogs remain plausible and similarity alone is not enough?

Yihua Xu, Yan Lyu

Plausible donorsEvidence mixStock decision
Accepted · Revision in ProgressSecond authorIEOM Irvine 2026

From Weak Signals to Supply Chain Reliability Risk: A Feedback-Driven Risk Accumulation Framework for Late-Delivery Prediction, Early Warning, and Risk Prioritization

How can matured operational signals accumulate into a leakage-controlled risk record that supports early warning and exception triage?

Yan Lyu, Yihua Xu, Jiani He, Jize Li, Rubing Qiang, Yidan Hu

Early-warning thresholdMatured signals → accumulated risk

06 Research approach

How recurring operational problems become researchable decision questions.

This is a research philosophy rather than a fixed methodology: the decision remains primary, while evidence and methods are shaped around what the operational setting requires.

  1. 01

    Recurring Operational Problem

    Research begins with decisions that practitioners repeatedly face.

  2. 02

    Decision Boundary

    The decision, available actions, and operational constraints are defined before a model is selected.

  3. 03

    Evidence Construction

    Evidence is organized around what was knowable at the moment the decision had to be made.

  4. 04

    Method Selection

    Methods are chosen to answer the operational question rather than applied uniformly.

  5. 05

    Validation

    Evaluation focuses on decision consequences, supported scope, and operational burden—not isolated metrics alone.

  6. 06

    Decision-Relevant Conclusion

    The objective is evidence that can support better operational decisions under real constraints.

07 Research outputs

A growing public research record.

The public record now connects early transportation-infrastructure research with current work on decision-aware intervention ranking for critical supply chains.

Published research Transportation Infrastructure Asset Management Technical report · 2019