Sami TayaraData & AI Governance
Home Original thinking

Original thinking

The Two Wings
AI Framework.

Assess business value and operational trust together, from the initial use case through the decision to deploy and scale.

Original thinking

Governance across value, trust, and the decision gate

The Two Wings AI Framework evaluates whether an AI initiative is worth doing and whether it is being done right. Each wing comprises six elevations, developed in parallel from AI-Worthy Processes and AI-Ready Data. Governance spans both wings and the AI Decision Gate: strategy defines value appetite, trust thresholds, and forbidden zones; use-case evidence informs the decision to proceed, strengthen readiness, or discontinue.

Value

The value wing

Assess the business opportunity, organizational readiness, and economics of deployment and scale.

1AI-Worthy Processes

Map business capabilities and processes to identify suitable AI use cases. Assess strategic relevance, process characteristics, and available data, then establish a prioritized register with accountable owners.

2People & Change Readiness

Assess adoption readiness, the capacity to redesign workflows, effects on roles, and reskilling requirements. Establish executive sponsorship and the changes needed to realize the expected benefit.

3ROI & Value Modeling

Develop the business case using expected benefits, full implementation and operating costs, and risk-adjusted assumptions. Use the analysis to prioritize investment and establish measurable value targets.

4Pilot Value Proving

Evaluate the pilot in a representative workflow against success criteria agreed in advance. Use observed results to inform the decision to advance, revise, or discontinue the use case.

5Value Capture & Ownership

Assign business ownership for benefits and measure realized value against the business case. Establish the reporting and review process needed to sustain benefits and address shortfalls.

6Portfolio & Scale Economics

Manage AI investment as a portfolio. Review operating costs, unit economics, dependencies, and realized benefits to determine which initiatives should scale and which should be revised or discontinued.

Reference disciplines: TOGAF Business Architecture, capability mapping, value streams, and capability-based planning.

Trust

The trust wing

Establish data readiness, accountable ownership, operational controls, and evidence of trustworthy performance.

1AI-Ready Data

Assess data quality, meaning, lineage, ownership, permissible use, and access controls for the intended AI use case. Establish the data improvements required for reliable development and operation.

2AI Visibility & Ownership

Maintain an inventory of models, agents, and embedded AI capabilities, with accountable owners and risk classifications. Include AI introduced through third-party applications and services.

3Model & Output Assurance

Evaluate model and output performance throughout the lifecycle. Apply benchmarks, fairness assessments, drift monitoring, and explainability requirements proportionate to the use case and its consequences.

4Agentic & Runtime Control

Define and enforce agent permissions, tool and data access, authorization of consequential actions, runtime monitoring, and intervention mechanisms. Retain evidence of actions and control decisions.

5Human Oversight & Accountability

Assign decision, review, override, and accountability roles. Define escalation procedures and the circumstances requiring human intervention before deployment.

6Compliance & Evidence

Maintain audit trails, assessment records, and incident evidence aligned with applicable obligations. Compliance evidence is produced through the controls and accountable practices established across the framework.

Reference frameworks and obligations: NIST AI RMF, ISO/IEC 42001, and the EU AI Act.

The lattice (AI Pilot Quadrant)

What it is

A framework for assessing AI use cases across Value and Trust, with six elevations on each wing and a common AI Decision Gate.

Who it’s for

Executive sponsors, data and AI leaders, and the architecture, risk, and compliance owners responsible for AI investment and deployment decisions.

When to use it

At use-case definition, pilot review, and subsequent deployment and scale decisions. Reassess when the system, data, operating context, or risk profile changes.

How to use it

Set the strategic bounds, register the use case, capture evidence, and assess the twelve elevations. Apply the value and trust thresholds at the AI Decision Gate, then document the verdict and its supporting evidence.

Paper and provenance

Methodological foundations.

The framework brings together data management, enterprise architecture, and the economics of AI. Its foundations draw on DAMA-DMBOK, TOGAF, the NIST AI Risk Management Framework, ISO/IEC 42001, and the EU AI Act. The research paper is available on SSRN. The published framework is at twaf.ai.

The Two Wings AI Framework: A Practical Guide ↗