AI Value Starts at Selection Part Two

Key Takeaways

• Target Root Causes Using the Iceberg Model: Diagnose operational issues layer by layer, structural changes are what deliver lasting impact, not superficial fixes.
• Automate Intervention Classification: AI models can classify operational events to flag whether what’s needed is relief, reform, adjustment, or a genuine cultural shift.
• Establish an Unbroken Lineage Architecture: Organisations should build a business architecture that links stakeholder value directly to AI use cases.
• Defeat “HIPPO” Prioritisation with Tangible KPIs: Without measurable KPIs, decisions end up resting on the HIPPO, reducing AI proposals to unverified hypotheses.
• Trace Lineage Bi-Directionally: Architecture should be validated both top-to-bottom and bottom-to-top, to confirm strategic alignment and surface any gaps.
• Deconstruct Decisions via DMN: Complex business decisions benefit from being structured hierarchically using DMN, which brings real clarity to grants allocation.
• Enforce Single-Point Executive Accountability: Each AI use case needs a designated, accountable owner to meet corporate governance requirements.
• Evaluate Proposals Across 8 Rigorous Pillars: Before deployment, initiatives should be evaluated across eight key dimensions to gauge effectiveness and impact.
• Accelerate Requirements with “Shadow Development” Prototyping: Architects can pair reference architectures and BPMN models with LLMs to prototype and validate quickly.
• Prioritise Data Governance Foundations: Successful enterprise AI depends on formal data contracts, clearly defined schemas, and robust data governance practices.

Webinar Details

Title: AI Value Starts at Selection Part Two
Date: 2026-08-27
Presenter: Howard Diesel
Meetup Group: African Data Management Community
Write-up Author: Howard Diesel

How does AI Address Underlying Systemic Problems Effectively?

Deploying AI without systems thinking tends to produce reactive firefighting, treating symptoms rather than solving the underlying problem. The Iceberg Model categorises challenges into four operational depths: surface events, recurring patterns and trends, systemic structures (policies and information flows, for instance), and cultural mental models or beliefs.

Visible (10%) [Events] → Reactive fixes
───────────────────────────┼──────────────────────────
Hidden (90%) │ [Patterns] → Trend analysis
│ [Structures]→ Policy & workflow reform
▼ [Mental Models] → Cultural & belief shifts

An AI classification model can diagnose incoming organisational events and recommend the right intervention layer for each. Emergency food relief, for instance, only addresses surface-level symptoms, while reforming tenant laws or establishing local farming structures resolves the root systemic constraints underneath. Systems don’t operate in a straight line either, symptoms and beliefs tend to reinforce each other through complex feedback loops rather than a simple top-down sequence.

Key Takeaways

• Address root causes: interventions aimed at systemic structures and mental models deliver lasting change, while surface fixes only offer temporary relief.
• Automate classification: machine learning models can be trained to categorise operational bottlenecks and assign the right intervention strategy.
• Recognise Systemic Loops: Organisations must account for non-linear feedback loops between immediate events and prevailing organisational beliefs.

FAQ

• How does the Iceberg Model differ from ITIL problem management? ITIL focuses on incident management and known error databases, whereas the Iceberg Model offers a broader framework that applies to operational, strategic, and belief-driven issues alike.

Figure 1 A Cross-functional Decision Workshop

Figure 2 The Iceberg Intervention Framework

Figure 3 The Gravity of the Quick Win

Figure 4 Depth 1: Event-level Interventions

Figure 5 Depth 2: Structure-level Interventions

Figure 6 Depth 3: Goal-level Interventions

Figure 7 The Intervention Diagnostic Matrix

Figure 8 The AGGPSA Connection: U05 Use Case

Figure 9 The U05 Automated Classification Engine

Figure 10 Decision D7: Portfolio Prioritisation

Are AI Initiatives Aligned with Enterprise Value Propositions?

AI initiatives frequently fail when the technical solution is disconnected from any demonstrable enterprise value proposition. A formal lineage architecture builds an unbroken, traceable thread running from high-level stakeholder outcomes all the way down to operational decisions and specific AI models.

[Stakeholder Value Proposition]
│
▼
[Business Outcome & KPIs]
│
▼
[Value Stream & Processes]
│
▼
[Operational Decision]
│
▼
[AI Use Case]

Every AI use case needs to explicitly answer three core governance questions: what decision is being improved, which stakeholders benefit, and which KPIs actually move. Deploying predictive analytics for educational “reading for meaning” programmes, for instance, requires mapping the interventions directly to literacy outcome metrics. That architectural discipline is what ensures investments produce defensible, evidence-based business results.

Key Takeaways

• Enforce Lineage: Link every machine learning implementation to targeted business decisions and specific value streams.
• Quantify Value: Require use-case proposals to specify targeted metric shifts and clear beneficiaries prior to funding.
• Ground the Theory of Change: Combine conceptual change frameworks with structural architecture to validate strategic feasibility.

FAQ

• Why must AI use cases link directly to business KPIs? Without direct links to measurable KPIs, proposals remain speculative hypotheses lacking quantifiable business evidence.

Figure 11 Connecting Interventions to Ultimate Impact

Figure 12 Business Architecture Lineage Explorer: Lineage Trace

Figure 13 Design Principle: how this Session Works

Figure 14 Five Artefacts you Leave with

How does the Interactive Architecture Navigator Enhance Models?

Enterprise architecture models often stay trapped in static presentations, which stops practitioners from navigating end-to-end dependency chains. An Interactive Architecture Navigator puts structured metamodels into practice, letting users drill down from high-level stakeholder personas all the way to operational execution nodes.

[Stakeholder Persona: Entrepreneur]
│
▼
[Value Proposition: Township Journey Upliftment]
│
▼
[Operational Decision: Grant Go / No-Go]
│
▼
[AI Decision Agent: Next-Best Intervention]

Underneath, the engine runs on a multi-tab relational data model defining personas, value propositions, capabilities, processes, decisions, and supporting AI agents. Grant allocation workflows, for instance, can be examined to check whether candidate models, drop-off prediction or next-best intervention algorithms have the data inputs they need.

Key Takeaways

• Replace Static Maps: Interactive graph and tree navigators allow real-time exploration of operational hierarchies and capability models.
• Trace Decision Points: Explicitly link specific process stages—such as grant evaluations—to candidate automated agents.
• Structure Source Data: Maintain relational metamodels covering personas, process stages, and data dependencies to power interactive views.

FAQ

• What is the underlying data structure powering the architecture navigator? The navigator runs on a multi-tab relational data model, maintained in Excel or structured JSON, that explicitly maps the relationships between stakeholder personas, processes, decision points, and potential AI models.
• How do personas factor into technical architecture mapping? Personas represent specific external or internal beneficiaries, township entrepreneurs or grant evaluators, for example, which keeps technical capabilities focused on concrete user needs and real-world workflows.

Figure 15 Knowledge Area Maturity Radar

Figure 16 Step 10: AI Use-case

Figure 17 Closeout, Evaluation & Dialogue

Is your Project Aligned with Critical Business Metrics?

Without a structured evaluation framework, technology prioritisation tends to default to the HIPPO, the Highest Paid Person’s Opinion. The Business Architecture Spine cuts down on that subjective decision-making by establishing an auditable, hierarchical baseline for every project.

Spine Element Governance Verification Question
────────────────────────────────────────────────────────────────────
Stakeholder Who is the target beneficiary?
Value Proposition What tangible value is created?
Value Stream Stage Where does this sit in the customer journey?
Process & Decision Which operational decision is being automated?
Business KPI What quantifiable metric will change?

If a team can’t demonstrate which business process, decision, and metric a proposed model adjusts, its evidentiary weight takes a real hit. Rigorous spine alignment is what provides the formal accountability modern governance codes like King V demand, keeping board-level signoffs grounded in empirical proof rather than assumption.

Key Takeaways

• Defeat Subjectivity: Replace subjective executive opinions with a verifiable, metric-driven evaluation spine.
• Downgrade Speculative Work: De-prioritise proposals that fail to define measurable KPI impacts.
• Meet Governance Standards: Fulfil executive governance mandates by proving the systemic worth of technological initiatives.

FAQ

• What is the “HIPPO” effect in AI project selection? HIPPO stands for the “Highest Paid Person’s Opinion”, a dynamic where technology investments get chosen based on executive preference rather than empirical business evidence.
• What happens if an AI proposal cannot be tied to a specific KPI? If a team can’t identify the exact decision, process, and business metric a model impacts, the proposal gets classified as an unverified hypothesis and de-prioritised from production funding.

Figure 18 The Anatomical Blueprint of the AI Enterprise

Figure 19 The Spine (the Golden Thread)

How can we Ensure Bi-directional Gap Analysis?

A common failure in enterprise modelling is mapping the path from strategy down to execution while missing broken links running the other way. Robust architecture needs bi-directional gap analysis, top-to-bottom and bottom-to-top, to catch orphaned initiatives and disconnected data assets.

Sticking to standard frameworks like the Business Architecture Guild’s BIZBOK keeps models grounded in proven enterprise canvas principles. Architectures need to align external customer journeys with internal value streams, which is what clarifies why an organisation exists and what value it delivers. Automated gap-export utilities can quickly reveal disconnected nodes too, which stops organisations funding orphaned use cases.

Key Takeaways

• Audit Bi-Directionally: Validate strategic alignment downward and trace operational execution upward to ensure zero structural gaps.
• Adopt BIZBOK Standards: Structure value streams, stages, and capabilities according to established architectural industry standards.
• Expose Disconnected Nodes: Use automated validation tools to locate missing relationships across the capability landscape.

FAQ

• Why is bi-directional gap analysis necessary? Evaluating top-to-bottom confirms business strategy has real execution paths, while evaluating bottom-to-top makes sure technical assets, models, and data products aren’t left orphaned without a supporting business purpose.
• How does BIZBOK fit into AI architecture? The Business Architecture Guild’s BIZBOK standard provides formal structures for mapping value streams and capabilities, keeping AI tools aligned with established industry architectural practice.

Figure 20 Tree Explorer

How can DMN Improve Organisational Decision-making Processes?

Complex organisational initiatives often stall because process boundaries are blurred and project sponsorship is spread too thin. Resolving that takes formal Decision Model and Notation (DMN) hierarchies, paired with explicit target operating models.

DMN breaks primary business decisions, portfolio grant allocation, say, down into contributing sub-decisions, identifying exactly which knowledge sources and data inputs feed each one. Alongside that, the operating model needs a named internal business owner responsible for funding, execution, and long-term outcomes. Separating external beneficiaries from internal budget owners this way is what prevents cross-functional paralysis.

[High-Level Decision: Grant Allocation]
┌──────────────────────┴──────────────────────┐
▼ ▼
[Sub-Decision: Eligibility] [Sub-Decision: Impact Score]
│ │
(Input Data & Business Rules) (Input Data & Knowledge Sources)

Key Takeaways

• Deconstruct Decisions: Apply DMN principles to break complex macro-decisions into transparent, automated sub-decisions.
• Assign Budget Ownership: Designate a single internal business role as the accountable budget holder and owner.
• Formalise Operating Roles: Standardise role definitions across grant making, technical quality, and portfolio governance.

FAQ

• Why use Decision Model and Notation (DMN) for AI use cases? DMN breaks complex macro-decisions down into transparent sub-decisions, making it clear exactly where automated AI predictions fit alongside business rules and human judgement.
• Why must governance assign accountability to a role rather than an individual? Assigning ownership to an organisational role (such as Head of Portfolio or Chief Risk Officer) ensures institutional accountability persists through staff turnover and satisfies governance standards like King V.

Figure 21 Policy, Model, and Glossary

Figure 22 Navigation Graph

Figure 23 Grant-making & Programmes

How do we Assess AI Use Cases Effectively?

Assessing an AI use case takes a dual perspective: looking upward to prove its systemic business worth, and looking downward to evaluate technical and ethical risk. Upward analysis anchors the initiative to business strategy, while downward analysis audits data readiness, system controls, and data products.

▲ Upward Trace: Systemic Business Worth
│ (Strategy, KPIs, Stakeholder Outcomes)
│
[ AI Use Case / Decision Node ]
│
▼ Downward Trace: Technical & Ethical Risk
(Data Products, Schemas, Quality Rules, Human Safeguards)

Organisations need to define data products complete with clear contracts, schemas, and automated quality rule validation before any model gets built. An ethical stewardship framework should also be built into the metric tree, to establish clear boundaries, operational guardrails, and escalation thresholds for AI systems.

Key Takeaways

• Balance Value and Risk: Trace upward to confirm business relevance and downward to verify technical feasibility and data health.
• Enforce Data Contracts: Demand formal data schemas, defined contracts, and automated quality scores prior to development.
• Embed Ethical Controls: Define explicit authority limits, human override pathways, and operational guardrails.

FAQ

• What is the difference between upward and downward lineage tracing? Upward lineage traces an AI use case back to strategic business outcomes and KPIs, while downward lineage audits technical feasibility, data schemas, data quality contracts, and ethical guardrails.
• What role do data contracts play before building an AI model? Data contracts define the expected schema structures, field constraints, and automated quality thresholds needed to make sure underlying data products are stable and reliable before model training even starts.

Figure 24 The Diagnostic Axis

Figure 25 Iceberg-depth Intervention Classification

Figure 26 Ecosystem Conditions Dataset – Data Contract

Figure 27 Ethical Stewardship

How does the 8-Pillar AI Framework Ensure Safety?

Moving projects safely from concept to production takes an objective readiness evaluation. The 8-Pillar AI Evaluation Framework grades initiatives systematically across core dimensions to prevent high-risk deployments:

• Relevance & Problem Definition: Target beneficiaries and validated decision improvements.
• Data Quality & Readiness: Contract completeness, schema stability, and baseline data quality.
• Cost Effectiveness & Sustainability: Outcome ROI, total cost of ownership, and value metrics.
• Human-Centred Design & Control: Confidence thresholds, appeal workflows, and human-in-the-loop triggers.
• Governance & Operating Roles: Clearly designated executive sponsors and assigned operational teams.
• Ethics & Safeguards: Hallucination mitigation, bias checks, and guardrail enforcement.
• Monitoring & Evaluation: Post-deployment feedback collection and drift detection.
• Technical Performance: Latency, infrastructure dependencies, and system integration.

[8-Pillar Assessment Scorecard]
│
┌──────────────┼──────────────┐
▼ ▼ ▼
[Stop] [Pilot] [Scale]
(Failing/Red) (Controlled) (Production)

Under corporate governance standards like King V, executive accountability needs to sit with specific job titles and roles, that’s what keeps institutional oversight uninterrupted.

Key Takeaways

• Apply Multidimensional Scoring: Evaluate all proposed models across the 8 standard pillars prior to enterprise roll-out.
• Define Confidence Thresholds: Trigger human-in-the-loop workflows whenever automated decisions fall below set certainty margins.
• Institutionalise Oversight: Assign formal AI responsibility to durable executive titles rather than individual employees.

FAQ

• What determines whether an AI model triggers a human-in-the-loop review? Models carry predefined confidence score thresholds, and any automated prediction that falls below the threshold gets automatically escalated to a human operator for review.
• What are the primary gates in the 8-pillar evaluation? Proposals get evaluated across eight dimensions, relevance, data quality, cost-effectiveness, ethics, and performance among them, producing a score that determines whether the initiative should be halted, piloted, or scaled.

Figure 28 Five Artefacts you Leave with

Figure 29 The Challenge

Figure 30 Eight Dimensions – One Shared Scorecard

Figure 31 Build 1: Problem, Value and Theory of Change

Figure 32 Evaluation Integrity Challenge

Figure 33 Table Decision: Stop, Pilot or Scale?

Figure 34 Model Eight-dimension Result

How does Rapid Prototyping Validate Operational Requirements?

Agile architecture depends on rapid prototyping to validate operational requirements before committing to heavy software engineering cycles. Pairing structured reference architectures with Large Language Models like Claude lets teams convert raw tabular data and process definitions into standard BPMN swim lane diagrams and interactive prototypes.

[Tabular Metadata / Process Steps]
│
▼ (LLM Prompting & Validation)
[BPMN Diagrams & Decision Logic]
│
▼ (Rapid Prototyping)
[Interactive Web UI / Schema Validation]
│
▼ (Handoff)
[Production Engineering (e.g. React)]

Modern architectures make use of lightweight GitHub Pages interfaces, version-controlled JSON data schemas, and ArchiMate-compliant model exports. This “shadow development” approach clarifies business requirements and data dependencies within a 7-day starter kit timeline, and rapid prototyping is what ensures production engineering teams build against requirements that are validated, backed by a renewed enterprise focus on core data governance.

Key Takeaways

• Accelerate requirements via prototyping: use LLMs to convert narrative process definitions into structured BPMN models and functional prototypes.
• Decouple Prototyping from Production: Validate schemas and workflows rapidly before handing blueprints to formal engineering teams.
• Anchor in Governance First: Align technical architectures with strong metadata and data governance practices before scaling enterprise AI.

FAQ

• What is “shadow development” in the context of architecture prototyping? It refers to rapid, low-overhead prototyping by product architects, using LLMs and lightweight code to refine requirements before committing enterprise software engineering resources.

Figure 35 Model Editor: Stakeholder

Figure 36 Lifecycle – BPMN 2.0 Swimlane

Figure 37 The Seven-day Blueprint: from AI Hype to Architected Reality