Key Takeaways
- Foundation in Proven Frameworks: AI project execution builds upon established analytical lifecycles, adapting CRISP-DM, PMI’s CPMAI (Continuous Project Management for AI), and the INFORMS CAP/DSBoK standards.
- Guardrails Before Business Understanding: Data governance guardrails, security standards, and privacy protections (specifically mitigating person re-identification risks) must be established before initiating business use cases.
- Ethics Precedes Financial Evaluation: Ethical assessments must occur prior to calculating business value or financial ROI, preventing executives from overriding ethical concerns in favour of projected monetary gains.
- Rigorous Control Gates: Project governance relies on explicit decision gates to kill, pivot, halt, pilot, or deploy use cases based on data supply stability and model confidence.
- Integration of CPMAI and DMBoK: AI project management (CPMAI) cannot operate in isolation; each phase must map directly to DAMA DMBoK data governance controls, metadata management, and business architecture (BizBOK).
- Multi-Layered Quality for Non-Deterministic AI: Quality evaluation must extend beyond raw data quality to include business quality, decision quality, and records quality. Non-deterministic LLM outputs require specialised metric ensembles (such as Dale Rutherford’s BME for hallucination and tonality).
- Process-Driven Architecture Over Silos: AI decisions and agentic workflows must follow end-to-end business processes rather than functional silos to prevent downstream compounding errors.
- Maturity-Based Phasing (Crawl, Walk, Run): Organisations should enforce strict rules and limit concurrent use cases during the initial “Crawl” stage before scaling up to parallel portfolios and automated decisioning in the “Run” stage.
- Total Cost of Ownership (TCO) Accounting: Evaluating AI ROI requires transparent value taxonomies and TCO modelling to account for non-deterministic model costs when scaling to enterprise user bases.
Webinar Details
Title: Staff, Cost, and Pace: Leading AI Delivery with the CPMAI Lifecycle for Data Managers
Date: 2026-11-05
Presenter: Howard Diesel
Meetup Group: African Data Management Community
Write-up Author: Howard Diesel
What is the CPMAI Framework for AI projects?
The Continuous Project Management for AI (CPMAI) framework builds directly upon established data mining standards to provide a structured methodology for AI initiatives. Organisations must establish foundational project management knowledge before launching complex machine learning models.
CPMAI extends the classic Cross-Industry Standard Process for Data Mining (CRISP-DM) six-stage lifecycle:
- Business Understanding
- Data Understanding
- Data Preparation
- Model Development
- Model Evaluation
- Model Operationalisation
Key certifications in the Data Science Body of Knowledge (DSBoK) include the Certified Analytics Professional (CAP) offered by INFORMS, alongside traditional project management credentials like the PMI PMP.
Key Takeaways
- CRISP-DM Heritage: CPMAI adopts and updates CRISP-DM’s classic six-phase iterative workflow for modern enterprise AI.
- Pre-Execution Preparation: Effective delivery requires completing readiness prep gates prior to defining initial business use cases.
- Professional Standards: Certified frameworks like INFORMS CAP and PMI CPMAI establish baseline competencies for AI delivery teams.
Frequently Asked Questions
- What is the relationship between CRISP-DM and CPMAI? CPMAI adopts CRISP-DM’s six core phases while adding upfront governance, platform readiness, and continuous operational feedback loops.
Figure 1 Aligning CPMAI and DAMA-DMBOK with Our AI Governance Series
Figure 2 www.informs.org
Figure 3 CRISP-DM Process Diagram
What are the Risks of Unchecked AI Data Ingestion?
Enterprise AI frameworks mandate implementing platform readiness checks and governance guardrails prior to initiating business use case execution. Establishing clear policies and risk controls protects organisations from severe data privacy failures.
Core governance guardrails include model cards, suitability and alternative assessments, solution classifications, and formal privacy addenda. These artefacts document model lineage, training validation records, and algorithmic rationale.
Unchecked AI data ingestion poses major privacy risks through secondary data linkage and person re-identification. For example, combining anonymised voting datasets with public census records in Dublin enabled total re-identification of individual voter choices.
Key Takeaways
- Front-Loaded Guardrails: Platform setup and regulatory policy definition must precede use case business scoping.
- Essential Artefacts: Model cards and suitability assessments provide mandatory documentation for compliance and model tracking.
- Re-Identification Risks: AI models can cross-reference disparate datasets to re-identify anonymised individuals, creating major liability.
Frequently Asked Questions
- Why must guardrails precede business understanding in CPMAI? Establishing security, privacy, and architectural controls upfront prevents teams from building non-compliant or unviable AI models.
- What is secondary data linkage in AI? It occurs when AI algorithms combine separate, individually anonymised datasets to re-identify specific individuals.
Figure 4 CPMAI Delivery Pipeline
Figure 5 One Deck, Five Sections
Figure 6 CPMAI and DAMA-DMBOK, in One Kit
How should we evaluate AI Initiatives effectively?
Evaluating AI initiatives requires connecting technical execution to an overarching enterprise architecture spine and rigorous financial modelling. Ethical assessments must take precedence over financial projections to prevent governance overrides.
The AI Architecture Spine aligns business outcomes and KPIs down through business processes, data value propositions, data products, and technical platforms.
Financial evaluation for non-deterministic Large Language Model (LLM) architectures must measure Total Cost of Ownership (TCO), Net Present Value (NPV), and operational cost shifts. Frameworks like Doug Laney’s Infonomics help structure value taxonomies for CFO approval.
Key Takeaways
- Ethics First: Ethical evaluations must occur before financial business case calculations to prevent high-value metrics from overriding safety.
- Architecture Lineage: The AI Architecture Spine links high-level business KPIs directly to underlying data products and infrastructure.
- Financial Rigour: Non-deterministic LLM systems require TCO and NPV accounting that tracks operational capacity shifts.
Frequently Asked Questions
- Why should ethical assessments occur before business value estimation? High financial promises can pressure executives to bypass ethical risks if monetary gains are calculated first.
Figure 7 Four Frameworks, One Delivery Method
How does AI Governance influence Use Case Decisions?
Effective AI governance relies on explicit control gates within CPMAI Lane 3 to continuously score, rank, and evaluate use cases throughout their lifecycle.
Governance committees must have the authority to kill, pivot, halt, pilot, or deploy AI use cases based on data supply quality and confidence scoring.
Pivoting involves stripping unstable features or critical data elements to establish a viable Minimum Viable Product (MVP). Decision flows must also be evaluated across end-to-end business processes (such as hire-to-retire) rather than within functional department silos.
Key Takeaways
- Five Governance Decisions: Executive gates must actively kill, pivot, halt, pilot, or deploy projects based on empirical evidence.
- Scope Reduction: Pivoting allows teams to salvage projects by removing unviable data elements to achieve a stable MVP.
- Cross-Functional Flows: AI decision gating must follow horizontal business processes rather than isolated organisational units.
Frequently Asked Questions
- How does a team pivot an AI use case under CPMAI? Teams pivot by removing unreliable data elements or complex capabilities to focus on a stable, high-confidence MVP.
Figure 8 The AI Architecture Spine: The Golden Thread
Figure 9 Prodago: Use-case Selection and Prioritisation
Figure 10 AI Evaluation: the BALANCE Control Plane
How does CPMAI align with Data Management frameworks?
Delivering sustainable enterprise AI requires mapping CPMAI project phases directly to established data management frameworks, including DMBoK, BizBOK, and BABOK.
The AI Delivery and Data Management Kit establishes four parallel execution lanes:
- Lane 1: Governance – Establishes charters, decision rights, risk tiers, and executive accountability.
- Lane 2: Foundations – Manages platform architecture, data domains, and metadata glossaries.
- Lane 3: Use Case Delivery – Executes phased project builds from scoping through deployment.
- Lane 4: Execution Rules – Enforces maturity-specific rules across Crawl, Walk, and Run stages.
Each CPMAI phase links to concrete DMBoK artefacts, including data quality profiles, security architectures, metadata glossaries, and named stewardship accountabilities across data domains.
Key Takeaways
- Framework Integration: CPMAI provides project delivery steps, while DMBoK supplies the underlying data management controls.
- Four Execution Lanes: Structured lanes separate governance, architecture foundations, use case execution, and maturity rules.
- Artefact Traceability: Every delivery stage links directly to specific quality reports, metadata glossaries, and security reviews.
Frequently Asked Questions
- What is the main difference between CPMAI and DMBoK? CPMAI governs AI project delivery workflows, whereas DMBoK provides the foundational data management and governance controls.
Figure 11 AI Delivery and Data Management Kit
Figure 12 DMBOK Controls Integration in CPMAI Delivery Phases
Figure 13 What We are Building
Figure 14 Reading the Kit
Figure 15 The Kit Architecture
Figure 16 DAMA-DMBOK in Each Lane
Figure 17 DAMA-DMBOK Chapters in Every CPMAI Phase
Figure 18 Governing Enterprise AI
How should organisations implement AI capabilities effectively?
Organisations should implement AI capabilities iteratively across maturity phases (Crawl, Walk, and Run) to manage competency limits and structural risks.
During the Crawl phase, teams strictly control single use cases using structured waterfall governance. Maturity scanning helps map opportunities across business capabilities; for example, a bank in Mauritius identified 500 use cases across core functions.
In agentic AI systems, small initial calibration errors can trigger severe downstream compounding errors along process flows, as demonstrated in automated emergency department triage systems.
Key Takeaways
- Maturity Phasing: Crawl stages require strict rules and single-use-case execution before scaling into parallel Agile runs.
- Opportunity Mapping: Capability heatmaps expose organisational blind spots and adoption resistance across business units.
- Compounding Risk: Autonomous agentic workflows amplify minor early decision errors into massive downstream operational failures.
Frequently Asked Questions
- What are compounding errors in agentic AI? They occur when a minor error made by an initial AI agent escalates across subsequent autonomous decision steps.
Figure 19 At a Glance
Figure 20 How the Playbook is Organised
Figure 21 Lane 3 of 4
Figure 22 One Lifecycle, Three Ways to Deliver it
What competencies are essential for managing AI delivery?
Managing AI delivery requires distinct project management competencies and specialised quality frameworks tailored for non-deterministic systems.
AI Project Managers focus on business alignment, decision rights, and governance oversight rather than hands-on code execution.
Evaluating AI quality demands going beyond basic data quality by building a “quality vault” across four dimensions: Business, Decision, Data, and Records quality.
Non-deterministic models require specialised evaluation metrics, such as Dale Rutherford’s BME ensemble metrics, to continuously track hallucinations, tonality, covariate drift, and response latency.
Key Takeaways
- Role Focus: AI Project Managers lead decision governance and stakeholder alignment rather than technical software coding.
- Four Quality Dimensions: Quality covers business effectiveness, decision accuracy, data fitness, and records integrity.
- Non-Deterministic Metrics: LLMs require ensemble metrics (BME) to monitor hallucinations, tonality shifts, and data drift.
Frequently Asked Questions
- How does non-deterministic AI quality differ from traditional data quality? Non-deterministic quality evaluates probabilistic outputs, hallucinations, and tonality using ensemble metrics rather than static rules.
How does AI delivery utilise structured tooling effectively?
Practical AI delivery relies on structured tooling to connect delivery gates with underlying data management controls.
The CPMAI Correlation Matrix in Excel consolidates multi-level mind maps, DMBoK chapter mappings, gate packs (G1–G6), and financial cost templates into a unified matrix.
Automated extraction workflows use NotebookLM and Model Context Protocol (MCP) server integrations to transform unstructured webinar transcripts into structured, multi-level visual mind maps adhering to Tony Buzan’s mapping principles.
Key Takeaways
- Integrated Tooling: Matrix toolkits bind delivery gate requirements directly to DMBoK control checklists and cost models.
- Automated Structuring: MCP integrations enable extraction of structured hierarchical mind maps directly from LLM contexts.
- Visual Synthesis: Hierarchical mind mapping provides clear visibility into multi-layered governance and architectural controls.
Frequently Asked Questions
- What is the CPMAI Correlation Matrix? An interactive toolkit that maps CPMAI project phases to DMBoK controls, gate evidence requirements, and financial templates.
Figure 23 What a Use-case Costs
How can we scale AI Project Management effectively?
Scaling AI project management across enterprise teams requires standardised research synthesis techniques and shared community knowledge packs.
Internal governance assets, including correlation templates, slide decks, and gate checklists, support structured learning when paired with mind mapping, flashcards, quizzes, and infographics.
Distributing open-access community packs via professional networks like LinkedIn enables data leaders to refine governance practices through real-world peer feedback.
Key Takeaways
- Structured Learning: Synthesising complex AI frameworks is optimised by pairing mind maps with quizzes and flashcards.
- Asset Standardisation: Centralised governance repositories ensure consistent execution across internal project managers.
- Community Feedback: Sharing governance toolkits publicly drives continuous refinement of data management standards.
Frequently Asked Questions
- How can data managers effectively digest complex AI frameworks? By combining hierarchical mind mapping with iterative self-assessment tools like flashcards, quizzes, and visual infographics.
- Key Takeaways
- What is the CPMAI Framework for AI projects?
- What are the Risks of Unchecked AI Data Ingestion?
- How should we evaluate AI Initiatives effectively?
- How does AI Governance influence Use Case Decisions?
- How does CPMAI align with Data Management frameworks?
- How should organisations implement AI capabilities effectively?
- What competencies are essential for managing AI delivery?
- How does AI delivery utilise structured tooling effectively?
- How can we scale AI Project Management effectively?