Generative AI & Programme Delivery

AI That Accelerates.
Humans That Lead.

Generative AI is reshaping how complex programmes are planned, delivered, and governed. This page sets out my practical understanding of GenAI — what it can do for project managers, how I apply it across the delivery lifecycle, and how I manage its limitations.

Understanding the Technology

What is Generative AI?

Generative AI learns patterns from vast datasets and produces original content — text, images, code, and data analysis — from a simple natural-language prompt. Unlike traditional AI that classifies existing data, GenAI creates new content grounded in contextual understanding.

For project and programme managers, this is transformative: complex deliverables that once took hours — charters, RAID logs, stakeholder communications, status reports — can now be produced as high-quality first drafts in minutes, leaving more capacity for the thinking and leadership that genuinely requires human judgement.

The key is using GenAI as an accelerator, not a replacement. The PM's role becomes one of directing, validating, and refining AI outputs — and knowing when not to rely on them.

The Distinction That Matters

Discriminative AI: "Is this project on track or not?"

Generative AI: "Draft a corrective action plan for a programme running 3 weeks behind schedule, with a fixed go-live date and an executive steering committee meeting next week."

8 Ways GenAI Helps Programme Managers

From initiation through close-out, GenAI reduces administrative overhead and sharpens delivery quality across every phase.

Time Saving

Status reports, RAID logs, meeting summaries, and project charters produced in minutes rather than hours — freeing capacity for stakeholder management and decision-making.

📢

Sharper Communication

Stakeholder emails, change-request memos, and board updates drafted and tailored to any audience, maintaining consistent tone across a complex multi-stakeholder programme.

🤖

Task Automation

WBS structures, sprint backlogs, milestone charts, and risk registers generated on demand — giving the team a solid baseline to refine rather than a blank page.

💡

Faster Ideation

Alternative approaches, edge-case scenarios, and risk identification surfaced rapidly — useful in programme recovery situations where speed of diagnosis matters.

📊

Better Decision Support

Project data consolidated and analysed faster than manual methods allow, enabling evidence-based decisions with stakeholders before positions have hardened.

⚠️

Risk Intelligence

RAID logs generated, likelihood and impact assessed, and mitigation strategies proposed across all 10 PMBOK knowledge areas — including risks that might otherwise be missed.

📈

Scalability

Complex, multi-workstream programmes handled without proportional increases in team overhead — particularly valuable in lean contract and consulting engagements.

🎯

Focused Human Effort

Administrative and documentation burden redirected to AI, so the PM's time goes where it counts: stakeholder trust, risk judgement, and commercial negotiation.

GenAI Across the Delivery Lifecycle

How I apply Generative AI at each stage of a programme — from initiation through close-out.

1

Initiating

Business case, project charter, stakeholder identification

AI Contribution

  • Generates first-draft project charters and business case summaries
  • Identifies potential stakeholders and suggests roles and responsibilities
  • Assesses feasibility by surfacing comparable programme patterns
  • Analyses charter completeness and flags gaps before sign-off
2

Planning

Scope, schedule, budget, risks, and the project management plan

AI Contribution

  • Builds WBS structures, milestone charts, and network diagrams
  • Generates RAID logs and proposes risk mitigation strategies
  • Drafts resource plans, RACI matrices, and communication strategies
  • Simulates planning scenarios to stress-test baseline assumptions
3

Executing

Team meetings, status reporting, change management, quality assurance

AI Contribution

  • Summarises meeting notes and extracts action items automatically
  • Drafts stakeholder updates tailored to executive or technical audiences
  • Generates sprint agendas and change-request documentation
  • Provides real-time decision support through instant data analysis
4

Monitoring & Controlling

Performance tracking, variance analysis, issue management

AI Contribution

  • Analyses programme data continuously to detect deviations early
  • Generates planned-vs-actual variance reports and trend analysis
  • Recommends corrective actions based on historical patterns
  • Provides predictive analytics on schedule and cost trajectories
5

Closing

Handover, lessons learned, contract closure, knowledge transfer

AI Contribution

  • Generates structured lessons-learned reports from project data and feedback
  • Drafts handover documents and operational team briefings
  • Produces close-out reports summarising outcomes against original objectives
  • Organises and archives project documentation for future reference

Integrating AI into a Large Programme

Successful AI integration in a complex programme requires more than selecting a tool. These are the principles I apply.

1

Define the Goals First

AI integration must be grounded in the programme charter. Define where AI adds value, align it to strategic KPIs, and include it explicitly in the requirements — not as an afterthought.

2

Assess Organisational Readiness

Infrastructure maturity, data availability, and cultural readiness all determine how quickly AI can be embedded. Rushing deployment ahead of readiness creates more risk than it removes.

3

Build the Right AI Team

Identify stakeholders who understand AI capabilities and assign clear ownership. AI integration without governance is a scope and quality risk like any other.

4

Set Realistic Expectations

Apply SMART criteria to all AI objectives. Stakeholders need to understand both what AI can do and what it cannot — unrealistic expectations are a leading cause of AI programme failure.

5

Include AI in the WBS

Embed AI tasks as formal work packages in the WBS and network diagram. Track AI integration milestones on the schedule — if it isn't tracked, it won't be delivered.

6

Verify Before Close-Out

AI deliverables must pass the same acceptance criteria as any other. Test outputs against requirements, gather stakeholder feedback, and don't close out until scope is fully verified.

Prompt Templates I Use

A selection of the prompts I apply in day-to-day programme work. Click any prompt to copy it.

Charter & Planning

Draft a project charter for [Project Name]. Include: goals, success criteria (SMART), scope in/out, assumptions, constraints, key milestones, roles, stakeholders, and a 1-page executive summary.

Click to copy ↗
Risk Management

Generate a RAID log for [Project Name] with 8 entries. Fields: ID, Type (Risk/Assumption/Issue/Dependency), Likelihood, Impact, Owner, Trigger, Mitigation, Status.

Click to copy ↗
Stakeholder Communication

Write a stakeholder update email to [CIO / Steering Committee] for [Project Name]. Include: current status, top 3 risks, decisions required from leadership, and the 2-week plan. Keep it under 300 words.

Click to copy ↗
Monitoring & Control

Generate a planned vs. actual progress report for [Project Name] up to [Date]. Include corrective action recommendations for any variances over 10%.

Click to copy ↗
Change Management

Prepare a Change Request summary for [Change Description]. Include: rationale, business value, scope/schedule/cost impact, risks, affected stakeholders, and a recommendation (Approve/Reject/Defer).

Click to copy ↗
Closure

Create a lessons-learned document for [Project Name]. Cover: what went well, what to improve, root causes of issues, and recommendations for future projects.

Click to copy ↗
Balanced Perspective

Limitations & Challenges

Effective use of GenAI requires clear-eyed awareness of where it falls short — and governance structures to manage that in a programme context.

⚠️

Inaccurate Output (Hallucination)

AI models can produce factually incorrect content — including hallucinated references and figures — which can undermine credibility if outputs aren't validated before use.

Mitigation: Cross-validate all numerical and factual claims. Define acceptance criteria for AI output before deployment. Never use AI output directly in a client deliverable without review. Confirm the LLM knowledge cutoff date.
🔒

Data Privacy & Security

Inputting sensitive programme data — commercial terms, personal data, IP — into public AI tools creates significant GDPR and confidentiality risks that are often underestimated.

Mitigation: Classify data before use. Use enterprise-grade tools (Claude Ai, ChatGPT) with data governance in place. Establish clear rules on what can and cannot be submitted to AI tools.
🎯

Limited Genuine Innovation

GenAI cannot generate truly novel ideas beyond its training data. It excels at synthesising and accelerating existing knowledge, not at the creative leaps that often define programme-level breakthroughs.

Mitigation: Use AI for ideation starting points and first drafts, not final outputs. Combine AI speed with human experience and domain knowledge to reach better decisions faster.
📋

Regulatory Uncertainty

The AI regulatory landscape — EU AI Act, sector-specific rules, evolving data protection law — is moving quickly and creates compliance risk for programmes that don't account for it.

Mitigation: Engage legal and compliance early. Build regulatory review into programme governance. Ensure AI recommendations are auditable and bias-reviewed before publication.
🔄

Rapid Technology Change

AI tools and capabilities are evolving at a pace that makes today's integration decisions potentially obsolete within months — creating technical debt risk in longer programmes.

Mitigation: Build AI model refresh cycles into programme governance. Monitor capability releases quarterly. Avoid deep dependencies on single-vendor AI products where possible.
🔗

Integration Complexity

Connecting AI tools to existing enterprise systems — ERP, PMIS, data warehouses — is non-trivial and requires the same architecture discipline applied to any major systems integration.

Mitigation: Map all system dependencies before AI integration begins. Use API-first design. Include AI integration as a formal workstream with its own RAID log and acceptance criteria.
Learning & Credentials

How I've Built This Knowledge

My GenAI capability has been developed through formal study alongside active application to programme delivery work.

🎓
IBM Generative AI for Project Managers Completed IBM's specialist certification covering GenAI fundamentals, project lifecycle application, programme integration, and responsible AI — assessed through graded coursework and applied exercises.
✍️
Advanced Prompt Engineering Completed advanced prompt engineering training, covering over a dozen prompt patterns that can be utilised to help resolve the most complex of problems that are identified in complex projects.
📡
Ongoing AI Training The speed of change with Generative AI is so fast, that if you are not maintaining ongoing AI study, you will be effectively going backwards. Hence ongoing AI related study is imperative.
⚙️
Active Application GenAI tools are embedded in my day-to-day work: predominantly Claude Ai for both work and personal tasks; Google Gemini / Microsoft Copilot / NotebookLM for graphics.

Prompt Engineering Patterns Matrix

A strategic reference guide for structuring large language model interactions to navigate complex technology delivery. Click any prompt block to copy it.

1. Context & Personas

01. Persona Pattern
Instructs the LLM to adopt a specific role or character, instantly inheriting specialized knowledge, tone, and professional constraints.
Act as an information security specialist. I will ask you different security-related questions and I need responses that meet industry best practice. Confirm the key source for each response.
Click to copy ↗
02. Audience Persona Pattern
Forces the LLM to tailor its explanation depth, vocabulary, and communication style to match a distinct target reader.
Explain large language models to me. Assume that I am a non-technical marketing executive with no background in computer science.
Click to copy ↗

2. Workflow & Interactive Control

03. Flipped Interaction Pattern
Reverses the dynamic: the LLM questions you iteratively until it extracts the exact parameters required to perform a task.
I would like you to ask me questions to help me diagnose a problem with my Internet. Ask me questions until you have enough information to identify the two most likely causes. Ask me one question at a time. Ask me the first question.
Click to copy ↗
04. Ask for Input Pattern
Strictly prevents generation until the user supplies explicit materials or datasets, preventing baseline drift or placeholder text.
From now on, I am going to cut/paste email chains into our conversation. You will summarise each person's points as sequential bullet points. At the end, list any open questions or action items addressed to me. My name is Branko. Ask me for the first email chain.
Click to copy ↗
05. Outline Expansion Pattern
Maintains high strategic control over long assets by having the model generate structures first, expanding segments one-by-one under your direct oversight.
Act as an outline expander. Generate a bullet point outline based on the input I give you, then ask me which bullet point to expand on. Create a new outline for the selected bullet point. At the end, ask me what to expand next. Ask me for what to outline.
Click to copy ↗
06. Game Play Pattern
Establishes interactive, rule-bound simulation spaces ideal for training scenarios, sandboxed testing, or immersive learning.
Create a cave exploration game for me to discover a lost language. Describe where I am in the cave and what I can do. I should discover new words and symbols for the lost civilisation in each area. Tell me about the first area and then ask me what action to take.
Click to copy ↗

3. Output Structure & Syntax

07. Template Pattern
Locks the LLM into a predefined output schema using explicit placeholders, ensuring machine-readable or consistently formatted responses every time.
Extract the data from this email using the template below. Do not alter the template structure. Template: Name: [Name] | Issue: [Issue] | Urgent: [Yes/No] Email: [paste email here]
Click to copy ↗
08. Meta Language Creation Pattern
Defines a compact custom shorthand or symbol system the LLM must learn and execute, dramatically compressing repetitive instruction overhead.
I want to create a custom shorthand for editing essays. Review these rules and reply only "READY". /FIX = Correct grammar and spelling only. /PUNCH = Rewrite to be shorter and more energetic. [A] = Rewrite for an academic audience. [B] = Rewrite for a casual blog post.
Click to copy ↗
09. Recipe Pattern
Supplies a goal and a partial sequence of known steps; the LLM identifies gaps and returns a complete, ordered action plan.
I would like to purchase a house. I know that I need to perform steps: make an offer and close on the house. Provide a complete sequence of steps for me. Fill in any missing steps.
Click to copy ↗
10. Menu Actions Pattern
Appends a persistent command menu to every response, allowing rapid single-keystroke follow-up actions without re-prompting from scratch.
Whenever I type "add FOOD", add FOOD to my grocery list and update my estimated bill. Whenever I type "remove FOOD", remove it and update the bill. Whenever I type "save", list cheaper alternatives. At the end, ask me for the next action. Ask me for the first action.
Click to copy ↗
11. Tail Generation Pattern
Anchors the output to a fixed, verbatim closing statement and instructs the LLM to reverse-engineer all preceding content to lead naturally to that endpoint.
Write a short persuasive argument. The document must end exactly with the following sentence — do not change a single word: "And that is why the most successful teams don't look for more hours in the day — they simply look for better focus."
Click to copy ↗

4. Refinement & Verification

12. Question Refinement Pattern
Instructs the LLM to surface a superior version of your query before answering, closing knowledge gaps you did not know existed.
From now on, whenever I ask a question, suggest a better version of the question and ask me if I would like to use it instead.
Click to copy ↗
13. Cognitive Verifier Pattern
Forces the LLM to decompose complex questions into targeted sub-questions, answer each independently, then synthesise a comprehensive final response.
When you are asked a question, follow these rules. Generate a number of additional questions that would help you more accurately answer the question. Combine the answers to the individual questions to produce the final answer to the overall question.
Click to copy ↗
14. Alternative Approaches Pattern
Prevents single-path thinking by requiring the LLM to generate multiple contrasting solution strategies with explicit trade-off analysis for each.
For anything that I ask you to write, determine the underlying problem I am trying to solve. List at least one alternative approach to solve the problem and compare/contrast it with the original approach implied by my request.
Click to copy ↗
15. Fact Check List Pattern
Compels the LLM to enumerate every verifiable claim embedded in its own output, creating a structured audit trail that accelerates human review.
Whenever you output text, generate a set of facts contained in the output. The set of facts should be inserted at the end of the output. The set of facts should be the fundamental facts that could undermine the veracity of the output if any of them are incorrect.
Click to copy ↗
16. Semantic Filter Pattern
Applies meaning-based redaction to a block of text — removing or retaining content by conceptual intent rather than keyword matching alone.
Filter this information to remove any personally identifying information or information that could potentially be used to re-identify the person. Rewrite the remaining text naturally so it retains its original readability.
Click to copy ↗
Copied to clipboard!