Bonus: BA in the Age of AI
BA in the Age of AI
Artificial intelligence is not arriving in the business analyst’s world — it is already here. In 2026, the question is no longer whether AI will change how BAs work, but whether you will be the professional who harnesses it deliberately or the one who scrambles to catch up.
This module is immediately practical. You will leave each topic with tools you can open on Monday morning: prompt templates, governance frameworks, requirements templates, and a clear picture of how to position your AI literacy in the ANZ market.
1. AI Reshaping the BA Role
2. Prompting for Requirements
3. AI-Assisted Process Modelling
4. Writing Requirements for AI Systems
5. BA as AI Governance Facilitator
6. Your ANZ AI Toolkit
How AI Is Reshaping the BA Role — Not Replacing It
The 3 Things AI Cannot Replace
AI compresses the low-value production work so the BA can spend more time on what only a human can do. These three capabilities determine whether a BA project succeeds or fails — and AI cannot replicate any of them.
Stakeholder relationships
Requirements do not live in documents — they live in people. The procurement manager with a silent objection. The IT architect who will not say the integration is difficult until you have built trust privately. The executive sponsor who needs re-enrolling every three weeks. AI cannot have coffee with these people, read the room, or navigate unspoken organisational dynamics. This relational intelligence is not a soft skill — it is the core skill.
Political navigation
Most BA work happens inside organisations with competing priorities, historical tensions, and conflicting incentives. AI can generate a stakeholder impact assessment template. It cannot tell you that the Head of Operations is only supporting this project because her rival championed the last one — and that you need to frame your recommendations to give her visible ownership.
Judgment under ambiguity
BAs routinely encounter situations where information is incomplete, stakeholders disagree, and a decision must be made anyway. AI operates on patterns from training data. It is not equipped to make judgment calls in novel situations where the stakes are organisational, relational, or ethical.
New BA Responsibilities Created by AI
AI Output Validation
When generative AI produces a document, someone has to check it. BAs are increasingly taking on this validation role because they have the domain knowledge and quality mindset to do it properly.
Prompt Governance
Organisations deploying AI are discovering wildly inconsistent outputs from the same tool. BAs are beginning to own approved prompt libraries — standardised prompts reviewed for accuracy and appropriate use.
Requirements for AI Systems
Writing requirements for a feature that uses a machine learning model is fundamentally different from writing requirements for traditional software. BAs who understand this difference are becoming highly sought-after.
Key insight: The BAs who will thrive are not the ones who know the most about AI technology. They are the ones who bring human judgment, stakeholder insight, and organisational context to AI-assisted analysis. That combination is rare and valuable.
✎ Practice Exercise
The AI Audit
Take your most recent BA assignment. List every task in the requirements phase. For each task: (1) Could AI have produced a useful first draft? (2) What did you add — judgment, context, relationships — that AI could not? Write 2–3 sentences for each. You are mapping where AI fits in your workflow and where your irreplaceable value lies.
Prompting for Requirements — AI as Your First Draft
4-Part Anatomy of a Good BA Prompt
Most BAs who try AI and find it unhelpful have made the same mistake: they treat the AI like a search engine rather than a well-read junior analyst who needs context and a clear brief. The quality of your output is almost entirely determined by the quality of your input.
Role framing
Tell the AI what role it is playing. ‘You are a senior business analyst…’ sets the register, vocabulary, and level of sophistication.
Context
Give the AI the project context it needs. Industry, organisation type, system in scope, primary users. The more specific, the better the output.
Task instruction
Be precise. Not ‘write some user stories’ but ‘write six user stories in the format As a [role] I want [feature] so that [benefit], for the payment reconciliation process described below.’
Constraints and format
Specify what you want excluded, the output format, and quality criteria. ‘Each story must include 3–5 acceptance criteria in Given/When/Then format. Avoid technical implementation details.’
5 Ready-to-Use BA Prompts
Prompt 1: User Story Generation
You are a senior business analyst working on a [industry] project for [organisation type]. The system in scope is [system], which will be used by [primary user roles]. Based on the following process description, write eight user stories: As a [role], I want [capability], so that [benefit]. For each story, include four acceptance criteria in Given/When/Then format. No technical implementation details. Process description: [paste notes here].
Prompt 2: Stakeholder Email Draft
You are a BA drafting a professional email to [stakeholder role] at [organisation type]. Purpose: [specific purpose]. Key points: [3–5 bullets]. Tone: professional and direct, not overly formal. Length: under 200 words. Do not use filler phrases. Start with the purpose immediately.
Prompt 3: Gap Analysis Starter
You are a senior BA conducting a gap analysis. Current state: [3–5 sentences]. Target state: [3–5 sentences]. Identify at least six specific gaps where the current state does not meet the target state. For each gap, suggest one approach to closing it. Present as a table: Gap ID | Gap Description | Current State | Target State | Suggested Approach.
Prompt 4: Process Description to BPMN Narrative
You are a BA translating a process description into a structured BPMN narrative. Convert this into a structured description identifying: swim lanes (participants), start event, each task in sequence, decision gateways with conditions, and end events. Label each element with a BPMN element type in brackets. Process: [paste notes here].
Prompt 5: Meeting Notes to Action Items
You are a BA reviewing raw meeting notes. Extract all action items, decisions, and open questions. Format as three sections: (1) Action Items — owner, description, due date; (2) Decisions Made — one-sentence statement per decision; (3) Open Questions — not resolved, with name of person who raised them. Meeting notes: [paste here].
Validation Checklist — Before Any Output Goes into a Deliverable
Factual accuracy — has the AI invented system names, process steps, or role titles not in your source?
Completeness — did the AI quietly drop the messy or ambiguous parts of your requirements?
Stakeholder appropriateness — is the language right for this specific audience, not a generic one?
Organisational context — does the output reflect how your organisation actually works?
Testability — can a tester actually use these acceptance criteria to verify the system?
✎ Practice Exercise
The 30-Minute Sprint
Choose a real or simulated BA scenario. Open ChatGPT, Claude, or Copilot. Use Prompt 1 to generate user stories. Then apply the validation checklist. Note how long generation took vs validation. Write down what the AI got right, missed, and invented. This is the exercise that builds accurate intuition for where AI is genuinely useful.
AI-Assisted Process Modelling and Analysis
From Interview to Draft Model — Faster
Instead of going from interview → blank page → draft model, you go from interview → AI-generated draft → refined model. The blank page problem disappears. Structured AI output also gives stakeholders something specific to react to — and people are much better at saying “that’s wrong, here’s what actually happens” than constructing an accurate process description from scratch.
The Interview-to-Draft Workflow
Most common AI failure in process modelling
AI models tend to capture the happy path well and miss the exception paths — the workarounds, the informal steps, the “what happens when X goes wrong.” Always ask your stakeholder: “What happens when X fails?” for each major step. Validate the AI draft with the people who actually do the work, not just the people who designed it.
Tools in 2026
Microsoft Copilot in Visio
Natural language input to create and modify diagram elements within the Visio environment. Best for organisations already on Microsoft 365. Works well for simple processes; complex exception paths still require manual work.
ChatGPT / Claude + draw.io
Most flexible approach for BAs not using Visio. Generate the process narrative in ChatGPT or Claude, then build the diagram in draw.io using the narrative as your specification. draw.io is free, widely used in ANZ, and exports to multiple formats.
Claude for process narrative
Particularly strong for readable, well-structured process descriptions from raw input. Claude’s output tends to be more naturally written than competing models, making it easier to share with non-technical stakeholders as a process summary.
✎ Practice Exercise
The Interview-to-Draft Workflow
Find a process you know well. Write a 200–300 word description including at least one exception path. Paste into Claude or ChatGPT with the BPMN narrative prompt. Review the output and mark: (a) what AI got right, (b) what it missed, (c) what it invented. Then build a simple diagram in draw.io from the corrected narrative. Estimate time vs building from scratch.
Writing Requirements for AI Systems
The AI Feature Requirements Pack — 11-Point Template
Writing requirements for traditional software assumes deterministic systems: given input A, the system produces output B — every time. AI systems produce probabilistic outputs. BAs who have not updated their skills are producing requirements that are either untestable or that miss the most important quality concerns. Use this template as a supplement to your standard requirements documentation whenever any AI or ML component is involved.
Feature Name and Purpose
Plain-language description of what the AI feature does and the business problem it solves
AI Approach
Classification, regression, NLP, recommendation, anomaly detection. The dev team specifies the model — the BA specifies what it must achieve
Input Data Requirements
What data the model receives: format, source, refresh frequency, pre-processing requirements
Training Data Requirements
Volume, recency, labelling, data sources, consent and provenance requirements, exclusions
Performance Requirements
Minimum acceptable accuracy/precision/recall. What constitutes ‘good enough’ to go live. What triggers remediation
Confidence Thresholds
What the system does when confidence is below a defined threshold: escalate to human, return null, flag for audit?
Bias and Fairness Requirements
Which demographic or protected groups must the system perform equivalently for? Testing methodology and reviewer
Explainability Requirements
Must the system explain its outputs? To whom? In what format? Appropriate level of detail per audience?
Human Override Requirements
Can a human override the AI output? Under what circumstances? Is the override logged?
Monitoring and Drift Requirements
How will model performance be monitored post-deployment? What triggers retraining? Who is responsible?
Regulatory Compliance
Applicable privacy, anti-discrimination, and sector-specific regulatory requirements (NZ Privacy Act 2020, AU Privacy Act 1988)
Writing Testable Acceptance Criteria for AI Features
✗ Deterministic AC (unusable for AI)
“Given a document is uploaded, when the system classifies it, then the classification must be correct.”
✓ Probabilistic AC (testable for AI)
“Given a test set of 500 documents with known classifications, when the model processes all 500, then overall accuracy must be ≥92%, with no protected category achieving accuracy below 88%, measured using the methodology in [linked test plan].”
ANZ Example: Document Classification in Government
A NZ local government agency classifying inbound correspondence (applications, complaints, OIA requests, general enquiries). Critical requirements: ≥95% recall on OIA requests (missing one is a compliance failure), flag any item below 80% confidence for human review, provide audit trail of every classification decision, and bias testing to ensure no performance difference for te reo Māori correspondence or non-native English writers. These requirements emerge from organisational and regulatory context — exactly the kind of contextual judgment a skilled BA brings.
✎ Practice Exercise
Rewrite a Requirement for AI
Take a traditional system requirement — ‘The system shall display the correct account balance’ or ‘The system shall route the application to the correct approver.’ Imagine this function is now performed by an AI component. Rewrite using the 11-point template above. What new questions must you ask? What performance thresholds would you specify? What happens when the AI is wrong?
The BA as AI Governance Facilitator
7-Risk Governance Framework for AI Initiatives
AI governance is one of the most talked-about topics in ANZ boardrooms in 2026 and one of the least well-understood at the operational level. The BA is the natural person to bridge this gap — stakeholder mapping, risk surfacing, requirements capture, facilitating between technical, legal, and business teams. This is business analysis work.
Data Privacy Risk
What personal information is processed? What third parties receive it? Is processing compliant with applicable privacy legislation?
Decision Accuracy Risk
What is the cost of a wrong output? What is the expected error rate? Is the error rate acceptable for this use case?
Bias and Discrimination Risk
Does the AI perform differently for different groups of people? What are the consequences if it does?
Accountability Risk
Is it clear who is responsible when the AI causes harm? Does the AI create a ‘diffusion of responsibility’ problem?
Dependency Risk
What happens if the AI vendor changes terms, increases pricing, or goes out of business?
Reputational Risk
If the AI makes a visible public error or media scrutiny arrives — what is the reputational exposure?
Regulatory Risk
Compliant with current regulation? How might the EU AI Act (influencing ANZ regulatory thinking) or upcoming NZ/AU AI regulation affect it?
Reframing Ethical Concerns — The Language That Works
Instead of (objection)
“I am concerned this AI could be biased.”
Try instead (requirements task)
“To complete the requirements, I need to document how we will test for bias and what our acceptable performance threshold is across different groups. Can we schedule 30 minutes with the data team?”
This reframe positions the concern as a requirements task, proposes a specific next step, and makes clear the governance question must be answered before the feature is complete — not as a separate, optional exercise.
✎ Practice Exercise
The AI Governance Audit
Think of an AI tool currently used — or being considered — in your workplace. Using the 7-risk framework above, write 2–3 sentences assessing the risk level and what mitigation or requirement would address it. Then draft 3 questions, using the reframing approach, that you would bring to the project sponsor or IT lead.
Your ANZ AI Toolkit
6 Tools ANZ BAs Are Using in 2026
Market signal (early 2026): ANZ senior BA contracting day rates for BAs with demonstrable AI governance or AI requirements experience are running 15–25% higher than equivalent roles without that specification. This premium is expected to persist for 2–3 years as organisations build out AI-enabled capabilities.
You do not need all six tools. You need one or two that you use consistently. A BA who uses ChatGPT every day for first drafts will outperform one with accounts across every AI platform but no systematic habits.
Positioning AI Literacy on Your Resume and LinkedIn
Generic claims carry little weight. Be specific about what you have done:
✗ Generic (ignored)
“Experienced with AI tools”
✓ Specific (differentiating)
“AI-assisted requirements development (ChatGPT, Microsoft Copilot, Claude); AI requirements specification; AI governance facilitation. Reduced first-draft requirements cycle time by 40% through systematic AI-assisted artefact generation.”
✎ Practice Exercise
Your Personal AI Workflow Audit
Return to the task list from Topic 1. For each task where AI could help, choose one tool and write one sentence describing how you would use it. Then identify three specific tasks in your current or upcoming work where you will use an AI tool this week. Be specific: ‘I will use the meeting notes to action items prompt in Claude after our Thursday workshop’ creates a workflow. ‘I will use AI more’ changes nothing.
Key Takeaways
- ✓
AI compresses low-value production work — first-draft user stories, BRD sections, meeting summaries — so you can spend more time on stakeholder relationships, political navigation, and judgment under ambiguity: the three things AI cannot replace
- ✓
The quality of AI output is almost entirely determined by the quality of your prompt — use the 4-part anatomy: role framing, context, precise task instruction, constraints and format
- ✓
Writing requirements for AI systems requires a different approach: probabilistic performance thresholds, confidence score handling, bias testing, explainability requirements, and model drift monitoring
- ✓
ANZ BAs working on AI-enabled systems must address explainability under the Privacy Act 2020 (NZ) and Australian Privacy Act 1988 — this is a requirement, not an optional good practice
- ✓
The CBBA + AI literacy combination is the most differentiated positioning available in the ANZ BA market right now — most credentialed BAs are still developing AI fluency, and most AI-literate professionals lack BA methodology depth
Ready to go further?
Earn the Certified Better Business Analyst (CBBA)
Go from foundational knowledge to a globally-recognised BA certification. 6 weeks, 100% online, built around a real-world case study — not just theory.
- ✓ 6 weeks of structured, practical content
- ✓ Real case studies — Innovate Bank & EcoEnergy
- ✓ Globally-recognised certificate & digital badge
- ✓ 12 months access, retake exam included
- ✓ Comparable courses cost USD $3,000+ — ours is $349 NZD
One-time payment. Instant access. 14-day money-back guarantee.