What AI Already Does of a BA's Job
A breakdown of roughly 50 business-analysis techniques from Karl Wiegers (Software Requirements) and BABOK v3: for each technique, an honest look at how much an AI assistant working with code already covers, and what still falls to the human. Not hype - an inventory: where you're already losing on speed if you do it by hand, and where nobody can replace you yet.
~58%
of formal BA techniques are already meaningfully automated (the "Full" plus "High" levels). Another 28% get substantial help. Purely human work remains for about 10%.
18%
40%
28%
Full80-100% AI does it on its own · 9 techniques (18%)
High60-80% AI carries the bulk, BA steers · 20 techniques (40%)
Medium30-60% AI helps noticeably, but the BA is in charge · 14 techniques (28%)
Low10-30% marginal assistance, mostly manual work · 4 techniques (8%)
None0-10% purely human work · 1 technique (2%)
The rating assumes working through an AI assistant with access to the code (Claude Code and similar tools). What follows: six blocks of techniques from BABOK/Wiegers.
1. Gathering Requirements
Requirements Elicitation
| Technique | Level | What AI does | What's left to the human |
|---|---|---|---|
| Interviews | Low | Generates interview scripts and domain-specific question lists, transcribes and summarizes notes | Conducting the interview itself, reading the interviewee's reactions, live follow-up questions |
| Workshops / JAD | Low | Prepares the agenda, pre-analyzes materials, summarizes outcomes, drafts follow-ups | Facilitation, group dynamics, conflict resolution, on-the-fly synthesis |
| Focus Groups | Low | Prepares the discussion guide, analyzes transcripts | Running the session, observing reactions |
| Observation | None | - | Physical observation of users in their environment |
| Questionnaires / Surveys | Medium | Drafts questions, analyzes results, finds patterns in responses, suggests refinements | Distribution, audience selection, interpreting context |
| Document Analysis | High | Reads and summarizes existing documents (Confluence, specs, policies), extracts requirements, cross-checks them against each other | Deciding which documents matter, validating extracted content against intent |
| Interface Analysis | High | Analyzes API contracts, system interfaces, and data formats straight from the code, finds integration points | Understanding the business context of interfaces |
| Prototyping | High | Generates HTML/CSS prototypes from a description, iterates quickly | Validation with real users, UX decisions |
| Brainstorming | Medium | Generates ideas, develops concepts, plays devil's advocate, finds gaps | Creative leaps, domain intuition, team energy |
| Reverse Engineering | Full | Reads existing code and extracts business rules, data flows, logic; reconstructs undocumented behavior | Verifying that the extracted rules match business intent |
2. Analyzing Requirements
Requirements Analysis
| Technique | Level | What AI does | What's left to the human |
|---|---|---|---|
| Use Cases & Scenarios | High | Generates scenario descriptions, alternate and exception flows, pre- and post-conditions from requirements | Checking completeness, business alignment |
| User Stories | High | Generates stories with acceptance criteria from epics and features, applies INVEST, adds edge cases | Prioritization, business-value assessment |
| Business Rules Analysis | High | Extracts rules from code, documents, policies; formalizes them into decision tables | Verifying the rules reflect actual business policy |
| Data Modeling | High | Generates ER diagrams, analyzes existing database schemas, suggests normalization | Domain decisions about relationships |
| Data Flow Diagrams | High | Generates DFDs from code analysis, traces data across systems | Validation against actual business processes |
| State Modeling | Full | Generates state diagrams from code or requirements, finds missing states | Checking edge cases with experts |
| Process Modeling (BPMN) | High | Generates BPMN from a description, analyzes existing processes | Validation against reality, optimization decisions |
| Decision Modeling (DMN) | High | Creates decision tables, checks completeness, finds contradictions | Validating the business logic |
| Concept Modeling | Medium | Generates a glossary, extracts domain concepts from documents and code | Semantic precision of the domain language |
| Functional Decomposition | High | Breaks features down into sub-functions, ties them to components in the code | Deciding where decomposition boundaries lie |
| Non-Functional Reqs | Medium | Surfaces NFRs from the code (performance patterns, security), suggests standards | Setting acceptable thresholds, business trade-offs |
| Acceptance Criteria | Full | Generates Given/When/Then, edge cases, checks completeness | Final business validation |
| Scope Modeling | Medium | Generates context diagrams, identifies system boundaries from the code | Deciding what's in scope and what isn't (policy and strategy) |
3. Specifying Requirements
Requirements Specification
| Technique | Level | What AI does | What's left to the human |
|---|---|---|---|
| SRS Document | High | Generates a structured SRS, keeps it IEEE 830-compliant, stitches sections together | Review, business sign-off |
| Data Dictionary | Full | Extracts from database schemas, code, APIs; automatically assembles complete data dictionaries | Validating the business descriptions |
| Glossary | High | Extracts domain terms from documents and code, generates definitions, finds discrepancies | Aligning on canonical definitions with stakeholders |
| Business Rules Catalog | High | Extracts and catalogs rules from code and documents, catches duplicates and conflicts | Checking against actual business policy |
| Requirements Attributes | Full | Auto-fills attributes (priority hints, complexity, source traceability) in Jira | Priority and business value are decided by the human |
| Sequence Diagrams | Full | Generates from code with full accuracy: reads actual method calls and interactions | Validation against intended behavior |
4. Validating Requirements
Requirements Validation
| Technique | Level | What AI does | What's left to the human |
|---|---|---|---|
| Reviews / Inspections | High | Checks for ambiguity, incompleteness, contradictions, testability; applies the DoR checklist automatically | Final approval, contextual and political judgment |
| Prototyping for Validation | High | Quickly generates clickable prototypes for validation with stakeholders | Running the validation session itself |
| Test Case Generation | Full | Generates test cases from acceptance criteria, boundary analysis, equivalence-class partitioning | Test-strategy decisions |
| Traceability Matrix (RTM) | Full | Builds and maintains the RTM across code, Jira, and tests; finds gaps automatically | Deciding on traceability granularity |
| DoR / DoD Checklists | Full | Checks every story against the checklist automatically, flags issues | Defining what goes into the checklist itself |
| Baseline Comparison | High | Diffs requirements between versions, surfaces scope changes | Deciding whether the changes are acceptable |
5. Managing Requirements
Requirements Management
| Technique | Level | What AI does | What's left to the human |
|---|---|---|---|
| Change Impact Analysis | High | Traces impact through the code, finds affected components, estimates blast radius | Assessing business impact, go/no-go decisions |
| Version Control | Full | Tracks all changes, diffs, history - native to the IDE | - |
| Traceability | Full | Automatically links requirements, code, tests, and Jira together | Initial setup of the links |
| Prioritization (MoSCoW, RICE) | Medium | Applies the framework structure, scores and ranks items; flags dependencies | Business value and stakeholder input |
| Risk Analysis | Medium | Identifies technical risks from the code, suggests mitigations, flags dependency risks | Assessing business risks and their likelihood |
| Estimation | Medium | Analyzes code complexity and historical velocity; suggests story points based on similar past stories | Team capacity, hidden complexity, the human factor |
6. Strategy and Solution Evaluation
BABOK: Strategy & Solution Evaluation
| Technique | Level | What AI does | What's left to the human |
|---|---|---|---|
| Stakeholder Analysis | Medium | Generates RACI templates, builds personas from org data | Understanding real influence and internal politics |
| SWOT Analysis | Medium | Generates the analysis from the provided context, researches competitors | Strategic judgment |
| Business Model Canvas | Medium | Generates the canvas from inputs, flags gaps | Strategic vision |
| Business Case | Medium | Structures the financial analysis, runs ROI template calculations | Real financial data, strategic decisions |
| Balanced Scorecard | Low | Template generation, KPI suggestions | Strategic alignment, goal-setting |
| Benchmarking | Medium | Researches and gathers benchmarks, analyzes competitor data | Judging relevance to your context |
| Vendor Assessment | Medium | Compares vendor documents, builds evaluation matrices | Relationships, trust, strategic fit |
| Root Cause Analysis | High | Analyzes the code for bugs, traces errors, builds Ishikawa diagrams from logs and data | Organizational root causes |
| Lessons Learned | Medium | Summarizes sprint and project data, surfaces patterns from Jira | Facilitating the team's retro |
| Metrics and KPIs | High | Calculates from Jira and code data, builds dashboards, trend analysis | Deciding which metrics matter |
| Solution Evaluation | Medium | Compares the actual code against requirements, runs gap analysis | Assessing business value after delivery |
The key takeaway for BAs
AI is best at what's textual and structural (generating, analyzing, checking documents), tied to code (anything that benefits from reading the actual implementation), and patterned (well-known frameworks: INVEST, Given/When/Then, decision tables).
AI doesn't replace what needs live interaction (facilitation, interviews, observation), political and strategic judgment (prioritization, stakeholder management, scope boundaries), and creative domain understanding (knowing why the business works the way it does).
The BA's role is shifting from "producing artifacts" to "validating AI-produced artifacts against business reality." The value now isn't in writing the SRS or the diagram - it's in catching where AI got something plausible but wrong.
That's exactly what AI Sandbox teaches: building specs and MVPs through an AI assistant and checking the result, instead of rewriting everything by hand. Join the new group.