A business analyst's raw material is conversation: interviews, workshops, complaint threads, and the fifteen-tab spreadsheet someone built in 2019. AI assistants are unusually good with this material. Paste a transcript, ask for the requirements it implies, the assumptions nobody stated, and the questions to ask next, and the draft arrives before you could have opened the template.
What changes is the ratio of writing to thinking. First drafts of user stories, acceptance criteria, process narratives, RACI charts, and options memos come from your notes. The hours go to interviewing better, resolving conflicts between stakeholders, and deciding what to build, which was always the part that mattered.
The failure modes are specific. Models invent requirements that sound reasonable and were never said, quietly resolve ambiguities that should have gone back to the stakeholder, and draw confident process maps of systems they have never seen. Every artifact needs a line back to its source, and confidential financials and customer data stay out of tools your organization has not approved.
Quick wins this week
- Paste a stakeholder interview transcript and ask for stated requirements, inferred ones (labeled), unresolved ambiguities, and five follow-up questions, each tagged with the quote it came from.
- Turn a vague request ('the report needs to be faster') into a user story with Given/When/Then acceptance criteria, then read the criteria back to the requester and watch which ones they push back on.
- Describe a process from a walkthrough and ask for a numbered step list with actors, systems, decisions, and handoffs; paste it into Miro as the skeleton of your swimlane.
- Draft the meeting recap from your Otter or Fireflies transcript: decisions, open questions, owners, dates. Fix the two things it got wrong before you send it.
What AI can do for business analysts, task by task
Requirements elicitation and interview synthesis
Record with consent, then give the transcript and your project context and ask for stated requirements, inferred requirements (labeled), constraints, and contradictions between what different people said. Insist on a quote or timestamp for every item, then check that nothing was invented; models fill gaps with plausible needs no one voiced.
User stories and acceptance criteria
Paste the requirement, the persona, and your story template and ask for the story, Given/When/Then criteria, edge cases, and the questions to answer before refinement. Check that each criterion is observable by a tester and that the model did not close an open question by guessing a business rule.
Process mapping and current-state documentation
Describe the process as you observed it, exceptions and workarounds included, and ask for a step table with actor, system, input, output, and decision points, plus the pain points a reader would flag. Use it as the skeleton for a swimlane in Miro, then walk it with the people who do the work; the model does not know about the sticky note that says 'skip step 4 on Fridays'.
Gap analysis, options, and the business case
Give the current-state findings, target outcomes, and constraints and ask for a gap table, three options including 'do nothing', and the assumptions each depends on. Supply every number yourself; the model structures the case, it does not estimate your volumes or savings. Check each figure before it reaches a deck.
Data pulls, SQL, and spreadsheet analysis
Paste the table schema (column names and types, never the rows) and ask for the SQL that answers your question with a comment on each join. Copilot in Excel or Julius can profile a sheet, build pivots, and flag outliers. Run generated queries against a test environment and reconcile one total against a report you already trust.
UAT scenarios and traceability
From the stories and criteria, ask for UAT scenarios covering the happy path, each exception, permission boundaries, and the test data each needs, every scenario traced to its story. Reject scenarios that test the model's assumptions rather than your requirements, and send untestable criteria back to the story owner.
Prompts for business analysts
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Extract requirements from an interview transcript
You are a senior business analyst. Below is a transcript of a stakeholder interview for [PROJECT NAME]. Project goal: [ONE-SENTENCE GOAL]. Stakeholder role: [ROLE]. Produce four lists: (1) requirements stated explicitly, (2) requirements you infer but they did not state, labeled INFERRED, (3) constraints, policies, and systems mentioned, (4) ambiguities and contradictions. Cite the quote or timestamp for every item and add nothing without a citation. Finish with five follow-up questions, ordered by how much a wrong assumption would cost. Transcript: [PASTE THE TRANSCRIPT]
Tip: Anything without a citation is the model's idea, not the stakeholder's; delete it or verify it.
Write a user story with testable acceptance criteria
You are a business analyst on an agile team. Turn the requirement below into a user story in the standard 'As a, I want, so that' form, then write Given/When/Then acceptance criteria covering the main path, each exception, and permissions. Every criterion must be observable by a tester without reading code. List the business rules you had to assume, each phrased as a question for [STAKEHOLDER NAME OR ROLE]. Do not invent rules to make the story complete. Requirement: [PASTE THE REQUIREMENT] Context: [SYSTEM, USERS, AND KNOWN RULES]
Tip: Read the assumed-rules list first; each one is a decision somebody has to make.
Turn a walkthrough into a process map skeleton
You are documenting a business process for a current-state analysis. From my notes, produce a numbered step table with columns: step, actor (role), system, input, output, decision or exception. Mark handoffs between roles and estimate nothing I did not state. Then list the pain points, workarounds, and manual re-keying a reader should notice, and the questions you would ask the people who do this work. Process: [PROCESS NAME] Notes from the walkthrough: [PASTE YOUR NOTES]
Tip: Walk the map with the people who run the process; the model only knows what you wrote down.
Structure the options section of a business case
You are helping a business analyst structure a business case for [DECISION OR PROBLEM]. Current-state findings: [PASTE FINDINGS]. Target outcomes: [LIST OUTCOMES]. Constraints: [BUDGET, TIMELINE, SYSTEMS, POLICY]. Produce a gap table (current, target, gap, impact), then three options including 'do nothing', each with scope, benefits, risks, dependencies, and assumptions. Use only the figures I provided; where a number is missing, write NEEDED instead of estimating. End with the questions a CFO would ask that we cannot yet answer.
Tip: The 'do nothing' option keeps everyone honest about the cost of the status quo.
Generate UAT scenarios traced to stories
You are a QA-minded business analyst preparing user acceptance testing for [FEATURE OR RELEASE]. From the stories and acceptance criteria below, write scenarios with: ID, story it traces to, role, preconditions and test data, steps, expected result. Cover the happy path, every exception in the criteria, permission boundaries, and one negative case per story. Flag any criterion that cannot be tested as written and say why. Add no scenarios for behavior that is not in the stories. Stories and criteria: [PASTE STORIES AND ACCEPTANCE CRITERIA]
Tip: Untestable criteria are a requirements bug; send them back before UAT starts.
Want a prompt for something else? Use the Prompt Builder or browse finance and analysis prompts.
Skills to build
Interviewing for the transcript
Why: The model can only synthesize what was said, so your questions now bound the quality of every generated artifact.
How: Ask for examples, exceptions, and what happens when it goes wrong; restate what you heard on the recording; close each interview by listing the decisions still open.
Traceability discipline
Why: Generated requirements look finished, and a plausible line with no source is how a phantom requirement reaches development.
How: Require a quote, timestamp, or document reference on every item, keep a traceability matrix in Airtable or your requirements tool, and review anything labeled inferred with its owner.
Building context the model can use
Why: A prompt with the goal, stakeholders, constraints, and template gets a usable draft; a prompt with only the transcript gets a generic one.
How: Keep a one-page project brief (goal, scope, glossary, systems, decision log) and load it into a Claude Project, a custom GPT, or a NotebookLM notebook at the start of every session.
Reading generated SQL and formulas
Why: A query that runs and returns a plausible number is more dangerous than one that errors, and your analysis inherits its join logic.
How: Learn enough SQL to read joins, filters, and aggregations, ask the model to explain each clause, and reconcile one output against a trusted report.
Facilitation and decision-making
Why: AI drafts artifacts faster than stakeholders can agree on them, so the bottleneck moves to decisions, and that is human work.
How: Bring generated options to workshops as straw men, use the model's list of contradictions as the agenda, and log decisions where the model can reference them next time.
Tools worth knowing
Claude
A careful writing and analysis assistant that shines on long documents.
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
Microsoft Copilot
AI inside Word, Excel, Outlook, and Teams, with your company's data protections.
Miro
Shared whiteboard for workshops and planning, with AI to cluster, summarize, and draft
NotebookLM
A research notebook that only answers from the sources you give it, with citations.
Otter.ai
Meeting transcription and summaries that follow you into Zoom, Teams, and Meet.
Cautions for business analysts
Business cases, vendor pricing, and customer records are exactly what a consumer chat account may retain or use for training. Use your organization's approved plan with no-training or zero-data-retention terms, send schemas rather than rows, and never paste confidential data into a consumer AI tool unless your organization has approved it.
Models complete patterns: given eight requirements they will suggest a ninth that no one asked for, and given an ambiguous rule they will pick an interpretation and state it as fact. Label inferred items, keep a citation on every requirement, and send ambiguities back to the stakeholder instead of accepting the guess.
Recording and transcribing stakeholders requires their consent, and some states require all-party consent; announce the recorder and honor requests to pause. Documents you feed the model (vendor RFP responses, emailed requirements, exported tickets) can contain text written to steer its output; read summaries of third-party material with that in mind and never let an automated pipeline act on them unreviewed.
Ask for a business case and the model will estimate volumes, savings, and adoption rates that look researched and are not. Supply your own figures, mark every estimate with its source, and treat a generated process map as a hypothesis until the people who run the process have corrected it.
Your 30-day plan
- Week 1: Confirm which assistant your organization allows and its data terms, then write a one-page project brief (goal, scope, glossary, systems, stakeholders) to load into every session.
- Week 1: Use AI only on your own notes and on transcripts you have consent to process; ask for requirements with citations and count how many inferred items turn out wrong.
- Week 2: Draft every user story and acceptance criterion with the assistant first, refine them in the team's normal ceremony, and note what reviewers changed.
- Week 3: Build one process map skeleton from a walkthrough and validate it with the people who run the process; try one SQL or Excel analysis with the schema-only rule.
- Week 4: Generate the options section of a live business case with your own numbers and UAT scenarios for a story in flight, then agree with the other analysts on how AI-drafted artifacts are labeled and traced.
Frequently asked questions
Will AI replace business analysts?
Can I paste stakeholder interviews into ChatGPT?
How do I stop AI from making up requirements?
Do I need to learn SQL if AI can write it?
Terms used on this page
Related roles
- AI for Product ManagersAI turns forty interview transcripts into themes, a rough idea into a PRD draft, and a napkin sketch into a clickable prototype, and it will also confidently invent a customer need that does not exist. Here is how product managers get the first without the second.
- AI for Project ManagersAI cannot run your project, but it can draft the status report, turn a messy meeting into an action list, and pressure-test your risk register in minutes. Here is how project managers use it without losing the plot.
- AI for Data AnalystsAI writes the SQL, the pandas, the DAX, and the sentences around the chart. It cannot know your data, so your job shifts to the right question, the checked join, and keeping regulated data out of the wrong tool.
- AI for ConsultantsAI synthesizes twelve interview transcripts into themes with quotes, builds the issue tree, argues against your recommendation before the client does, and drafts the SOW. The judgment, the relationship, and the confidentiality obligations are yours.