

In my BA work, the Business Requirements Document is one of the most-produced and most-debated artefacts. I’ve watched stakeholders argue about its length, format, and necessity in every organisation I’ve worked with. Most BRDs I see are too long for executives, too short for engineers, and too generic for either. AI has dramatically changed how I produce BRDs. The right prompts and a disciplined review process turn BRD writing from a multi-day chore into a 2-3 hour exercise that, in my experience, produces sharper output.
A working BRD has these sections:
Length: 15-30 pages for medium projects, 30-60 for large. Anything longer is unread.
A working workflow:
Total time from synthesis to final BRD: 2-3 days vs 1-2 weeks pre-AI.
A master prompt that consistently produces strong BRDs:
“You are a senior business analyst at [industry] company. Below is the synthesised research from 18 stakeholder interviews and supporting documents. Generate a BRD with these sections: Executive Summary, Business Context, Scope (In/Out), Stakeholders, Functional Requirements, Non-Functional Requirements (cover security, performance, scalability, accessibility, compliance), Assumptions, Constraints, Dependencies, Acceptance Criteria, Glossary. Tone: precise, no marketing language. Maintain traceability comments noting which interviews support each requirement. Length: 4,000 words. Flag sections where the input is insufficient and propose what additional research is needed.”
The BA edits 25-40% of the output and adds judgement on strategic sections.
Executive Summary: 1 page, lead with business impact. AI tends to write generic summaries. Edit aggressively for organisational voice.
Business Context: AI captures stated context well. Add political and historical context only humans know.
Scope: AI generates clear in/out lists. Validate boundaries with stakeholders.
Stakeholders: AI extracts mentioned stakeholders. Add political dynamics manually.
Functional Requirements: AI generates from synthesis. Validate completeness against stated needs.
Non-Functional Requirements: AI tends to generate generic NFRs. Push for specific numbers (e.g., “response time under 2 seconds for 95% of requests” not “system should be fast”).
Assumptions: AI lists obvious assumptions. Add organisational assumptions manually.
Constraints: AI captures stated constraints. Add unstated organisational constraints.
Dependencies: AI infers from synthesis. Validate with technical and operational teams.
Acceptance Criteria: AI generates from requirements. Make sure each AC is testable.
Different audiences need different framings of the same BRD:
A useful tailoring prompt:
“Take this BRD and rewrite the executive summary for a C-level audience. Lead with business outcomes. Length: 1 page. Tone: confident, no jargon.”
Before finalising:
A 30-minute pass through this checklist catches most issues before stakeholder review.
Traceability links requirements to:
AI tools (Jama, Modern Requirements, IBM DOORS Next) automate much of this. For BAs without dedicated tools, a simple traceability matrix in Excel or Notion suffices.
| Tool | Strength |
| Confluence AI | Strong if Atlassian-stack |
| Notion AI | Lightweight, flexible |
| Microsoft 365 Copilot | Native to Word and SharePoint |
| Jama Connect | Dedicated requirements management |
| General LLM (Claude, ChatGPT) | Most flexible |
For most BAs, a general LLM plus a documentation platform is sufficient.
These are the patterns I see most often when BAs lean on AI for BRD work. I’d flag the first two as the ones that have done the most damage to teams I’ve coached.
Modern BAs increasingly maintain BRDs as living documents:
This pattern eliminates the “stale BRD” problem that plagues classical BA work.
Sign-off rituals matter more than the document itself:
AI helps draft sign-off summaries, track who has reviewed, and surface remaining open questions.
Logan Hutchinson has 25+ years of experience leading AI innovation at Cruise, Motorola, Siemens, and Drift, building Level 5 autonomous systems, enterprise AI platforms, and breakthrough healthcare automation products at scale.
QUICK FACTS
After review and editing, yes. Disclose AI use.