Why generic AI chat struggles here
Ask ChatGPT or Claude directly for "a BRD" and you'll get a plausible-looking document with the wrong numbering scheme, requirements that mix business and solution language, and details it quietly invented because you didn't give them. A BA then has to catch every one of those before it goes to a stakeholder — which is often slower than starting from a blank page.
What it does
- Asks before drafting. Confirms whether you need a BRD or FRD, the project name and sponsor, and scope boundaries — only the questions that would block the draft.
- Never invents. Stakeholders, dates, systems or figures not in your input go into Assumptions or are flagged
[TO CONFIRM: …]. - Numbers hierarchically. BR-01, FR-01.1 — one requirement per line, atomic, testable, no compound "and/or" requirements.
- 16-section structure. Document control, executive summary, background, problem statement, objectives, scope, stakeholders, business and functional requirements, non-functional requirements, data requirements, business rules, assumptions, risks, open questions, and a glossary.
- Quality pass included. Checks every requirement is testable, uniquely numbered, and traceable to an objective before returning it.
Example
Illustrative example, not a real client.
customer portal needs self-serve password reset. currently has to call support, ~40 calls/wk. security wants MFA. must work on mobile. no budget for new infra, use existing auth provider.
BR-01 (Must) Users shall be able to reset their password without contacting support. Rationale: reduces support call volume (~40/wk observed). FR-01.1 (Must) The system shall require MFA verification before a password reset is completed. CONSTRAINT: Must use existing auth provider — no new infrastructure. [TO CONFIRM: mobile browser support matrix not specified in source notes]
How it works with Claude
Takes about 10 minutes to add the skill to Claude, and it's a one-off.
Paste in your notes or brief. It'll ask back if it's missing something that would block the draft.
A numbered, structured BRD or FRD lands, ready for your review pass.
FAQs
Can AI write a full BRD from workshop notes?
A well-built AI skill can produce a solid first-pass BRD structured with numbered requirements, scope, stakeholders and assumptions from raw notes. It still needs a BA to review it, confirm assumptions, and validate the requirements are accurate.
What's the difference between a BRD and an FRD?
A BRD describes the business need, scope and benefits in business language. An FRD describes system behaviour in detailed functional requirements, in solution/system language. This skill can draft either.