AI-Augmented Requirements Elicitation

What Elicitation Is and Why It Matters

Requirements elicitation is the process of drawing out what stakeholders actually need — as opposed to what they say they want, which are often two different things. It involves workshops, interviews, observation, and document analysis. The quality of everything downstream — design, development, testing — depends on how well elicitation was done.

AI does not replace elicitation. No AI tool can substitute for the conversation where a stakeholder reveals, in the seventh minute of a one-on-one, that the real constraint is a regulatory audit coming in Q3 that no one mentioned in the group workshop. That disclosure happens in a trust relationship.

What AI does is compress the preparation and synthesis work on either side of the conversation.

Pre-Session Preparation With AI

Before a requirements workshop or stakeholder interview, a BA typically prepares:

• A question guide covering the scope areas
• Background research on the domain and stakeholder context
• A review of any existing documentation (legacy system specs, previous project artifacts)

AI accelerates all three.

Generating a question guide:

I'm conducting a requirements elicitation session with
the finance operations team for a new accounts payable
automation system. The current process is manual — invoices
are received by email, checked against purchase orders in
a spreadsheet, and approved in two sign-off stages.

Generate a structured question guide for a 90-minute workshop.
Cover: current process pain points, approval workflow edge cases,
exception handling, integration with existing ERP,
audit and compliance requirements, and success metrics.
Format: topic heading + 3-5 questions per topic.

The output is a draft — review it, remove questions that don't apply to your stakeholders, and add the ones you know the AI doesn't know to ask based on your domain context.

Document gap analysis:

Here is the existing system specification for the legacy
accounts payable module: [paste document]

I am scoping a replacement system. Identify:
1. Business rules stated explicitly in this document
2. Business rules implied but not stated
3. Areas where the specification is ambiguous or incomplete
4. Questions I should ask stakeholders to resolve the gaps

Format: numbered list per category.

This produces a prioritized list of gaps to probe in the session — work that previously required hours of manual document analysis.

Post-Session Synthesis With AI

After an interview or workshop, the BA typically:

• Reviews notes or transcripts
• Extracts requirements
• Identifies gaps and follow-up questions
• Drafts initial requirements artifacts

All of this is AI-compressible.

Transcript to requirements:

Here is the transcript from a 60-minute requirements
workshop with the finance operations team: [paste transcript]

Extract:
1. Functional requirements — what the system must do
2. Non-functional requirements — performance, security, compliance
3. Business rules — constraints and logic the system must enforce
4. Ambiguous statements — things that were said but need clarification
5. Follow-up questions — gaps that need a subsequent session

Format each as a numbered list. For ambiguous items,
include the original quote and the two most likely interpretations.

The AI output gives you a structured starting point for the requirements document — but you validate it. The "ambiguous statements" section is particularly valuable: it identifies the items that are most likely to generate rework if left unresolved.

Cross-session gap analysis:

I've conducted interviews with three stakeholder groups:
[paste combined synthesis notes]

Identify:
1. Requirements that appear in all three groups
2. Requirements that appear in only one group
3. Conflicts — where two groups have stated incompatible requirements
4. Gaps — business areas that none of the groups addressed
   but that are likely to be important

Format: category heading + numbered list per category.

Finding conflicts across stakeholder groups is one of the highest-value BA activities — and it's tedious to do manually across long interview transcripts. AI does this in seconds.

Real-Time Synthesis in Workshops

For workshops where you capture notes in real time, AI can support live synthesis if you have a note-taker or can brief AI with running notes during breaks:

Mid-session gap check:

Here are the notes from the first hour of our workshop
on the accounts payable system: [paste running notes]

We have 30 minutes remaining. What critical topics
have we not yet covered that were in the scope?
What follow-up questions should I ask in the remaining time?

This turns a break into a structured "what are we missing?" audit that improves the quality of the second half of the session.

Validating AI-Generated Requirements

Every AI output from elicitation synthesis requires validation against the source material. The validation checklist:

• Source traceability: Can each requirement be traced back to a specific stakeholder statement? If you can't trace it, the AI may have inferred it. Flag inferred requirements for confirmation.
• Completeness check: Does the output include all the requirements you recall from the session, or only the ones that were stated most clearly? Ambiguous or offhand comments often get dropped.
• Conflict preservation: Did the AI resolve conflicts silently (picking one interpretation) or flag them? Silently resolved conflicts are the most dangerous — they look like complete requirements but aren't.
• Business rule accuracy: Business rules are often domain-specific and nuanced. Verify AI-extracted business rules against your own domain knowledge and the source transcript.

Key Takeaways

• AI compresses pre-session preparation and post-session synthesis — not the elicitation conversation itself
• Pre-session: use AI to generate question guides, gap-analyze existing documentation, and prepare domain background
• Post-session: use AI to extract requirements from transcripts, identify ambiguous items, and compare across stakeholder groups
• Every AI-generated output requires validation against the source — traceability, completeness, conflict preservation, and business rule accuracy
• The BA's judgment in validating AI output is the new high-value activity

---

Try It: Take notes or a transcript from any meeting where requirements were discussed. Run it through Claude or ChatGPT with the post-session synthesis prompt. Review the output against your own recollection. Note specifically what the AI captured correctly, what it missed, and what it may have inferred that wasn't explicitly stated.