Can AI extract requirements from handwritten meeting notes?

Yes. Using a well-constructed prompt with confidence tagging, Claude extracted 6 business requirements, 14 stakeholder requirements, 20 solution requirements, 7 integration points, all open conflicts, and 10 action items from five photographed pages of handwritten kickoff notes — in approximately two minutes. The key is prompt design: insisting on confidence levels ([HIGH], [INFER], [UNCERTAIN]) and preventing premature synthesis of conflicting stakeholder positions.

How much time does AI save in business analysis work?

In a practical test, the Week 1 post-kickoff synthesis — which took approximately 28 hours across five days in 2016 (typed notes, meeting minutes, requirements list, interview plan, conflict log) — was compressed to roughly 4 hours in one afternoon using AI. This is an order-of-magnitude improvement, not a marginal efficiency gain.

What can AI not do in business analysis?

AI cannot read the room — it cannot see that a stakeholder looked uncomfortable or that a concern carried more emotional weight than the words suggest. It cannot supply domain context (knowing whether a regulation is mandatory or customer-driven). It cannot elicit the unsaid — requirements that were in the air but never voiced. And it cannot navigate stakeholder politics. AI handles structure; the BA handles judgement.

The Business Analyst’s Reinvention: Five Pages from 2016, Two Minutes in 2026

I gave my decade-old kickoff notes to Claude. Here is what AI extracted, what it missed, and what every working BA should take from it.

Earlier this week, while sorting through an old shelf at home, I found a consulting notebook from 2016. Inside, five pages of dense handwritten notes from a kickoff meeting for a Laboratory Information Management System (LIMS) project at an Indian downstream refinery. Blue ink on the body, red sketch-pen for the margin stars and critical questions, the date 1 Mar 16 written compactly in the top-right corner.

I sat with the notebook for a few minutes. Then I had an idea I would not have had a year ago: What would happen if I photographed those pages and gave them to Claude?

What This Notebook Actually Contains

The kickoff was 90 minutes long. Nine people in Conference Room B at 10:30 in the morning — the customer’s GM Operations as Sponsor, the Lab Head, the Lab Operations Manager, a Senior Chemist, the QA and Compliance lead, the IT Infrastructure Manager, the Instrument Engineering Lead, the DCS and Process Control engineer, and myself as the BA.

The five pages capture, in order:

Page 1 — Attendees, sponsor’s opening, business drivers, an early conflict (cloud-ready vs on-prem)
Page 2 — Pain points by stakeholder. Strike-throughs, red star margin annotations, boxed quick-wins
Page 3 — Scope discussion with a freehand system context diagram — LIMS at the centre, seven surrounding systems with arrows
Page 4 — Compliance must-haves, the sample lifecycle as a seven-step flow
Page 5 — Risks, three numbered conflicts with status, actions for the next two weeks

Where This Sits in BABOK

Three knowledge areas are in play: Business Analysis Planning and Monitoring (background), Strategy Analysis (Page 1 — sponsor’s current state and desired future state), and dominantly Elicitation and Collaboration — the kickoff is the first formal elicitation event. BABOK is explicit that elicitation is iterative and incremental.

The central question: what is the BA actually doing with these five pages between the end of the kickoff and the start of the next elicitation round? That is where 2016 and 2026 diverge sharply.

What Kickoff Notes Used to Mean — The 2016 Workflow

Walk through the chain of documents that flowed from those five handwritten pages:

Day 1–2: Typed Word document structured by stakeholder, quotes preserved verbatim. 6–8 hours.
Day 2–3: Formal Kickoff Meeting Minutes (3–5 pages). 6–8 hours.
Day 4: First-draft requirements list in BABOK’s layered structure. 6–8 hours.
Day 5 morning: Stakeholder Interview Plan matrix. 2–3 hours.
Day 5 afternoon: Conflict and Open Questions Log. 2–3 hours.

Beyond Week 1:

Weeks 3–6: Business Requirements Document (BRD), 40–80 pages
Weeks 5–8: Software Requirements Specification (SRS)
Weeks 6–10: Use Cases and User Stories
Throughout: Requirements Traceability Matrix
Weeks 10–14: Acceptance Test Plan

The kickoff scribble pad was the seed. Eight major documents and roughly fourteen weeks of work eventually flowed from it. The discipline produced excellent results — but it was painfully slow.

The Experiment — Turning to Claude

I photographed all five pages. Five JPEGs uploaded to Claude. Then a single carefully-constructed prompt.

The prompt asked Claude to extract requirements in BABOK-aligned form: stakeholders present, layered requirements (business, stakeholder, solution, transition, integration), direct quotes, open conflicts, open questions, action items, implied requirements, and scope statements. Every item tagged with confidence level: [HIGH], [INFER], or [UNCERTAIN].

Two critical prompt design choices:

Confidence tagging — transforms output from “here is what I read” to “here is what I read, here is what I am guessing, here is what I could not read clearly”
No premature synthesis — preserving each stakeholder’s position in their own framing so the BA can see conflicts

These prompt-design decisions take fifteen minutes of thought up front and save five hours of confusion downstream.

What AI Extracted

In roughly two minutes, Claude returned:

All nine stakeholders correctly identified with roles, including margin-flagged absence of Marketing and Dispatch
6 business requirements
14 stakeholder requirements organized by speaker
14 functional and 6 non-functional requirements
7 integration requirements from the hand-drawn system context diagram
4 transition requirements with risks
All three open conflicts with both positions preserved
4 verbatim quotes flagged for the SRS
6 implied requirements
10 action items by owner

Highlights of what AI handled well:

Strike-throughs correctly interpreted as self-corrections
Direct quotes preserved verbatim with attribution
Cross-page contradiction caught (Sponsor’s “cloud-ready” vs IT Manager’s “cloud is out”) flagged as Open Conflict #1 with status UNRESOLVED
Red margin annotations preserved as the BA’s judgement, not content

The Honest Verdict — What AI Missed

A few words genuinely unreadable — correctly flagged as [UNCERTAIN]
One stakeholder’s surname incomplete — AI faithfully recorded what was there
Cross-page reasoning relied on the BA’s red-ink margin flags
Domain context not supplied (21 CFR Part 11 identified but regulatory applicability unknown)
People were invisible — AI could not see that Mr Suresh looked uncomfortable, or that Dr Meena’s concern carried more weight than words suggest
The unsaid stayed unsaid — requirements in the air but never voiced (training, change management, politics)

“AI handles structure. The BA handles domain judgement. AI extracts what is on the page. The BA elicits what is not on the page.”

The Time Math

2016 Week 1 synthesis: ~28 hours across 5 days

2026 Day 1 synthesis: ~4 hours in one afternoon

The day-one synthesis work has compressed by an order of magnitude — from one working week to one working afternoon.

What This Really Means for the BA — The Reinvention

The four to six hours saved on day one is not the headline. What the BA does with those saved hours is.

Three options:

Option 1: Pocket the time

Leave at 4 PM. The cost-saving accrues to the consulting firm. The BA does not grow. This is the dumbest possible use and quietly commoditizes BA work.

Option 2: Compress the project timeline

Start stakeholder interviews on day 2 instead of week 2. Reasonable — but still measures the BA in hours billed.

Option 3: Reinvest the saved time in the work AI cannot do

Take extracted requirements back to each stakeholder in person for confirmation. Walk the lab floor a second time. Read the 80-page audit report cover to cover. Spend an hour over coffee probing the real concerns. Follow up the cloud contradiction with the Sponsor the same week.

“This is the work that produces a dramatically better BRD. Fewer requirements changes during development. A sponsor who feels genuinely heard. A development team that builds the right thing the first time. The constraint has always been time. AI just removed the constraint.”

Closing Thought

The skills that defined a competent BA in 2016 — careful documentation, structured synthesis, rigorous traceability — are now table stakes that AI handles in minutes. The skills that will define a senior BA in 2030 — judgement, elicitation, political navigation, deep domain context, reading the unsaid — are the skills the BAs of 2016 also had, but did not have time to fully exercise.

“AI is not making the BA’s job smaller. It is making the BA’s most valuable work, finally, possible at scale. Reinvention is not a single decision. It is a practice. Let us get on with it.”

Originally published in the Reinvention in the AI Era newsletter on LinkedIn.