AI-Enabled Business Analyst Toolkit

Introduction

30 Second Summary

Every company has processes that eat time and money. Business Analysts are the people who untangle the mess, figure out what's actually going wrong, and design the fix. With AI tools, they can do this faster and more rigorously than ever before.

In this project, you will use Claude Pro as your AI co-pilot to produce a complete set of Business Analysis deliverables for a fictional logistics company exploring AI-powered automation. You'll synthesize stakeholder interviews, map processes, write AI-specific requirements, and build a quantified business case.

What You'll Build

You'll walk someone through a Notion portfolio page showing how you untangled a messy logistics problem into five clear, interconnected deliverables, each produced with AI assistance and refined with your own business judgment.

By the end of this project, you'll have:

• A stakeholder interview synthesis that distills contradictory input from three stakeholders into structured themes, root-cause pain points, and measurable success criteria you can present with confidence.
• As-Is and To-Be swimlane process diagrams in Lucidchart that expose current bottlenecks and mark exactly where AI acts autonomously, where humans review, and where humans decide.
• AI-specific requirements with guardrails featuring testable acceptance criteria, escalation triggers, and a catalog of what the AI must never do, paired with a quantified business case showing projected cycle time and error rate improvements.
• Secret Mission: Design a UAT test plan with edge-case scenarios and a guardrails validation checklist for AI components.

Are there any prerequisites?

You'll need a Claude Pro subscription, a free Lucidchart account, and a free Notion account. No coding experience is required.

Before We Start

Before we dive in, let's lock in what you're about to build and why it matters. You'll produce a complete set of Business Analysis deliverables for a realistic AI consulting engagement, using Claude Pro to accelerate your work while applying the human judgment that separates a Senior BA from a junior one.

Set Up Your AI-Enabled BA Workspace

Every consulting engagement starts with structure. Before you interview stakeholders or map processes, you need a place where each deliverable lives and an AI assistant that understands the engagement context.

In this step, you will set up a Notion page to house your five BA artifacts and a Claude Project loaded with the Meridian Logistics scenario. Once both are ready, Claude can reference your engagement details across every artifact you produce.

In this step, get ready to:

• Create a Notion project page with five artifact sections.
• Create a Claude Project loaded with the Meridian Logistics scenario brief.
• Test that Claude references your uploaded context in its responses.

Create your Notion project page

A consulting BA organizes deliverables by engagement. Your Notion page will hold five sections, one for each artifact you produce in this project.

• Open your browser and navigate to notion.so. Sign in to your account.
• In the sidebar, click the + button next to Private to create a new page.
• Title the page Meridian Logistics - AI Exception Handling.
• Inside the page body, type ## Stakeholder Discovery followed by a space to create an H2 heading. Press Enter twice to leave space below it.
• Repeat for each remaining section: type ## As-Is Process, then ## To-Be Design, then ## AI Requirements, then ## Business Case.

You should now see five H2 headings stacked on the page, each with empty space beneath it ready for artifacts.

Why structure the page this way?

Each heading corresponds to one artifact you will produce in the next steps. When you deliver BA work to a client, they expect a clear table of contents. This page mirrors that structure from the start.

Don't see the headings?

Make sure you type ## followed by a space before the heading text. Notion converts it to an H2 heading automatically. If it appears as plain text, delete the line, retype ## with a trailing space, then type the title.

Still having trouble formatting your Notion page?

Create a Claude Project with scenario context

Claude Projects let you upload documents and set custom instructions so every chat within that project shares the same context. You will load the full Meridian Logistics scenario brief so Claude can reference it across all future prompts.

• Open a new browser tab and navigate to claude.ai/projects. Sign in with your Claude Pro account.
• Click + New Project in the upper right corner.
• Name the project Meridian Logistics BA.
• In the project instructions area, paste the following custom instructions:

You are a Senior Business Analyst assistant working on a client engagement for Meridian Logistics. Your role is to help produce professional BA deliverables including stakeholder synthesis documents, process maps, requirements specifications, and business cases.

Context: Meridian Logistics is exploring AI-powered automation for their delivery exception handling process. The engagement involves three key stakeholders (VP Operations, CS Manager, IT Director) with different priorities and concerns.

When producing artifacts:
- Use clear, structured formatting with headers and sections
- Include specific metrics and numbers from the scenario context
- Flag assumptions that need stakeholder validation
- Distinguish between facts from the scenario and your recommendations
- Follow consulting BA conventions (quantified impacts, actionable recommendations)

What do these instructions do?

Project instructions tell Claude what role to adopt and how to format responses for every chat within this project. You do not need to repeat this context in future prompts.

Claude performs best when project instructions cover general context and role guidelines. Task-specific instructions go in the chat itself.

Your project instructions are saved. Now you need to create the scenario brief document and upload it to the project's knowledge base.

• Open any text editor on your computer (TextEdit on macOS, Notepad on Windows, or Google Docs).
• Paste the following content as the first section of your scenario brief:

# MERIDIAN LOGISTICS - AI EXCEPTION HANDLING
## SCENARIO BRIEF

### COMPANY OVERVIEW
Meridian Logistics is a regional delivery company operating across the Pacific Northwest (Washington, Oregon, Idaho). They handle 12,000-15,000 deliveries per week across B2B and B2C channels. The company has grown 40% in 2 years but operations processes have not scaled.

### THE PROBLEM
Delivery exceptions (damaged shipments, late deliveries, wrong items, address issues) consume disproportionate resources. The CS team handles ~800 exception tickets per week. Average resolution time is 48-72 hours. Customer satisfaction dropped from 4.2 to 3.6 over the past year. The VP of Operations wants to explore AI-powered automation for classifying, routing, and resolving delivery exceptions.

### STAKEHOLDER PROFILES
1. Marcus Chen - VP of Operations: Owns exception handling end-to-end. Primary goal: reduce resolution time to under 4 hours for at least 70% of cases. Has budget authority.
2. Sarah Okafor - Customer Service Manager: Manages 12 CS reps. Team spends 60% of time on repetitive classification/routing. Wants automation for simple cases. Worried about job impact on her team.
3. David Park - IT Director: Owns technical infrastructure (ServiceNow, SAP). Requires human approval for financial transactions over $200. Non-negotiable on data security and customer PII.

### CURRENT-STATE METRICS
- Exception tickets per week: ~800
- Average resolution time: 48-72 hours
- Manual classification accuracy: 85%
- Misrouted tickets requiring re-assignment: 22%
- CS rep time spent on exceptions: 60% of working hours
- Customer satisfaction (exception-related): 3.6/5.0
- Refund approval cycle: 24 hours average
- Top exception types: Damaged (35%), Late (28%), Wrong Item (20%), Address Issue (12%), Other (5%)

What does this section establish?

This is the factual foundation of the engagement. It gives Claude the company context, who the stakeholders are, and the quantified pain points. Every artifact you produce will reference these numbers and names.

Confirm your document contains the company overview, stakeholder profiles, and metrics. Now add the interview notes below the metrics section.

• Paste the following raw interview notes below the metrics in the same document:

### RAW INTERVIEW NOTES

**MARCUS CHEN (VP Operations):**
"We're drowning. Two years ago we had 400 exceptions a week and it was manageable. Now we're at 800 and climbing. My team is burning out. I need resolution time under 4 hours for at least 70% of cases. I don't care how we get there - AI, automation, whatever - but the board is asking why our CSAT is tanking and I need answers by Q2. The real killer is the classification step. Reps spend 15-20 minutes per ticket just figuring out what type of exception it is and who should handle it. If we could automate that alone, we'd save thousands of hours per year."

**SARAH OKAFOR (CS Manager):**
"My team is exhausted from the repetitive work. They're skilled people doing data entry. Every ticket comes in, they read it, categorize it, figure out who handles it, route it manually. Then they wait. The specialist teams take 18 hours on average to pick up a routed ticket. By the time it's resolved, the customer has already called back twice. I want full automation for the simple cases - damaged item under $50? Just auto-approve the replacement and send the customer a message. Don't make my rep spend 20 minutes on a $30 item. Let them focus on the complex stuff where they actually add value."

**DAVID PARK (IT Director):**
"I'm open to AI but I need guardrails. We can't have a system auto-approving refunds without human oversight - not for anything over $200. Our compliance team would shut us down. Integration-wise, whatever we build needs to talk to ServiceNow for tickets and SAP for refund processing. I don't want another standalone tool that creates data silos. Data security is non-negotiable. Customer PII stays encrypted, the AI model doesn't get trained on our customer data without explicit consent frameworks. And no customer-facing AI responses without a human reviewing them first. One bad automated email and we're on social media for all the wrong reasons."

Why include raw interview notes?

These notes contain contradictions, emotional language, and implicit priorities. In Step 2, you will prompt Claude to synthesize them into structured insights. The rawness is intentional. A BA's job is to find signal in messy stakeholder input.

• Save the file as scenario-brief.md (or .txt) to your Desktop.
• Return to your Claude Project page at claude.ai/projects. Open the Meridian Logistics BA project.
• On the right side of the project page, find the project knowledge base area.
• Click the + button to add content, then upload your scenario-brief.md file.

You should see the file appear in the project knowledge base. Claude will process this information and use it as context in all chats within the project.

File not uploading?

Ensure the file is saved with a .md or .txt extension. Claude accepts text files, documents, and code snippets in the project knowledge base. If your browser blocks the upload, try dragging the file directly into the knowledge base area.

Help me upload a file to my Claude Project knowledge base.

Test your AI workspace

The workspace only works if Claude can actually reference the scenario you uploaded. Time to verify that the project knowledge is active.

• Inside your Meridian Logistics BA project, start a new chat.

Before you send this test prompt, think about what you expect the response to include. If the project knowledge is working, what specific details should Claude mention?

• Type the following prompt and send it:

What are the three main bottlenecks in Meridian Logistics' current exception handling process, and what metrics support each one?

Claude's response should reference specific numbers from your scenario brief: the 15-20 minute classification time, the 18-hour specialist queue wait, and the 24-hour refund approval cycle. It should also cite metrics like the 800 tickets per week and 48-72 hour resolution time.

What if the response is generic?

If Claude gives a vague answer without citing specific numbers from the scenario, your project knowledge may not be active. Check that the file appears in the knowledge base on the right side of the project page. Try clicking on the file to confirm it uploaded completely.

If the file shows but Claude still ignores it, start a fresh chat within the project. The first chat after uploading sometimes needs a fresh start to pick up new knowledge.

Your workspace is ready. You have a structured Notion page waiting for artifacts and a Claude Project that understands the full Meridian Logistics engagement. Next up, you will use Claude to synthesize the raw stakeholder interviews into a structured discovery document.

Synthesize Stakeholder Interviews

Your workspace is ready. Claude has the scenario context, and your Notion page is waiting for artifacts. Now it is time to produce the first deliverable a phData BA creates after discovery workshops: the stakeholder interview synthesis.

Raw interview notes are messy and contradictory. Three stakeholders each described their pain differently, with conflicting priorities. The BA's job is to find the signal: what are the real problems, where do stakeholders align, and what does success look like?

In this step, get ready to:

• Prompt Claude to produce a structured synthesis from the three stakeholder interviews.
• Refine the synthesis by challenging vague statements and resolving contradictions.
• Finalize the document in Notion with a BA Commentary callout box.

Prompt Claude for a structured synthesis

The scenario brief you uploaded in Step 1 contains raw interview notes from three stakeholders: the VP of Operations, the Customer Service Manager, and the IT Director. Each described the delivery exception problem from their own vantage point.

Your first move is to get Claude to distill these notes into a structured document. You will provide a specific template so the output is organized the way a Senior BA would present it to a client.

• Switch back to your Meridian Logistics BA project in claude.ai.
• Start a new chat within the project by clicking Start a new chat.
• Paste the following prompt to request a structured synthesis:

Using the raw interview notes from all three stakeholders (VP Operations, CS Manager, IT Director) in the project knowledge base, produce a structured Stakeholder Interview Synthesis document.

Use this template:

1. Themes and Patterns - recurring ideas across all three interviews
2. Stakeholder Alignment Matrix - where they agree and disagree
3. Pain Points with Quantified Impact - each pain point tied to a number (time, cost, volume)
4. Contradictions and Tensions - direct conflicts between stakeholder priorities
5. Proposed Success Metrics - measurable criteria for project success

For each section, cite which stakeholder(s) the insight came from.

Why this template structure?

These five sections mirror what a consulting BA delivers after a discovery workshop. The Alignment Matrix and Contradictions sections are particularly valuable because they surface political dynamics that derail projects if left unaddressed.

Before you read Claude's response, think about what you expect. Do you think Claude will produce specific numbers, or will it default to vague language like "improve efficiency"?

• Press Enter to send the prompt and read through the full response.
• As you read, note two things: where Claude pulled specific numbers from the interviews, and where it fell back on generic phrasing.

You should see a structured document with all five sections. Claude will likely get the themes and alignment right, but pay attention to the Pain Points section. Generic statements like "manual processes are slow" are not useful to a stakeholder presentation.

Claude not referencing the scenario context?

Make sure you started the chat inside your Meridian Logistics BA project, not on the main Claude chat page. The project knowledge base is only available within the project.

Check that the project name appears in the header of your chat window. If it does not, close the chat and open a new one from the project page.

Help me troubleshoot why Claude is not using my uploaded scenario brief in its responses.

Refine the synthesis with human judgment

Claude's first output is a draft, not a deliverable. A Senior BA's value comes from pushing back: making vague claims specific, resolving contradictions that Claude left as "both views have merit," and identifying which pain points are root causes versus symptoms.

• In the same chat, send the following refinement prompt to challenge vague language:

Revise the Pain Points section. Every pain point must include a specific number from the interviews. Replace vague phrases like "improve efficiency" with measurable statements.

For example, "manual exception classification" should become "manual classification takes 15-20 minutes per ticket, with CS reps handling 40-60 exceptions daily."

Also in the Success Metrics section, make every metric a specific target with a number, not a direction. "Reduce resolution time" should become "reduce exception resolution time from 48-72 hours to under 4 hours."

• Send the prompt and review whether Claude now uses specific numbers from the scenario.

You should see the Pain Points section transform. Each point should now reference concrete metrics: ticket volumes, time-per-task, and resolution cycle lengths.

• Now send this follow-up to resolve the stakeholder contradictions:

Revise the Contradictions and Tensions section. Don't just list both sides neutrally. For each contradiction, add a "BA Recommendation" that proposes a resolution.

For example: The CS Manager wants full automation of exception handling. The IT Director insists on human approval for any refund over $200. Propose a resolution that satisfies both: AI handles classification and routing automatically, but refunds above $200 require human sign-off before processing.

Also identify which pain points are root causes versus symptoms. Manual reclassification is a symptom. The root cause is lack of a consistent classification taxonomy.

• Send the prompt and read Claude's revised contradictions section.
• Review whether the BA Recommendations feel realistic. If any resolution feels too simple or ignores organizational politics, push back with a follow-up prompt in your own words.

What does good refinement look like?

The goal is not to get Claude to produce a perfect document in one shot. The goal is to demonstrate the AI-enabled BA workflow: prompt, validate, push back, refine.

Your judgment adds value at every iteration. Claude knows the words. You know which contradictions will actually block a project and which success metrics a VP of Operations will care about.

Claude's refinements still too generic?

If Claude keeps producing vague output, try being more directive. Instead of asking it to "be more specific," give it the exact format you want: "State the metric as: [current state number] to [target number] within [timeframe]."

Help me write a more directive prompt to get Claude to produce specific, measurable pain points instead of vague improvement statements.

Finalize and add BA Commentary in Notion

The refined synthesis is ready to move into your portfolio. But a polished document alone does not demonstrate your value as a BA. The BA Commentary callout box is what separates "I used AI" from "I used AI with professional judgment."

• Return to your Meridian Logistics - AI Exception Handling page in Notion.
• Click into the Stakeholder Discovery section.
• Copy the final refined synthesis from Claude (all five sections: Themes and Patterns, Stakeholder Alignment Matrix, Pain Points with Quantified Impact, Contradictions and Tensions, Proposed Success Metrics) and paste it into this section.
• Format the section headings using Notion's Heading 2 blocks so each section is clearly separated.

Now add the commentary that demonstrates your professional judgment.

• At the bottom of the Stakeholder Discovery section, type /callout to insert a Notion callout block.
• Title the callout BA Commentary.
• Inside the callout, write three short paragraphs covering:

Make the commentary your own

Do not ask Claude to write the BA Commentary for you. This section is where YOU demonstrate critical thinking. Write it in your own voice based on what you actually observed during the prompt-refine cycle.

• Review your completed Stakeholder Discovery section in Notion.

You should see a structured synthesis document with five clearly headed sections, each containing specific numbers and stakeholder attributions. Below it, your BA Commentary callout box should show three distinct paragraphs explaining where AI helped, where you added value, and what questions remain.

Synthesis feels too long or unfocused?

A good synthesis is concise. If any section runs longer than half a page, trim it. Each pain point should be 1-2 sentences with a number. Each theme should be a single clear statement. If Claude gave you paragraphs where bullets would suffice, reformat.

Help me tighten my stakeholder synthesis so each section is concise and scannable for a client presentation.

Your first artifact is complete. You have demonstrated the core AI-enabled BA workflow: prompt Claude with structure, validate critically, push back on vague output, and finalize with your own professional commentary. Next up, you will turn the pain points you just documented into a visual process map that stakeholders can immediately understand.

Map the As-Is Process in Lucidchart

Your stakeholder synthesis revealed three massive bottlenecks eating 48-72 hours of resolution time. But a text document full of findings is not what you show a VP of Operations when you need sign-off.

You need a visual map of the current process that makes the pain impossible to ignore. A swimlane diagram assigns every step to a specific actor, exposing the handoffs and wait times that text alone cannot convey. In this step, you will use Claude to draft the process sequence, then build it as a formal BPMN 2.0 diagram in Lucidchart.

In this step, get ready to:

• Generate a structured current-state process description with Claude.
• Build a swimlane diagram in Lucidchart using BPMN 2.0 shapes.
• Annotate the three major bottlenecks and embed the diagram in Notion.

Generate the current-state process with Claude

Before you open a diagramming tool, you need the process steps written out in sequence. Claude already has the scenario context from your uploaded brief, so it can draft the full exception handling flow for you to validate.

• Switch back to your Meridian Logistics BA project in claude.ai.
• Start a new chat and paste the following prompt:

Based on the Meridian Logistics scenario brief, describe the current-state exception handling process step-by-step.

Structure it as a swimlane breakdown with four actors:
- Customer
- CS Rep
- Specialist Team
- Finance

For each step, include:
- Which actor performs it
- What the handoff looks like (email, system queue, manual transfer)
- Estimated time or wait time at that step
- Any decision gates (e.g., "Is refund needed? Yes/No")

Format as a numbered sequence I can translate directly into a BPMN swimlane diagram.

What does this prompt do?

You are asking Claude to structure the process as a numbered sequence mapped to specific actors. This format translates directly into a swimlane diagram because each step already names its lane (actor), its shape type (decision gates become diamonds, activities become rectangles), and its connections (handoffs between lanes).

• Review Claude's output for logical completeness.
• Check that the flow starts with the Customer reporting an issue and ends with resolution or closure.
• Verify that Claude included at least one decision gate (e.g., "Does this need a refund?" or "Is the exception classifiable?").
• Adjust the sequence if any step seems out of order or if a handoff between actors is missing.

You should now have a numbered list of 10-15 process steps, each tagged with an actor name and a time estimate.

Claude's output too generic?

If Claude produces vague steps like "process the exception," push back. Reply with: "Be more specific about what the CS Rep physically does at step 3. What system do they open? What fields do they fill in?" The scenario brief contains enough detail for Claude to be concrete.

Help me get more specific process steps from Claude for my As-Is map.

Build the swimlane diagram in Lucidchart

With your process sequence ready, it is time to turn it into a formal BPMN 2.0 swimlane diagram. Each actor gets their own horizontal lane, and the process flows left-to-right across lanes as work passes between people.

• Open lucidchart.com in your browser and sign in to your account.
• Click + New and select Blank Document to create a fresh canvas.
• Enable the BPMN 2.0 shape library by clicking the Shapes button in the upper-left corner of your window.
• In the window that appears, check the BPMN 2.0 box, then close the panel.

You should now see BPMN 2.0 shapes (pools, events, gateways, tasks) available in the shape library on the left side of your canvas.

What is BPMN 2.0?

Business Process Model and Notation is a standardized visual language for documenting business processes. Ovals represent start and end events, rectangles represent activities, and diamonds represent decision points. Swimlanes organize everything by who is responsible.

• Drag a Pool shape from the BPMN 2.0 library onto the canvas.
• Use the Properties bar to add lanes until you have four swimlanes within the pool.
• Label each lane from top to bottom: Customer, CS Rep, Specialist Team, Finance.

Your canvas now shows a four-lane pool ready to receive process steps.

• Place an oval (start event) in the Customer lane on the far left. Double-click it and label it Exception Reported.
• Working left-to-right, drag rectangle shapes (activities) into the lane matching each step from your Claude output.
• For each decision gate in your process (e.g., "Refund needed?"), use a diamond (gateway) shape instead of a rectangle.
• Connect each shape to the next by hovering over a shape until you see the red node, then dragging an arrow to the next shape.
• When the process crosses from one actor to another (a handoff), draw the arrow from the shape in one lane to the next shape in the receiving lane.
• End the diagram with an oval (end event) labeled Exception Resolved in whichever lane handles the final closure step.

Your diagram should now show the full exception handling flow from left to right, crossing between lanes at each handoff point.

Keep flow direction consistent

BPMN best practice is to keep all flow moving left-to-right, even when crossing between swimlanes. Never route arrows upward or backward unless you are depicting a loop.

Shapes not appearing in the library?

Make sure you checked the BPMN 2.0 box specifically (not just "Flowchart" or "Containers"). Click the Shapes button in the upper-left corner again to verify the box is checked.

Help me find the BPMN shapes in Lucidchart.

Annotate bottlenecks and embed in Notion

A process map without annotations is just a pretty picture. The annotations are what make stakeholders feel the pain. You will highlight the three bottlenecks your synthesis identified with red callout shapes so they jump off the page.

• Locate the activity shape in the CS Rep lane where manual classification happens.
• Drag an Annotation shape from the BPMN 2.0 artifacts section onto the canvas near that activity.
• Double-click the annotation and type: BOTTLENECK: Manual classification takes 15-20 min per ticket.
• Select the annotation shape, click the Fill color icon in the formatting toolbar, and choose a red outline or red fill to make it visually prominent.

• Repeat for the second bottleneck: add an annotation near the Specialist Team queue activity with the text BOTTLENECK: Specialist queue wait averages 18 hours. Set it to red.
• Repeat for the third bottleneck: add an annotation near the Finance refund approval activity with the text BOTTLENECK: Manual refund approval loop adds 24 hours. Set it to red.

Your diagram now has three red annotation shapes calling out the exact pain points from your stakeholder synthesis.

Why red annotations?

When you present this diagram to stakeholders, the red annotations create an instant visual gravity. The VP of Operations does not need to read every box to understand where the process bleeds time. The red shapes draw the eye directly to the 48-72 hour gap.

• Title your Lucidchart document Meridian Logistics As-Is Exception Handling Process by clicking the untitled document name at the top of the editor.
• Take a screenshot of the completed diagram, or export it as a PNG by clicking File then Export then PNG.
• Return to your Notion page and scroll to the As-Is Process section.
• Embed the screenshot or exported PNG into the As-Is Process section by typing /image and uploading the file.

Before you move on, take a moment to look at the finished diagram. What do you expect to see?

You should see a four-lane swimlane diagram with process steps flowing left-to-right, decision diamonds at branching points, and three red annotation shapes highlighting the manual classification (15-20 min), specialist queue wait (18 hours), and refund approval loop (24 hours).

Diagram looks cluttered or overlapping?

If shapes overlap or arrows cross awkwardly, select all shapes in one lane and use the Distribute Horizontally option from the alignment toolbar. This spaces shapes evenly. Then reroute any crossing arrows so they do not overlap.

Help me clean up my BPMN diagram layout in Lucidchart.

Your As-Is process map is locked in. You have a visual artifact that makes Meridian's pain undeniable. Next up, you will design the To-Be process that shows exactly where AI can cut those 48-72 hours down to minutes.

Design the To-Be Process with AI Integration Points

Your As-Is map exposed three bottlenecks that consume 48-72 hours of resolution time. Now you design the future state that eliminates them with AI.

The critical skill here is drawing the automation boundary. You need to be precise about where AI acts autonomously, where it drafts-and-suggests, and where humans must remain in the loop.

In this step, get ready to:

• Prompt Claude to propose a To-Be process with three autonomy levels per AI capability.
• Build a second swimlane diagram in Lucidchart with an AI System lane and color-coded automation zones.
• Add comparison annotations showing projected time and accuracy improvements.

Prompt Claude for the To-Be process design

Each bottleneck in the As-Is map represents a specific failure mode: slow classification, inefficient routing, and a manual approval loop. You need Claude to propose an AI capability that directly addresses each one, along with clear autonomy boundaries.

• Switch back to your Meridian Logistics BA project in claude.ai.
• Start a new chat within the project and enter a prompt like the following:

Based on the three bottlenecks in the As-Is process (manual classification taking 15-20 minutes, specialist queue wait averaging 18 hours, and refund approval loop adding 24 hours), propose a To-Be process design that addresses each bottleneck with a specific AI capability:

1. AI classification replacing manual triage
2. AI-powered routing to specialist queues based on exception type and severity
3. AI-drafted resolution responses with auto-approval for low-value cases

For each AI capability, identify three autonomy levels:
- Full automation (AI acts without human involvement)
- Draft-and-suggest (AI proposes, human approves)
- Human-only (AI does not participate)

Structure the output as a swimlane process breakdown with five actors: Customer, CS Rep, Specialist Team, Finance, AI System. Include handoff points and decision gates where the AI's confidence level determines the routing path.

What does this prompt do?

This prompt gives Claude the specific bottlenecks to solve and constrains its design within a practical framework.

By requesting three autonomy levels per capability, you force Claude to think about the boundaries rather than defaulting to "automate everything." The five-actor structure prepares output you can translate directly into a Lucidchart swimlane.

• Review Claude's response and check for three things:
• If Claude's output is too generic or missing risk-based boundaries, push back with a follow-up prompt:

The autonomy boundaries need to be more specific. For each AI capability, define:
- What confidence threshold triggers full automation vs. human review?
- What dollar amount or risk level forces human-only handling?
- What happens when the AI encounters an exception type it has never seen before?

Also ensure the CS Manager's preference for full automation is balanced against the IT Director's requirement for human approval on refunds over $200.

Why push back on Claude here?

Claude tends to propose clean, optimistic automation boundaries. In real consulting, the boundaries are messier because different stakeholders have different risk appetites.

By referencing the contradiction you surfaced in Step 2 (CS Manager wants full automation, IT Director insists on human approval for refunds over $200), you demonstrate the BA skill of carrying stakeholder context forward across artifacts.

• Save Claude's refined output. You will use this as the blueprint for your Lucidchart diagram.

Build the To-Be swimlane diagram in Lucidchart

Your As-Is diagram has four lanes. The To-Be diagram adds a fifth lane for the AI System, making automation handoffs visually explicit.

• Switch back to Lucidchart and create a new blank document.
• Title it Meridian Logistics To-Be Exception Handling Process.
• From the Containers shape library on the left panel, drag a pool onto the canvas.
• Add five swimlanes to the pool, labeled: Customer, CS Rep, Specialist Team, Finance, and AI System.

Adding a fifth lane

Select the pool, then right-click an existing lane header and choose Insert After to add the AI System lane at the bottom. You can also use the + icon that appears when hovering over the pool edge.

• Place the process steps using BPMN 2.0 shapes from the shape library you enabled in the previous step:
• Place decision diamonds at each confidence-threshold routing point:
• Connect all shapes with sequence flow arrows moving left to right.

Your diagram now shows the process flow, but you cannot yet distinguish automation zones from human zones at a glance.

• Apply color coding to each activity shape using the fill color icon (paint bucket) in the formatting bar at the top of the canvas:

Why three colors?

This color scheme maps directly to the three autonomy levels from your Claude prompt. When a stakeholder looks at this diagram, they can instantly see how much of the process is automated (green), how much requires human oversight (amber), and what remains fully manual (red).

You should now see a five-lane swimlane diagram with colored shapes showing a clear visual distinction between AI-driven, AI-assisted, and human-only activities.

Shapes not showing color?

Make sure you are selecting the shape itself (not the text label inside it) before clicking the fill color icon. If the color applies to the text background instead of the shape, click away to deselect, then click the shape border to reselect.

help me fix color coding in my Lucidchart To-Be diagram

Add comparison annotations and embed in Notion

A To-Be diagram on its own shows the new process, but stakeholders want to know the projected impact. A comparison annotation translates the design into numbers they can evaluate.

• From the shape library, drag a text or annotation shape onto the canvas next to your process flow.
• Enter the following projected improvements:
• Style the annotation box with a visible border so it stands out from the process shapes.

Where do these numbers come from?

The "before" numbers come from the pain points you quantified in the stakeholder synthesis (Step 2). The "after" targets come from Claude's To-Be design proposal and represent realistic expectations for AI classification and routing systems, not guarantees.

In a real engagement, you would validate these targets with the engineering team before committing them to stakeholders.

• Export or screenshot the completed To-Be diagram.
• Switch back to your Notion page and scroll to the To-Be Design section.
• Embed or paste the To-Be diagram screenshot into this section, positioned alongside your As-Is map for easy comparison.

Your Notion page now shows the As-Is and To-Be diagrams side by side. A reader can immediately see the transformation from a four-lane manual process riddled with red bottlenecks to a five-lane AI-augmented process with clear green automation zones.

Diagram too small to read in Notion?

Export from Lucidchart at a higher resolution (use File > Export > PNG and set the resolution to 2x). Alternatively, embed the Lucidchart document directly using an embed block in Notion so viewers can zoom in.

help me embed my Lucidchart diagram in Notion so it's readable

You now have a To-Be process design that visually communicates exactly where AI takes over, where humans stay in the loop, and what improvement the client can expect. Next, you will translate these design decisions into formal requirements with testable acceptance criteria.

Write AI-Specific Requirements with Guardrails

Your To-Be diagram shows where AI acts in the process. Now you need to specify how it must behave. Traditional user stories break down for AI systems because AI outputs are non-deterministic.

You cannot write "the system shall classify exceptions correctly" and call it done. You need performance thresholds, acceptable error boundaries, fallback behaviors, and human escalation triggers. This is the artifact that separates a BA who has worked on AI projects from one who has not.

In this step, get ready to:

• Generate AI-specific user stories using the four-layer requirement pattern.
• Refine acceptance criteria into Given/When/Then format with measurable thresholds and a guardrails catalog.
• Finalize the requirements document in Notion with BA commentary.

Prompt Claude to generate four-layer user stories

The four-layer requirement pattern structures each AI requirement into four explicit layers: functional outcome, permissions and constraints, quality thresholds, and safe failure behavior. Each layer adds specificity that a traditional user story would leave vague.

What is the four-layer requirement pattern?

Layer 1 (Functional outcome) defines what the AI does. Layer 2 (Permissions and constraints) defines what it is and is not allowed to access or modify.

Layer 3 (Quality thresholds) defines how well it must perform, with measurable numbers. Layer 4 (Safe failure behavior) defines what happens when it cannot meet those thresholds.

Alongside each user story, you will also produce companion specifications using the AI Requirements Triangle: acceptable error boundary, context retrieval specification, and human escalation trigger.

• Switch back to your Claude Project chat for Meridian Logistics BA.
• Send the following prompt to generate your initial user stories:

Generate user stories for the three AI capabilities in our To-Be process:
1. AI-powered exception classification
2. AI-powered routing to specialist queues
3. AI-drafted resolution responses

For each capability, structure the requirement using the four-layer pattern:
- Layer 1: Functional outcome (what the AI does)
- Layer 2: Permissions and constraints (what it can/cannot access or modify)
- Layer 3: Quality thresholds (measurable performance targets)
- Layer 4: Safe failure behavior (what happens when thresholds aren't met)

Also provide AI Requirements Triangle companion specs for each:
- Acceptable error boundary
- Context retrieval specification
- Human escalation trigger

Base all details on the Meridian Logistics scenario context you have.

What does this prompt do?

This prompt asks Claude to produce structured requirements for each of the three AI capabilities you designed in Step 4. By naming the four layers and the triangle specs explicitly, you force Claude to address each dimension rather than generating vague user stories.

Claude will generate a set of user stories with companion specifications. You should see structured output covering all three capabilities with each of the four layers addressed.

• Review Claude's output for specificity. Reject any acceptance criterion that is not testable or measurable.
• Flag statements like "improve efficiency" or "handle exceptions accurately" that lack concrete numbers.
• Note which Layer 3 (quality thresholds) and Layer 4 (safe failure) entries are genuinely specific vs. vaguely stated.

Spot the vagueness

Claude tends to generate strong functional outcomes (Layer 1) but weak quality thresholds (Layer 3). Look for entries that say things like "high accuracy" or "fast response time" without a number attached. Those are the ones you will sharpen in the next substep.

Refine acceptance criteria and add the guardrails catalog

Claude's initial output gives you a skeleton. Now you apply the judgment that makes these requirements production-ready: explicit numbers in Given/When/Then format, and a guardrails catalog that defines what the AI must never do.

• Send this follow-up prompt to Claude to convert the acceptance criteria into Given/When/Then format:

Rewrite each user story's acceptance criteria in Given/When/Then format with explicit numbers. Use these thresholds from our To-Be design:

- AI classifier confidence threshold: 85%
- Auto-classification time target: under 30 seconds
- Routing accuracy target: 95%
- Resolution time target: under 4 hours for 70% of cases
- Refund approval limit for auto-processing: $200

For each story, provide:
- A "When confidence is ABOVE threshold" path (happy path)
- A "When confidence is BELOW threshold" path (fallback behavior)

Example format:
Given a delivery exception email,
When the AI classifier confidence is above 85%,
Then the system auto-classifies and routes to the correct specialist queue within 30 seconds.
When confidence is below 85%,
Then the system flags for human review and presents top-3 suggested categories.

Why Given/When/Then with two paths?

AI systems need two acceptance paths because they will not always meet their confidence threshold. The "above threshold" path defines success. The "below threshold" path defines graceful degradation. Both must be testable.

Without the fallback path, the system has no defined behavior for uncertain cases. That is where AI systems fail silently in production.

Claude should produce Given/When/Then criteria for each story. Review the output and confirm the numbers match the thresholds from your To-Be design.

• Verify that each user story now has at least two Given/When/Then paths (above-threshold and below-threshold).
• Confirm the 85% confidence threshold, 30-second classification time, 95% routing accuracy, and $200 refund limit appear in the appropriate stories.
• Check that the AI Requirements Triangle specs reference these same numbers consistently.

Now add the guardrails catalog. This defines what the AI must never do, regardless of confidence level.

• Send this prompt to Claude to generate the guardrails catalog:

Create a Guardrails Catalog for the Meridian Logistics AI system. This defines what the AI must NEVER do regardless of confidence level.

Include these four mandatory guardrails:
1. Must never approve refunds over $200 without human sign-off
2. Must never access customer payment data
3. Must never override a human agent's classification
4. Must never respond to customers directly without review

For each guardrail, specify:
- The rule (what is prohibited)
- The trigger condition (when this guardrail activates)
- The enforcement mechanism (how the system prevents violation)
- The escalation path (what happens if a violation is detected)

What is a guardrails catalog?

A guardrails catalog is a non-negotiable constraint document. Unlike quality thresholds (which have graceful degradation paths), guardrails are hard boundaries. The AI does not get to degrade gracefully past a guardrail. It simply must not cross it.

These boundaries come from business risk appetite and regulatory requirements, not from the AI's capabilities. That is why they require human judgment to define.

• Review Claude's guardrails catalog output. Confirm all four mandatory guardrails are present with enforcement mechanisms.
• Check that each guardrail has a testable enforcement mechanism, not just a statement of intent.

Claude's guardrails too generic?

If Claude produces enforcement mechanisms like "the system will prevent this" without specifying how, push back. Ask: "How specifically would the system enforce this guardrail? Name the technical control." Good answers include permission flags, input validation rules, or approval workflow gates.

Help me get more specific guardrail enforcement mechanisms

Finalize the requirements document in Notion

Your requirements are now scattered across multiple Claude chat messages. You need to consolidate them into a single structured document in Notion and add the BA commentary that makes this a portfolio artifact, not just AI output.

• Return to your Notion page "Meridian Logistics - AI Exception Handling" and scroll to the AI Requirements section.
• Paste in your finalized user stories, organized with each story as a heading followed by its four layers, Given/When/Then criteria, and AI Requirements Triangle specs.
• Paste the guardrails catalog as its own section below the user stories.

Ask Claude to format for you

You can ask Claude to consolidate everything into a single formatted document ready for pasting. Try: "Combine all the user stories, Given/When/Then criteria, AI Requirements Triangle specs, and guardrails catalog into one structured document." This saves manual assembly time.

Now add the BA Commentary section. This is what transforms Claude's output into a professional deliverable that shows your critical thinking.

• Below the guardrails catalog in your Notion page, add a callout block titled BA Commentary.
• Document what Claude generated well. Typically: functional outcomes (Layer 1), the structure and format of user stories, and the general shape of Given/When/Then criteria.
• Document what required human judgment to fix. Typically: the specific threshold numbers (85% confidence, $200 limit), the guardrail boundaries based on business risk appetite, and the escalation triggers that reflect organizational accountability.
• Note anything you would validate with stakeholders before finalizing (e.g., is 85% the right confidence threshold for auto-classification, or should it be higher for this industry?).

Why does the BA Commentary matter?

Anyone can paste Claude's output into a document. The commentary proves you understand which parts of the requirement need human validation and which are safe to accept from AI. This is exactly the judgment a Senior BA role requires: knowing where to trust the tool and where to override it.

Before you move on, scroll through your completed AI Requirements section. Do you expect to see at least four complete user stories with Given/When/Then criteria, the guardrails catalog, and your BA Commentary?

• Scroll through the full AI Requirements section in your Notion page from top to bottom.

You should see at least four user stories (one per AI capability, plus at least one additional story covering a cross-cutting concern like audit logging or confidence reporting). Each story has its four layers, Given/When/Then acceptance criteria with explicit numbers, and AI Requirements Triangle specs. Below them sits the guardrails catalog with four non-negotiable rules. At the bottom, your BA Commentary explains where AI helped and where you applied judgment.

Fewer than 4 user stories?

If you only have three stories (one per capability), go back to Claude and ask for a fourth covering a cross-cutting concern. Good candidates: "As an operations manager, I want an audit trail of all AI decisions" or "As a CS manager, I want confidence reporting so I can monitor AI accuracy over time."

Help me create a fourth user story for a cross-cutting concern

Your requirements document now defines exactly how the AI must behave, when it must defer to humans, and what it must never do. Next up, you will turn these capabilities into a quantified business case that proves the ROI to Meridian Logistics' leadership.

Build the Quantified Business Case

Your requirements document defines what the AI must do and how it must behave. Now stakeholders need to know why the investment is worth making.

A quantified business case translates your process improvements into dollars and time savings that executives can act on. This is the artifact that turns your analysis from "interesting" to "funded."

In this step, get ready to:

• Generate an initial ROI model with Claude covering cycle time, error rate, and capacity metrics.
• Validate assumptions and refine the numbers with your own business judgment.
• Finalize the business case in Notion with BA Commentary on where human judgment shaped the model.

Generate the initial business case with Claude

Executives approve projects based on projected return, not technical elegance. Your business case needs to connect every AI capability from the To-Be design to a specific, measurable improvement.

• Switch back to your Meridian Logistics BA project in Claude.
• Start a new chat within the project.
• Prompt Claude to build a quantified business case for the AI exception handling system. Ask it to structure the output as: (1) Executive Summary with one-paragraph value proposition, (2) Current-State Cost Analysis quantifying the cost of manual exception handling using the scenario metrics, (3) Projected Improvements broken into three categories (cycle time reduction, error rate improvement, capacity gains), (4) Implementation Cost Estimate with phased investment, and (5) ROI Summary with projected payback period.
• Ask Claude to reference the specific numbers from the scenario: exception resolution currently takes 48-72 hours, routing accuracy sits at 85%, and each CS rep handles approximately 40 tickets per day.

Review Claude's output. You should see a structured business case with projected numbers for each improvement area.

What should I see?

Claude should produce a multi-section document with specific projections: cycle time dropping from 48-72 hours to under 4 hours for the majority of cases, routing accuracy improving from 85% to approximately 95%, and each agent handling more tickets without additional headcount.

If the output is vague or missing specific numbers, ask Claude to "quantify every claim with a specific percentage or dollar figure based on the scenario metrics."

Why four metrics?

The business case focuses on four metrics that align with how consulting engagements measure AI value: cycle time reduction, error/misclassification rate improvement, capacity/throughput gains, and CSAT improvement.

Each metric traces directly back to a bottleneck you identified in the As-Is diagram and a solution you designed in the To-Be process. That traceability is what makes the business case credible.

Validate assumptions and refine the numbers

Claude generates plausible-sounding projections, but plausible is not the same as defensible. Your job as a BA is to pressure-test every assumption.

• Challenge Claude on the cycle time reduction claim. Ask: "What percentage of exceptions can realistically be resolved in under 4 hours? Not all exceptions are equal. Break this into tiers by complexity."
• Push back on the routing accuracy target. Ask Claude: "What is a realistic accuracy ramp? The system won't hit 95% on day one. Show me a 3-month, 6-month, and 12-month projection."
• Ask Claude to add implementation costs. Prompt: "Add a cost section that includes: model training/fine-tuning costs, integration development, change management and training, and ongoing monitoring. Estimate a phased 12-month implementation budget."

Review the refined output. The business case should now show a more realistic ramp-up curve rather than instant results.

• Apply your own judgment to the capacity gains section. Consider: if the AI handles 70% of routine exceptions automatically, the remaining 30% (complex cases) still require specialists. Adjust the headcount impact accordingly.
• Verify that every improvement metric traces back to a specific capability in your To-Be design. If the business case claims a benefit that your requirements document does not support, remove it.
• Ask Claude to produce a final, clean version incorporating all your refinements. Review it one more time for internal consistency.

Why challenge the numbers?

Claude defaults to optimistic projections because the training data is full of marketing materials and case studies that report best-case outcomes. A Senior BA delivers defensible estimates, not aspirational ones.

The phased ramp-up you added (3-month, 6-month, 12-month) reflects how AI systems actually perform: they improve over time as they encounter more data and edge cases.

Finalize the business case in Notion with BA Commentary

• Return to your Notion page and scroll to the Business Case section.
• Paste the refined business case from Claude into this section.
• Format the document using Notion's heading blocks for each major section (Executive Summary, Current-State Cost Analysis, Projected Improvements, Implementation Costs, ROI Summary).
• Add a callout block at the bottom titled "BA Commentary" with the following content:

Your BA Commentary should address three points:

• Which assumptions Claude generated that you accepted and why (e.g., the 70% auto-resolution rate aligns with industry benchmarks for similar classification tasks).
• Which numbers you adjusted and the reasoning behind your changes (e.g., added a phased ramp-up because AI systems require training data accumulation).
• What you would validate with the client before presenting this business case (e.g., actual CS rep hourly cost, current monthly exception volume, existing SLA commitments).

Connect your artifacts

Your business case is strongest when it explicitly references the other four artifacts. Mention the As-Is bottlenecks by name, point to the To-Be automation zones that deliver each benefit, and cite the acceptance criteria from your requirements document as the success measures.

Before you move on, predict what your completed Business Case section should contain. Then scroll through it in Notion.

You should see a structured document with an ROI model showing three quantified improvements: exception resolution time dropping from 48-72 hours to under 4 hours for 70% of cases, routing accuracy improving from 85% to a target of 95%, and increased ticket volume handled without additional headcount. The BA Commentary callout should be visible at the bottom.

Business case feels too vague?

If your business case reads like a generic AI pitch instead of a Meridian Logistics-specific document, check that every claim references a specific scenario metric (the 48-72 hour resolution time, the 85% routing accuracy, the 40 tickets per rep). Ask Claude to "ground every projection in the Meridian Logistics scenario data."

Help me make my business case more specific to the scenario

All five artifacts are now complete. Your Notion page tells a complete story: from messy stakeholder input, through documented current-state pain, into a precise future-state design with guardrails, and finally a quantified justification for the investment. Next up, you can take on the Secret Mission to add a UAT test plan that proves your AI system actually works as specified.

Secret mission

Build a UAT Test Plan for AI Components

AI systems fail in ways that look like confident success. A fluent response can still be wrong, unsafe, or non-compliant. Can you design a UAT test plan that catches what traditional testing misses?

Clean Up Your Resources

Clean Up Your Resources

This project runs entirely on free-tier tools and your existing Claude Pro subscription, so there are no ongoing costs to worry about. Decide whether to keep your artifacts for your portfolio, or remove them entirely.

Resources you used:

• Claude Project: "Meridian Logistics BA" (in claude.ai).
• Lucidchart diagrams: As-Is Exception Handling Swimlane and To-Be Exception Handling Swimlane.
• Notion page: "Meridian Logistics - AI Exception Handling" with all six artifact sections.

Keep everything running

No action needed. Choose this if you plan to continue building on these artifacts or share them as part of your portfolio.

• Your Notion portfolio page is ready to share with potential employers or attach to job applications.
• You can export your Lucidchart diagrams as PNG or PDF for inclusion in slide decks or PDF portfolios.
• Your Claude Project retains all chat history and uploaded context for future reference.

Pause - I'll come back to this later

There is nothing to pause in this project. All artifacts are stored in free cloud services with no running processes or compute costs.

• Your Notion page, Lucidchart diagrams, and Claude Project will remain available whenever you return.
• Consider exporting your Lucidchart diagrams as PNG before stepping away, in case you want offline copies.

Delete - I don't want to use this again

Remove all project artifacts from your accounts. Follow each step below to fully clean up.

• Delete the Claude Project: Open claude.ai, go to the Projects page, find "Meridian Logistics BA", click the three-dot menu in the upper-right corner of the project tile, select Delete, then confirm by clicking Yes, delete.
• Delete the Lucidchart diagrams: Open lucid.app, click Documents from the left-hand navigation, select each diagram (As-Is and To-Be), click the three-dot More menu at the bottom-right corner of the thumbnail, and select Move to Trash. To permanently delete, open Trash from the left-hand navigation, click Empty trash in the top-right corner, then select Delete Forever.
• Delete the Notion page: In Notion, hover over "Meridian Logistics - AI Exception Handling" in the left sidebar, click the ••• menu that appears, and select Delete. The page moves to Trash and is automatically permanently deleted after 30 days.

Nice Work!

Nice Work!

You just produced a complete set of AI-enabled Business Analysis deliverables for a realistic consulting engagement. Your Notion portfolio page now showcases the exact artifacts a Senior BA at a data consultancy delivers to clients.

You've learned how to:

• Build As-Is and To-Be swimlane process diagrams in Lucidchart with color-coded automation boundaries, human-in-the-loop gates, and projected performance improvements at each bottleneck.
• Write AI-specific requirements using the four-layer pattern (functional outcomes, permissions/constraints, quality thresholds, safe failure) with Given/When/Then acceptance criteria and a guardrails catalog defining what the AI must never do.
• Execute the AI-enabled BA workflow end-to-end: prompt Claude with structured templates, validate critically against business context, refine with human judgment, and finalize portfolio-ready artifacts for stakeholders.
• Secret Mission: Designed a UAT test plan for AI components including edge-case scenarios, adversarial guardrail-violation tests, and a go-live validation checklist.

Ready to quiz yourself?