Leveraging Claude for Business Analysis Work on an Enterprise Project
- Gleb Bryskin
- 14 hours ago
- 7 min read
In an enterprise project, an AI setup helps business analysts gather requirements, decompose them into user stories, add acceptance criteria, and more. Here’s the AI setup that saves us time and effort.
BY GLEB BRYKSIN & ANASTASIA EVSTRATOVA
Surface-level AI Use
AI has been around for a while and has proven extremely helpful to IT teams. However, the use of AI tools was inconsistent across our team: developers, QAs, and business analysts used AI autonomously, without sharing common knowledge or a shared toolset.
That may not sound like an issue, but it was. Without a unified approach, the team’s use of AI was significantly less efficient than it could have been, so the BAs continued to face the familiar issues: continuous scope creep, requirements drift, fully manual ticket creation, etc.
We wanted to fix that, not by creating a specific AI setup solely for business analysts, but by using an all-in-one solution for the whole team.
Single AI Tool for the Whole Team
The goal for our team was to set up an AI-assisted software development lifecycle that the whole team will use: designers, BAs, developers, and QAs. Each team would use its own agent, tailored to its needs and grounded in the shared context.
Business analysts needed a toolset to analyze feature requests, decompose features, draft requirements, create tickets, and more.
Developers would benefit from a toolset that would help with implementation and code review.
QA engineers would use AI to create test cases and drive automated tests.
UI/UX designers could leverage AI tools to speed up prototyping.
The value would lie in the common knowledge base, so any user story, any code snippet, and any test won’t contradict each other.
AI Setup We Built for BAs
Since this article is focused on business analysis, we’re going to describe only this part of the AI SDLC setup we got:
Let’s dive into some details:
workspace.py is a single source of truth that generates and validates the entire workspace.
Artisyn SDK handles a shared catalog of skills for the whole team.
BA-agent is a Claude-based tool that invokes BA skills, such as ba-story-authoring, grooming, create-tasks, and many more.
The knowledge base is shared across the whole team and is based on multiple sources, such as project documentation, Azure DevOps Wiki, and codebase.
MCP Connectors are used by ba-agent to access Azure DevOps and Figma.
One Agent Doing BA’s Legwork
Let’s look closer at the ba-agent. It’s a Claude-powered assistant to business analysts on our project. It takes three inputs:
Knowledge base
Figma
Azure DevOps
The ba-agent handles the entire BA routine: grounds every user story in the knowledge base, reads the user interface from Figma, analyzes features in Azure DevOps, decomposes them into user stories, drafts requirements as INVEST stories, and writes them directly into Azure DevOps. Let’s focus on how we use the ba-agent in our daily business analysis activities.
Clickable Prototypes for Requirements Gathering
Grounded in the knowledge base, the ba-agent knows the project well enough to generate clickable HTML prototypes in minutes. This has proven to be a powerful tool for requirements-gathering sessions with stakeholders, helping to collect their feedback and facilitate brainstorming. Best of all, clickable HTML prototypes built by Claude help BAs and stakeholders detect what’s missing in the original requirements – the blind spots and unspoken requirements not mentioned in any of the documents.
Context-based Requirements Decomposition
The BA-agent can decompose features into INVEST user stories using one of the patterns: by user action, by role, by data, by platform, and so on. First, it creates a user story map for BA review. At this point, Claude doesn’t draft the requirements and will do so only after a BA confirms the decomposition.
For example, if there’s a feature request to enable customers with pets to change flights, the following tickets can be created by the ba-agent:
Platform | User story |
web | Change seats in booking with pet |
web | Change flights in booking with pet |
mob | Change seats in booking with pet |
mob | Change flights in booking with pet |
Feature → user story map
The decomposition always requires a BA review and confirmation before the BA agent proceeds to the next steps.
Adjustable AC Format
Our ba-agent is fine-tuned to write the acceptance criteria in different formats – Gherkin scenarios or bullet-point acceptance criteria, depending on the team’s preferences. Before drafting user stories, Claude asks which format to follow.
BULLET-POINT Happy path — list updates after a new in-app booking Precondition: the logged-in user commits a new in-app booking.Trigger: they open the Trips → Upcoming tab.Expected: the booking appears with no manual refresh; ordered by departure date; cache updated. | GHERKIN Happy path — same behaviour Given the logged-in user has committed a new in-app bookingWhen they open Trips → UpcomingThen the booking appears, no refreshAnd it is ordered by departure date |
The BA agent can be updated further. In the example above, instead of conventional bullet points, an alternative bullet-point format is used. Adjustability of the BA agent enables BAs to change how it works.
After BA review and confirmation, Claude automatically creates the tickets in Azure DevOps, and the agent-created tag is added to all such tickets. Before authorizing Claude to add tickets, BAs fine-tune Claude’s output, and manual adjustments can be made to the tickets in ADO afterward.
Faster Refinements
The AI setup also allows BAs to speed up refinements. The team plays planning poker to provide estimates in story points. For our project, it’s also necessary to provide an estimate of development and QA effort, in hours, for the subtasks. During refinement, the team estimates the effort in hours, and then, after authorization from the BAs, Claude creates the subtasks.
Dev task | 24h |
QA task | 10h |
Estimated effort | 34h |
Also, Claude updates the ticket status to Ready for Dev after refinement, letting team members know which user stories can be added to sprints.
Three More BA Work Accelerations
The work of BAs isn’t just user stories – a lot of time is spent on managing requirements, prototypes, and sign-offs. Three extra skills help BAs do it faster:
Update stories from calls – a call transcript runs through the meeting-notes skill, then gets matched against a specific user story, and then Claude lands edits directly in Azure DevOps. An updated version of a ticket is available in no time instead of spending time on manual work.
Prepare prototypes for brainstorming – very often stakeholders mistake clickable HTML prototypes for the to-be system. Therefore, Claude helps BAs downgrade the quality of the clickable prototypes so that stakeholders don’t get confused during the requirements workshops and other sessions.
Generate the sign-off document – a special skill was created to generate requirements documents for sign-off. It pulls the feature and linked user stories, analyzes the original request, and generates an HTML file for pasting straight into Loop.
Claude Drafts, BA Approves
So how do we draw a line between what Claude does and what the BAs do? In our project, it turns out to be simple: Claude generates; the business analyst makes decisions and approves. The AI agent is lightning-fast at creating a first version of BA artifacts, but nothing gets published without human supervision, review, and confirmation.
Activity | Claude | Business analyst |
Requirements gathering | Builds clickable HTML prototypes | Runs sessions, brings the context, defines the scope, gives feedback to Claude |
Requirements decomposition | Proposes a user story map following a chosen pattern | Confirms or reshapes the decomposition before anything is published |
User stories authoring | Drafts user stories with ACs in the agreed format | Reviews user stories, requests changes |
Ticket creation | Writes the confirmed tickets to Azure DevOps and tags them | Authorizes publishing of user stories and adjusts the tickets in ADO |
Scope refinement | Creates Dev / QA subtasks and moves states to Ready for Dev | Facilitates the session and triggers subtask creation |
Sign-off artifacts generation | Generates the requirements document for review | Shares the sign-off documents with the stakeholders and tracks progress |
Who does what in our AI SDLC
Even though it seems that Claude does everything on its own, the BA still owns the process and decisions. Nothing gets done by AI without human supervision and review: at every step of the BA workflow, the business analyst has an upper hand and controls what the AI does. Only this combination of human-AI effort ensures efficiency of the AI SDLC.
Applicable to Legacy and New Functionality
One worry with any AI setup is whether it’s tailored to enterprise or it can also handle greenfield projects. In the context of an enterprise project, some new modules or applications might be added from scratch in parallel to existing web or mobile applications.
For us, this isn’t a problem as the AI setup is perfectly capable of building new products. For example, if we need to design a new application that will be partially connected to existing apps, the same ba-agent and the same coding/QA agents can be used to achieve great results. Such a task has been set for our project – the ba-agent proved to be a powerful tool that helped analyze the request, create clickable HTML prototypes, prepare the requirements, and obtain sign-off.
Does It Actually Pay Off?
Since the AI setup has been launched quite recently, here are some statistics over a month:
44 | 135 | ~5 |
user stories authored, with acceptance criteria | Dev / QA tasks created, decomposed and estimated | BA-days saved (estimate) |
The stats come from Azure DevOps on the project. Taking a user story as an example, the average time to create it is roughly 45 minutes, whereas with Claude's assistance, it falls to around 12 minutes. What may seem like not much per single user story can add up to substantial time and cost savings.
The statistics above show impressive time savings from using the AI in everyday BA activities. One might think that there isn’t much left to humans – Claude handles everything. That isn’t the case: a human stays in the loop. Claude drafts, decomposes, and creates prototypes, but it’s the business analyst who sets tasks, reviews, and approves. Every single deliverable requires a thorough review from a BA.
Advice for Other BAs
If you’re looking to benefit from AI on your project, you don’t need the same setup to get good results. Prompt Claude to create skills and tailor them to your needs, or re-use the BA skills from another team, then point Claude to the tickets (be it Azure DevOps or JIRA/Confluence) to create a project context – and you will be able to boost your BA performance.
AI adoption is in its early stages, so don’t hesitate to experiment and try new approaches to build a setup tailored to your team's needs.
Art of Business Analysis training schedule
News and articles on business analysis:
