How should an analyst start using AI at work?

Start with tasks where you own the input and can verify the output in under a minute: turning a half-formed doubt into a precise question, drafting a stakeholder email that asks for a specific decision, converting workshop notes into decisions and owned actions, and rewriting a vague ticket into something testable. Then build a project context pack so every later prompt is grounded, and agree the guardrails before connecting an assistant to any real system.

In what order should an analyst adopt AI tooling?

Context pack first, because everything depends on grounding. Then guardrails, because the controls must exist before the capability. Then the meeting pipeline, which pays back immediately at low risk. Then reading the codebase, then read-only Jira and Confluence through MCP, then your notes vault. Then the delivery artifacts: test plan, API collections, reporting. Write access to systems of record comes last, once you know how often the assistant is wrong about your specific data.

What is MCP and why does it matter for analysts?

The Model Context Protocol is an open standard for connecting AI assistants to external tools and data through one common interface. For analysts it turns Jira and Confluence from things you copy out of into things an assistant can read directly, acting as the account that authorized it and therefore respecting existing permissions. Connect read-only, scoped to one project key and one space, on a dedicated service account.

What data should never be pasted into an AI assistant?

Five classes: identified customer data such as names, account numbers, IBANs and transaction references; credentials including API keys, tokens and connection strings; unreleased commercial material such as pricing and contract drafts; personal data about colleagues; and production extracts of any kind, even ones you believe are anonymous. Mask by preserving the shape of the data rather than blanking it, and confirm your organization approved tooling before pasting anything.

Will AI replace business analysts?

It makes drafting, enumeration and summarisation cheap, which were always the least valuable and most time-consuming parts of the role. What grows in value is verification against the real system, judgment about which risks matter, knowledge of undocumented behavior that no document contains, and accountability for a release decision. An analyst whose contribution was typing up other people conclusions is exposed; one who can prove what a system does is worth more than before.

Learning path

The AI Analyst

The AI Analyst is a twelve-part path for working analysts, from your first prompt to an AI-connected delivery week. It starts with the four tasks that pay back on day one, questions, emails, workshop follow-ups and vague tickets, then builds the project context pack and the guardrails. From there it connects a model to your real systems: the codebase, Jira and Confluence through the Model Context Protocol, your meetings, and your own notes. It ends with the artifacts you are accountable for: a complete test plan with traceability, an API collection gating merges in CI, and the weekly numbers you can defend.

Written by Ahmed at Analyst Engineering, a Senior Technical Business Analyst with 10+ years in banking and payments. Every prompt in the series is one I run on real delivery work.

The whole series is one loop

Strip the tooling and every part of this path is one of five verbs applied to a different artifact. Learn the loop and it survives the next three tool releases.

1. Ground. Give the model your material before it answers: the pack, the schema, the contract, the vault.
2. Read. Use it to read what you cannot read at scale: a codebase, 140 tickets, a 90 minute transcript.
3. Verify. Check every claim against the system. A file and a line, a ticket key, a query you read.
4. Write. Let it produce only what is mechanical and reversible, under per-action approval.
5. Prove. Thread identifiers through requirements, conditions, cases, and tests so coverage is demonstrable.

The core series, in 12 parts

If you prefer one continuous read, the twelve parts run in order from your first prompt to the full operating system. The stages below add companion articles at each level.

Free account

Get the next part when it ships

A free account, no password: an email link signs you in. It saves your progress on the Labs, unlocks the solutions, and, if you tick the box, sends you new practitioner playbooks as they publish.

Your email is used to sign you in. Nothing else, unless you ask. Privacy.

Stage 1

Prompt: get useful output, safely

For you if: You have opened a chat window, been unimpressed, and stopped. Or you use it daily for emails and suspect there is more.

By the end: You can turn a doubt into the right question, drive an email to a decision, structure a workshop follow-up, and ground every prompt in a project pack instead of hunting for documents. You know exactly what never goes into a prompt.

Stage 2

Connect: point it at your real systems

For you if: Your prompts are grounded and you are tired of copying things out of Jira, Confluence, and the repository to paste them in.

By the end: You can read a codebase you did not write, connect an assistant to Jira and Confluence read-only through MCP, let it write back under approval, turn every meeting into artifacts within the hour, and query fourteen months of your own notes.

Stage 3

Deliver: build the artifacts that ship

For you if: The plumbing works and you want it pointed at the things you are actually accountable for: the test plan, the suite, the numbers.

By the end: You can build a full test plan with a traceability matrix, generate an API collection with real assertions behind a CI gate, write SQL you can defend, automate the weekly pack, and describe the whole system in an interview.

If you only do one thing

The 90 day build order

DAYS 1-14 Build the context pack and agree the rules of engagement.
DAYS 15-30 Run the meeting pipeline on every workshop, published within the hour.
DAYS 31-45 Read one repository, connect Jira read-only, start the vault.
DAYS 46-70 One test plan end to end, then the collection in CI.
DAYS 71-90 Automate the weekly pack, then enable write access last.

Go further

Guides and 1:1 coaching

The full playbooks behind this path, and a direct line if you want help wiring it into your own programme, whatever stage you are at.

AI at Work: MCP, RAG, and Agents The Technical Skills Guide for BAs Book a 1:1 coaching call

22 articles across three stages.

Free account

Practice on real artifacts

The Labs give you a fictional payments platform with an OpenAPI contract, Kafka events, logs and a database to work through. Exactly the material this workflow is built to chew through, and a free account saves your progress.

Your email is used to sign you in. Nothing else, unless you ask. Privacy.