Sharper, Not Smaller: The Business Analyst in the Age of AI

Day 4 | Building AI Into Your BA Workflow, Step by Step

We have the principle, and we have both lists. Today we put it to work. This is the part I get asked about most, because building AI into a real workflow is where good intentions either turn into results or quietly fall apart. The mistake almost everyone makes is treating AI as a vending machine. You type a request, take whatever comes out, and move on. The output is generic, the quality is uneven, and the promised time saving never quite arrives. The fix is to stop thinking of it as a machine you query and start thinking of it as a step in a process you have designed, with a deliberate handoff on either side. The rule that holds the whole thing together is simple. At every step, the tool drafts and the human decides. Keep that handoff clean, and you get the speed without losing the judgment. Blur it, and you are back in Monday's meeting with no answer to the why. I have built versions of this on several live projects now, and the pattern holds every time. The teams that get value are not the ones with the cleverest prompts. They are the ones who decided, in advance, exactly where the tool stops and the human starts. The handoff is the whole design. Everything else is detail. Here is how it runs across a typical project lifecycle.

Discovery

Before workshops, ask the tool to generate a thorough question set for the process you are exploring, then edit it hard against what you actually know about this organisation. It is a check against your blind spots, not a script. During sessions, keep your full attention on the people and let a tool handle the recording. The reading of the room stays yours entirely, as we covered yesterday. Afterwards, have the tool produce a first-pass summary of the transcript so you reach the real reading faster.

Handoff: the tool drafts questions and summaries. You decide what to ask, what matters, and what the silence in the room meant. One warning on discovery. Do not let a generated question set narrow your curiosity. It is a floor, not a ceiling. The best questions you ask will be the ones that come to you in the room, in response to something a person said and the tool could never have anticipated.

Requirements

Take your messy discovery notes and have the tool shape them into a first draft in your standard template. This kills the blank-page problem. Then you do the work that counts: checking each requirement against what you heard, finding contradictions, surfacing assumptions nobody stated, and resolving ambiguity by going back to people, not to the tool. The draft is clay. The shaping is yours.

Handoff: the tool drafts and structures. You decide what is true, what is in scope, and why. This is the step where the artefact-and-accountability distinction earns its keep. The draft will look finished. It will be formatted, confident, and complete-looking. That polish is exactly the trap. A good-looking draft built on contradictions you have not yet resolved is more dangerous than a messy one, because it tempts you to stop thinking. Treat the polish with suspicion.

Analysis

For process work, sketch a first-pass map or have the tool turn a written description into a draft diagram, then walk it step by step with the people who live the process. They will tell you which steps are real, which are theatre, and which exist for reasons buried in history. For data, let the tool suggest what to examine and sketch the questions worth asking, then bring the business meaning yourself, because it cannot.

Handoff: the tool produces a starting structure. You apply the context, the history, and the judgment about what it means. Process maps are where this pays off fastest. A tool will draw you a clean, logical flow. The real process is rarely clean or logical, and the gaps between the diagram and the reality are usually where all the interesting problems hide. Use the draft to start the conversation, not to end it.

Validation

Ask the tool to generate first-pass test cases and acceptance criteria from each requirement. It will give you the routine scenarios fast. Your job is the cases it could never imagine: the project-specific edge conditions, the regulatory traps, the weird real-world inputs you have seen break systems before. Use it to clear the obvious so you have attention left for the dangerous.

Handoff: the tool covers the predictable. You add the cases only experience reveals. A pattern I have noticed is that the tool is excellent at the cases everyone remembers to test and useless at the ones that actually break systems in production. Those edge cases come from scars, from having watched a particular kind of input cause a particular kind of failure. You carry those scars. The tool does not. That is why validation is a partnership, never a handover.

Handover and communication

When you need the same content in three registers - a technical spec, a plain-language summary for the business, a short note for an executive - have the tool produce the variations from your approved source. Reformatting across audiences is mechanical, and it is genuinely useful here. But read every version before it goes out, because the moment it carries your name, you own it.

Handoff: the tool reformats and rephrases. You own the message and every word that leaves under your name.

Two safeguards that keep this honest

First, never let a tool touch confidential information it should not see. Know what your organisation permits, which tools are approved, and where the data goes. Stakeholder trust includes guarding what people told you in confidence, and feeding it into an unapproved system breaks that trust whether or not anyone finds out.

Second, build a read-before-send reflex and never break it. Every piece of AI-assisted output gets read by you, line by line, as though you wrote it. If you cannot explain or defend a sentence, it does not go out. This single habit is the difference between AI that makes you sharper and AI that sets you up for a silent room.

Start small

Do not rebuild your whole workflow this week. Pick one step. Summarising discovery notes is a good first one: low risk, high time saving, easy to check. Run it for a fortnight until the handoff feels natural, then add the next step. Built this way, deliberately and one handoff at a time, AI stops being a novelty you occasionally poke at and becomes a quiet part of how you work. And here is the reassuring part. Built this way, you never have a single point where the tool is trusted with something it should not be. Every handoff is small, checkable, and reversible. You are not betting your reputation on a black box. You are using a fast assistant for the mechanical parts and keeping your hands firmly on every decision that carries weight. That is what makes it sustainable rather than a gamble you get away with for a while.

Tomorrow we close the loop. Why all of this leaves the analyst role sharper rather than smaller, and how to position yourself so the people who decide your next move can see it. I am bringing a simple tool to make it stick.

Go out and be successful.

Oluwatosin Ogunkoya  |  Flotog BA Insights  |  www.flotogbainsights.com