Map the as-is before you design the to-be.

Welcome to Issue #04.
Every Saturday I ship one field-tested lesson from live BA × AI delivery — the stuff I actually use on client work this week, not repackaged theory.

A few weeks ago a client handed me a beautiful process diagram and said "this is how it works." I spent one afternoon talking to the three people who do the work, and none of them followed it. They had a shadow process — a spreadsheet here, a WhatsApp approval there, a step everyone skips because the system is slow. The diagram described a company that doesn't exist. If I had redesigned from that diagram, I'd have shipped a fix for a problem nobody had. You can't improve a process you haven't actually seen. Here's how I use AI to see it fast.

Reconstruct the real process

Three moves that turn scattered, contradictory inputs into an honest as-is map — usually in an afternoon, not a week of workshops.

1. Feed the mess in, ask for the steps out

I collect the raw material as it actually is — call notes, a couple of emails, screenshots of the tool, the one Excel everyone secretly runs on — and paste it in with a structured prompt: "From these inputs, reconstruct the current process as a numbered sequence of steps. For each step note the actor, the trigger, the system used, and the output." You're not asking for an opinion; you're asking the model to compress a pile of fragments into a spine you can react to. The first draft will be wrong in interesting ways — that's the point.

2. Make it mark every gap and guess

The dangerous map is the one that looks complete. So I follow up: "Flag every step where the inputs don't tell you what actually happens, and every place you inferred a connection rather than being told it." Now the map has honest holes in it — and those holes are your interview questions. I walk into the next call with "what happens between approval and invoicing?" instead of "so, walk me through your process" — and I get a real answer in two minutes instead of twenty.

3. Separate the official path from the real one

Ask the model to lay the documented process next to the reconstructed one and list where they diverge. Every divergence is a signal: a workaround means the official step is too slow, a skipped control means nobody believes it matters, a manual re-key means two systems don't talk. The gap between the diagram and the behaviour is where the actual work is. That list — not the pretty diagram — is what I bring to the redesign conversation.

What NOT to do

• Don't let the AI design the to-be yet. Its job here is to describe reality, not improve it. A model that jumps to "here's a better process" buries the messy truth you needed to see under a clean recommendation.
• Don't trust the map until a human who does the work confirms it. The reconstruction is a hypothesis. It gets you to a sharp first draft in an hour — but the person at the desk is the only one who can tell you it's right.
• Don't skip the shadow steps because they're awkward. The spreadsheet nobody mentions in the meeting is usually the most important node in the whole map. Chase it.

Try this week

Take one process you're about to change. Gather whatever raw inputs you already have — notes, a screen recording, one email thread — and ask AI to reconstruct the current steps and flag its gaps. Bring that gap list to the next conversation instead of a blank page. You'll find at least one step that isn't on any official diagram — and that's the one worth your attention.

See you next Saturday.
— Dibya Mishra, Senior Business Analyst