How to map a business process before you automate it

It means writing down the process as it really runs, not as the onboarding doc or the manager's mental model describes it. Pick one recurring piece of work, an invoice getting approved, a lead moving from enquiry to quote, a new hire getting set up, and walk it step by step with the two or three people who actually do it. You are capturing the as-is, the version with the copy-paste from three different spreadsheets and the email someone always has to chase, not the to-be version you wish existed. Automation built on the wish list breaks the first time reality does not match it.

For every step, note five things: what triggers it, what goes in, what comes out, who owns it, and whether a decision or a judgement call happens there. A sticky note or a plain spreadsheet works fine for this; you do not need diagramming software for a process under twenty steps. The trigger and the owner matter as much as the task itself, because automation needs to know exactly when to start and who gets pulled in when something looks wrong.

Write down the exceptions too, the order that arrives with no PO number, the client who always replies by phone instead of email. Most processes look clean until you hit the cases that do not follow the happy path, and those exceptions are usually where an automated version quietly fails first.

Once the steps are on paper, look for the ones that are repetitive, rule-based and high-volume, the same input producing the same shape of output every time: pulling fields off an invoice, sorting an inbox by request type, drafting a first-pass reply to a common question. Those are strong candidates. Steps where someone is exercising judgement, approving a discount, choosing a vendor, deciding how to handle an angry client, are not, no matter how repetitive they feel to the person doing them. The map tells you which is which; without it you are guessing from a demo.

RAND Corporation interviewed 65 experienced AI and machine learning practitioners about why AI projects fail, and named a misunderstood or miscommunicated problem as the leading cause: a project built on a version of the work that does not match how the work actually happens. A process that has never been mapped is exactly that kind of misunderstood problem. Automating it does not fix the copy-paste, the chasing, the manual workaround, it just runs the mess at higher speed and makes it harder to see where the mess is. Map first, simplify what the map shows you, and only then decide what gets automated.