Every few weeks we’re asked to “add AI” to a workflow that nobody in the company can fully describe. Someone owns the beginning, someone else owns the end, and the middle lives in three inboxes.
Automating an unmapped process doesn’t remove the chaos — it encodes it. The most valuable step in any automation project costs nothing: writing down what actually happens today.
Map before you automate
A workflow map needs only three columns: the step, who performs it, and what system it touches. The surprises appear immediately — approval loops nobody remembers approving, data re-entered three times, a spreadsheet that turns out to be the real system of record.
Step Owner System ----------------- ---------- -------------- Receive request Support Shared inbox Check eligibility Support CRM + spreadsheet Draft response Support Email Approve Manager Email (verbal!) Update records Ops CRM, manually
Once the map exists, automation candidates rank themselves. High volume plus low judgment equals automate now. High judgment plus high error cost equals assist, don’t replace — generate a draft and keep the human approval.
If a step’s approval happens verbally or in chat, that’s not a blocker — it’s the finding. Formalizing that gate is usually worth more than the automation itself.
A scoring model that works
We score each step on three axes before writing any code:
- Volume — how many times per week does this step run?
- Judgment — could a well-briefed new hire do it on day one?
- Error cost — what happens if it’s done wrong once?
- Data readiness — does the input exist digitally, anywhere?
Projects scoped this way ship in weeks and survive contact with reality. Projects scoped as “add AI to operations” produce demos. If you’re considering automation, start with the map — it’s the cheapest consulting you’ll ever not pay for.