How to Map Your Business Processes for AI Automation

What to Capture That Standard Process Documentation Misses

• All input sources and formats — not just the primary format, but every variant that exists in practice
• All output destinations — where does the result go, and in what format does each destination expect it?
• Decision logic — for every decision point in the process, what information is used and what are the rules?
• Exception patterns — what are the most common exceptions? How are they currently handled? How often do they occur?
• Edge cases known to practitioners — the scenarios that break the standard process at least once a month
• Quality signals — how does a practitioner know when they have done this correctly vs when something needs more attention?
• Tacit knowledge — what do experienced practitioners do differently from the documented procedure?

How to Extract Tacit Knowledge: The Interview Approach

The most important source of process knowledge is the practitioners who do the work every day, not the managers who own the process. Practitioners know the exceptions, the workarounds, and the informal rules that never made it into any documentation.

Interview technique: sit with a practitioner while they process real work. Ask them to narrate what they are doing and why. Focus specifically on: "What do you look for that tells you this one is different?" and "What do you do when X happens?" — these questions surface the tacit knowledge that determines whether AI will handle exceptions correctly.

Assessing Automation Readiness from the Process Map

• High readiness: inputs are consistent and well-structured, decision rules are explicit and documented, exception rate is low (<5%), output format is defined
• Medium readiness: inputs vary in format but follow known patterns, most decision rules are documented with manageable exceptions, output format is adaptable
• Low readiness: inputs are highly variable or unstructured, decision rules are largely tacit, exception rate is high (>20%), output format varies by recipient
• Not automatable now: process depends on relationship context, real-time negotiation, or judgement that cannot be formalised — address the formalisation gap first

The Documentation That Makes Automation Feasible

The output of a process mapping exercise for AI automation should be a structured document covering: input taxonomy (all input types and variants with examples), decision tree (all decision points with rules and data sources), exception catalogue (all known exceptions with current handling), output specification (format, destination, validation rules), and quality criteria (what does correct look like?). This document becomes the specification for the AI system and the ground truth for evaluation.