Why mapping comes first
"Use AI on this process" is not an instruction you can act on — a process is a dozen different steps with different data, risk and judgement. You have to make the work visible first. A process map breaks the whole into discrete, assessable steps; only then can you judge each one on its own merits. Skipping the map is how organisations automate the glamorous step and leave the actual bottleneck untouched.
The right question: automate, augment, agent
"Can AI do this?" is the wrong question because it's binary and the answer is almost always "partly." The useful question has three answers, in increasing order of autonomy:
- Augment — AI assists; a person stays in the loop and decides. Low risk, fast to adopt. (Drafting, summarising, suggesting.)
- Automate — AI performs the step; a human reviews the output. Suited to high-volume, rule-shaped work with clear right answers.
- Agent — AI runs the step end to end and acts, within guardrails. Reserved for well-understood, low-exception, reversible steps.
A step-by-step method
- Map the process into discrete steps with one action each.
- Score each step on the dimensions below — not the process as a whole.
- Assign a tier (augment / automate / agent) from the score.
- Check the data and the risk — is the input data available and clean? Is a mistake reversible and cheap, or irreversible and costly?
- Sequence the candidates by value and readiness, and start with the surest.
Scoring a step for readiness
A step's readiness isn't a feeling — it's a few concrete factors, which is exactly what the CREATE framework scores: cognitive complexity, repetition and volume, the evidence and data available, how well automation aligns to the step's goal, the transformation impact, and the exception rate. A step that is repetitive, data-rich, low-complexity and low-exception is ready for a high tier; a step that is judgement-heavy, sparse on data, and full of exceptions stays human — or gets augmented, not automated.
Sequencing: don't boil the ocean
The fastest way to fail is to try to automate everything at once. Rank the ready steps by value × readiness and do the surest, highest-value one first. Early wins fund and de-risk the harder ones, and a working automation on a real step teaches you more than a year of planning. This sequencing — surest and most valuable first — is exactly what a transformation roadmap is for.
Describe a process. Get the whole analysis.
LeanBean turns a plain-language description of any process into a professional map — with Lean waste, SIPOC, design-thinking, AI-readiness and capacity analysis built in, and a Transformation Coach that sequences the fixes. Free to start.
Build your first map free →Keep learning in the Garage
Short, practical lessons on Lean, process mapping and AI-ready operations — the same thinking that powers LeanBean, in two minutes or less. Follow along and build the habit.
Common questions
How do you assess a process for AI readiness?
Map the process into discrete steps, then score each step (not the whole process) on factors like complexity, volume, data availability and exception rate. Assign each a tier — augment, automate or agent — check the data and the risk, and sequence the candidates by value and readiness.
What's the difference between augment, automate and agent?
Augment means AI assists while a human decides; automate means AI performs the step and a human reviews; agent means AI runs the step end to end within guardrails. The right tier depends on the individual step's complexity, volume and risk — one process can have steps at all three tiers.
Why map the process before applying AI?
Because 'use AI on this process' isn't actionable — a process is many steps with different data, risk and judgement. Mapping breaks it into discrete, assessable steps so you can judge each on its merits, instead of automating the wrong one.
What makes a step hard to automate?
A high exception rate. If 'it depends' is the honest answer to most cases, the exceptions are the real work — and that's the part automation handles worst. Repetitive, data-rich, low-exception steps automate well; judgement-heavy, sparse, high-exception steps stay human or get augmented.
Related guides: What is AI readiness? · The CREATE framework · Process map to transformation