TutorialsThe Groundwork: Four Concepts and One Formula
State machine, data flow, routing rules and automation boundary in plain language — then the heart of the lesson, the six-step breakdown formula, unrolled step by step.
Why these four concepts first
The biggest obstacle to breaking down a process is not drawing the diagram — it is describing it. You carry a complete thing in your head, yet what comes out is one sentence: "a customer reports a fault, we send someone to look." The details in between — who triggered it, where the conclusion is stored — are so familiar that they never get said.
The four concepts are four rulers for measuring what your description is missing.
State machine: what this object counts as right now
The identities a work object passes through from birth to death. A requirement, for example: pending → in progress → awaiting confirmation → done. Note it is not a list of steps; it answers what the thing counts as at this moment — and the state decides who can move it forward next.
Data flow: what each step reads, writes and stores
Without data flow, automation breaks at step two — because nobody knows where step one's conclusion lives. Which table, which field, who reads it. All of it gets written down.
Routing rules: which branch this message should take
Something arrives at one entry point; how does the system decide which path it follows? A confirmation goes to the approval flow, a new request goes to registration — mix them up and tickets cross over.
Automation boundary: what work AI can take
The test is simple: pure information handling and rule-based judgement can be automated; creativity, business decisions and physical operations stay with people. Settle the boundary before discussing tools.
One line to remember the order: settle the states (what it counts as), then the data (where it is written), then the branches (which path it takes), and only then who does what.
Six common scenarios
Start with how this method works in situations you already know.
| Scenario | What the work actually is | How AI helps |
|---|---|---|
| Onboarding a new hire | Turning tacit knowledge into a handover-ready SOP | You narrate your experience, AI structures it into a step table |
| Cross-team collaboration | Design → development → testing → launch, with several handovers | It draws the state machine and pins down each step's inputs, outputs and waiting points |
| Customer tickets | How complaints and requests get triaged, assigned, followed up and closed | It defines the routing rules and the escalation paths |
| Approvals and expenses | Submit → multi-level approval → payment → archiving | It checks for missed returns, timeouts and delegated approvals |
| Content production | Topic selection → draft → review → revision → publish | It marks which stages AI can draft directly and which need human review |
| Procurement and suppliers | Enquiry → comparison → order → receipt → reconciliation → payment | It surfaces hidden branches such as repeated orders or rejected deliveries |
One formula (the heart of the lesson)
The six-step breakdown formula: trigger → state machine → step table → data flow → boundary questions → automation levels
| Step | What you do | Output | In one sentence |
|---|---|---|---|
| 1. State the trigger | Who / when / which event starts it | A one-line trigger condition | "Someone @-mentions the bot with an image in the group chat" |
| 2. Draw the state machine | Every state from birth to death, plus the arrows | State list + transition diagram | Pending → in progress → done |
| 3. Break out the step table | Trigger, input, action, output, executor for each step | One table | What S1 did, and who uses its output |
| 4. Map the data flow | Which table, field and location each step reads and writes | Data pipeline | Status in the main table, images in an attachment field |
| 5. Probe the boundaries | Concurrency, failure, timeout, missing values, permissions… | Exception handling list | "What if two people confirm at the same time?" |
| 6. Mark automation levels | Fully automated / semi-automated / manual per step | Responsibility matrix | AI generates the image, a person picks the supplier |
The order cannot be changed. The most common failure is jumping to step 6 — "I want AI to handle my tickets" — while the states and branches are still undefined, so what AI takes over is a fog. Steps 1 to 4 turn the fog into structure, step 5 prevents the crash, and only step 6 is about tools.
What each step is really asking
The general prompt formula
Every step of the six means talking to AI, and the skeleton is always the same:
Role + task + context + format + constraints
- Role: whose position it speaks from (business process analyst / QA engineer / AI adoption consultant);
- Task: what this round must produce;
- Context: your raw description, the previous round's output;
- Format: state machine as an arrow diagram, fixed six columns for the step table, branches written as "yes → path X";
- Constraints: no more than 10 steps, waiting points annotated with duration, sorted by severity.
The single most valuable line
Add this to the end of a prompt:
Before you start, ask me the three questions you think I am most likely to have overlooked.Its value is that it forces AI to surface what you did not know you had left unsaid. The process in your head is incomplete, and you cannot feel the gaps — so the best use of AI is not to execute, but to question.
The principle behind the whole lesson: AI produces the first draft and hunts for gaps; you judge and decide.
What it gives you is "a structurally complete candidate", not "your process". Which parts do not match reality, which rules will not fly in your company — only you know. That is exactly where your value lies.
Key points from this lesson
| Remember this | Because |
|---|---|
| States are nouns, actions are verbs | Mix them up and you cannot draw the state machine |
| A step with no describable output can usually go | The six columns are a sieve, not decoration |
| Data flow decides whether the process can run itself | If step one's conclusion is not stored, step two cannot connect |
| Do not exhaust the boundary cases alone | Having AI sweep them from a QA angle is far cheaper than firefighting |
| Classify first, then prioritise | Trying to automate everything at once lands nothing |
Breaking Down Workflows with AI
A six-step breakdown formula, 3 hands-on prompts and 5 templates — turn one fuzzy sentence into a runnable SOP in 15 minutes.
Hands-On: Three Prompts That Open the Process Up
Working through a customer repair ticket as the main case, three prompts in relay produce the state machine, step table, boundary list and automation assignment — with each prompt in full and the one thing to watch.