Many people think they don't know how to use AI, so they scramble to learn tools, hoard prompts, and bookmark every Skill they find. But the real problem is usually not that AI isn't capable — it's that you haven't decomposed your business, goals, and judgment for it. You tell AI "write me some copy," and it can indeed write — but what comes out mostly isn't what you wanted, because you never told it what kind of copy you want, what process stands behind it, or how you define the outcome.

This tutorial makes a single point: how to turn the experience you can't quite articulate into a structured Skill that AI can execute.

What Decomposition Really Is: Turning Feel into Structure

Decomposition, compressed into one sentence: turn feel into structure.

Here's an awkward scenario everyone has lived through: you're reading an article, someone asks what you think of the headline, you glance at it and say "this headline doesn't work." They press: "why not?" — and you're stuck. All you can say is "call it instinct. I've done this for years, and my gut tells me it doesn't work."

Your judgment is probably right — you do have the experience. But you can't articulate why. Very few people can actually take that experience apart: which specific piece of experience made you think it doesn't work? Did it fail to hit a pain point? Or did it fail to build contrast and tension? Without that process, your experience stays stuck at the "gut feel" level — impossible to pass on, impossible to sell, and impossible to feed into AI.

The core move of decomposition is forcing yourself to answer that "why." You think it doesn't work — fine, keep asking: why doesn't it work? And that "because," pushed one level further down, is because of what? Layer by layer, you squeeze the vague feeling into a clear structure.

What Decomposition Actually Breaks Down: Goal → Action → Judgment

Once you know the essence, the next question is what exactly gets broken down. The order matters — get it wrong and everything falls apart:

Step 1: Define the Goal

What is this thing actually supposed to achieve? This step is the one most often skipped, and it's the most critical. Break down actions before defining the goal, and everything you produce is wasted effort.

Step 2: Break Down Actions Based on the Goal

To achieve this goal, what actions are needed? Which factors influence it?

Step 3: Make the Judgment Explicit

Within each action, what's the standard for good versus bad? This is the step most people miss. How well does each action need to be done to count as good? On what grounds do you say this step was done right? These judgment standards normally run silently in your head — you have to dig them out one by one and write them down.

Tip: If judgment isn't made explicit, it can't be passed on and can't be fed into AI. Before decomposing any task, force yourself to write one sentence first: "The goal of this task is ___." If you can't write it, don't break anything down yet.

Practicing with the GMV Formula: Turning "Mysticism" into Controllable Variables

Take a formula anyone who has run a business knows: GMV = Traffic × Conversion × Average Order Value.

Everyone can recite the formula, but the value isn't in reciting — it's in breaking it down further. Let's zoom in on the "Traffic" factor and walk it through goal → action → judgment:

Layer one: Traffic itself is a goal. What determines traffic? Take short-video hits as an example: the influencing factors are the opening, the completion rate, interaction (likes and comments), and topic selection.

Layer two: Pick "completion rate" and break it down further. What determines completion rate? Three metrics: the opening, content value, and information density.

Layer three: Pick "the opening" and keep going. How do you define a good opening? How do you make the opening better? Keep decomposing until the granularity is fine enough to act on directly.

GMV = Traffic × Conversion × Average Order Value
      │
      ┌────────┴────────┬──────────┬─────────┐
    Opening       Completion   Interaction   Topic
                    rate
                      │
               ┌──────┼──────┐
            Opening  Content  Information
                     value     density
                      │
   How do you define a good opening? How do you make it better? ← Break it down to this level and you can act on it directly

Before decomposition and after decomposition are two completely different states:

Four Core Concepts: SOP, Workflow, Prompt, Skill

What you produce through decomposition ultimately lands in three forms:

SOP-Style Skill: Like an Intern

Write the "how to" as steps: fixed, no thinking required. It suits work where the steps are both fixed and tedious: expense reports, pulling periodic reports, batch-uploading videos.

Workflow-Style Skill: Like a Senior Operator

Add judgment at every step — it can judge, and it can redo work. It suits complex decision scenarios: writing a hit piece of copy, building a hit account from zero.

Tip: Same task, two logics, two Skills. Use SOP for repeated execution, workflows for complex decisions. Where do the judgment criteria come from? Your experience is the judgment criteria.

A Practical Method for Writing Your Business as a Skill

Feed It Experience Documents — You Can Write Skills Without Writing Prompts

You don't need to write prompts from scratch. Feed AI the experience documents you already have (operating manuals, retrospective notes, judgment criteria checklists) and let it distill them into a Skill structure for you.

Position the Skill at 70 Points, Not 100

Don't expect a Skill to nail it in one shot. A Skill gives you a 70-point baseline; the remaining 30 points come from iterating as you use it.

Make the Skill Generic: Base It on Material Types, Not Industries

If your Skill only works in one specific industry, its reusability is very low. Design the Skill around the type of material (copy, video, images) rather than the industry, and it can be reused across industries.

Business Skills Need Quantifiable Goals

"Write good copy" is not a goal; "write a promotional post with a click-through rate no lower than 5%" is. A quantifiable goal gives the Skill a standard for judging the quality of its own output.

Break Complex Tasks into Multiple Skills + One Orchestrating Skill

Don't try to solve everything with a single Skill. Break the complex task into multiple sub-Skills, then use one orchestrating Skill to coordinate them.

FAQ

• "What if the thing I broke down doesn't execute well in AI?" First check whether you skipped the "make the judgment explicit" step. In most cases, poor AI performance comes from unclear judgment criteria, not from AI being weak.
• "I don't have experience — can I still decompose?" Yes. Start with AI contrast training: work through the task yourself first, then hand it to AI, and see where the differences lie. The difference points are exactly the judgment criteria you need to fill in.
• "How fine-grained should the decomposition be?" Until the granularity is fine enough to act on directly. If you've decomposed something and still don't know what the next step is, it isn't fine enough.
• "What's the real difference between a Skill and a prompt?" A prompt is a one-time instruction; a Skill packages the capability for a whole class of tasks. A prompt is "help me write a piece of copy"; a Skill is "you are my copy assistant — every time I give you material, you produce copy following this process."

Use Cases

• Content creators: break down your hit-headline judgment criteria into a Skill, and AI-written headlines go from 50 points to 90
• Operations staff: break down social-media promo copy and community management scripts into reusable Skills
• Business managers: break down core business processes into Skills so both your team and AI can execute them
• Personal brands: "productize yourself," turning years of experience into Skills that can be delivered and scaled

Core advice: the next time you think "AI is so dumb," don't blame it first — go back and check whether you clearly broke down the goal, the process, and the requirements. Treat "breaking things down for AI to hear" as a deliberate practice, not a stopgap.