Problem-First AI Workflow Decomposition
Start with a costly task, break it down, then assign AI selectively.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
This framework reverses the common tool-first approach to AI adoption. Instead of opening ChatGPT and asking what it can do, identify one recurring business task that feels slow, expensive, or heavy. Document the task as a miniature standard operating procedure, including its component sections, decisions, research needs, and production steps. Then evaluate each step separately to determine where AI can generate ideas, conduct research, create an outline, draft material, or organize information. A human remains responsible for judgment, personalization, and final approval. Repeating the workflow creates a measurable improvement loop: compare time and output quality, refine the instructions, and gradually automate more of the process. Rick’s newsletter illustrates the mechanism, falling from roughly five hours to 30–45 minutes per edition.
Origin
Rick Mulready demonstrated the method through his weekly AI newsletter on The Jasmine Star Show, explaining how he decomposed each section before deciding where AI could help.
Core principles
- 01Begin with a business problem rather than an AI tool.
- 02Focus on one heavy or time-consuming task at a time.
- 03Decompose the task before choosing where AI belongs.
- 04Keep human review in every production workflow.
- 05Improve the workflow through repeated use and feedback.
How to run it
- 1
Choose One Bottleneck
Select one recurring task that takes too long, costs too much, or consistently creates friction. Keep the initial scope narrow rather than attempting to transform the whole business.
Pro tip Use a task with a clear recurring output, such as a newsletter, podcast package, or report.
Watch out Starting with an AI tool instead of a defined problem encourages unnecessary or ineffective experimentation.
- 2
Map the Existing Process
Write every action, decision, input, and section involved in producing the output. Treat this as a miniature SOP.
Pro tip Document the process as it is actually performed, not as you imagine it should work.
Watch out Skipping hidden steps will produce incomplete prompts and unreliable outputs.
- 3
Assign AI by Step
Determine where AI can help generate ideas, research, summarize, outline, draft, classify, or organize. Do not assume the entire task should be delegated at once.
Pro tip Begin with low-risk, time-intensive steps that are easy to review.
Watch out End-to-end delegation without checkpoints can amplify errors and generic output.
- 4
Provide Instructions and Context
Give AI the objective, relevant source material, output expectations, and constraints for each assigned step. Reuse successful instructions in future cycles.
Pro tip Structure instructions as clearly as an SOP written for a new team member.
Watch out A vague request such as writing a newsletter about marketing will usually return generic copy.
- 5
Apply Human Judgment
Review the output, correct inaccuracies, add personal insight, and align the result with the business’s voice and standards. Treat AI as a production partner rather than the final publisher.
Pro tip Identify recurring edits and incorporate them into the next version of the instructions.
Watch out Publishing an unreviewed AI draft risks factual errors, weak differentiation, and loss of trust.
- 6
Measure and Iterate
Compare the new workflow’s completion time and output quality with the previous process. Refine one weak step at a time until the result becomes dependable.
Pro tip Track both time saved and the percentage of the output that survives human editing.
Watch out Abandoning the workflow after one poor attempt prevents the iterative gains Rick describes.
In the wild
Rick divided his newsletter into sections such as filtered AI news, personal insights, and a use case of the week. He mapped the research, topic selection, outlining, and writing steps, then used AI selectively across them while retaining human editing and approval.
→ The weekly production time fell from approximately five hours to 30–45 minutes.
A business can map the steps that follow publication of a podcast or YouTube video, then let AI extract platform-specific ideas and drafts while a person performs quality assurance and scheduling.
→ One source asset can support multiple channels without recreating every post manually.
Common mistakes
Starting with the Shiny Tool
Opening a popular AI product and searching for something to do with it reverses the useful sequence. Define the business problem before selecting the tool.
Using One Vague Prompt
A broad request omits the task structure, context, standards, and desired outcome, so the resulting copy is likely to feel generic.
Treating the First Draft as Final
AI output still requires human judgment, editing, personalization, and approval. Copying it directly into a public channel removes the differentiating human element.
Is it for you?
Best for
Business owners who want to reduce the time spent on a repeatable content, administrative, or operational task.
Not ideal for
One-off tasks whose process is unknown, unstable, or too consequential to delegate partially to AI.
From the transcript
“they look at the tools, they look at ooh, ChatGPT, ooh, Claude, ooh, Gemini or whatever.”
“So I'm just breaking each thing that I need to do down into the steps and then figuring out, okay, where can I. Help me…”
“And then there's always the human element, always.”
From the episode
What 7-Figure CEOs Know About AI That You Don’t with Rick Mulready
Rick Mulready