m2ai-portfolio Workflow Decomposer
Takes a function, role, or process description and outputs a list of discrete, scoreable workflows — the atomic unit required before any AI build/buy/hire/wait decision can be made.
When to trigger
- User says "workflow decomposer", "decompose this role", "break this function into workflows", "what are the workflows in X"
- Before scoring workflows for AI investment decisions
- When auditing a team, department, or product for AI opportunity mapping
Phase 1: Accept input
Accept one of:
- Function or role description ("Account Executive", "Invoice Processing", "Customer Onboarding")
- Process narrative — a paragraph describing how work gets done
- Job description or process doc — paste or file path
If the input covers multiple clearly distinct functions, ask which to decompose first. Keep scope to one function at a time for clean output.
Phase 2: Decompose into atomic workflows
Break the function into discrete workflows. A workflow is atomic when it:
- Has a clear trigger (what starts it)
- Has a clear output (what it produces)
- Can be evaluated independently (success/failure is observable)
- Would be assigned to a single person or role in a non-AI org
Apply two decomposition rules:
- Split on decision type: Routine execution vs. judgment calls are separate workflows
- Split on frequency tier: Daily/weekly/monthly recurrences are separate workflows even if structurally similar
Aim for 5–15 workflows per function. If the list exceeds 15, group into sub-functions first.
Phase 3: Annotate each workflow
For each workflow, add:
| Field | Description |
|---|---|
name |
Short label (verb + noun, e.g., "Process invoice exceptions") |
trigger |
What kicks it off |
output |
What it produces |
frequency |
How often it runs (daily / weekly / ad-hoc / event-driven) |
avg_duration |
Estimated time per instance |
data_in |
Key inputs (systems, file types, human context required) |
data_out |
Key outputs (reports, records, decisions, messages) |
Phase 4: Output
## Workflow Map: [Function Name]
Total workflows identified: N
| # | Name | Trigger | Output | Frequency | Duration | Data In | Data Out |
|---|------|---------|--------|-----------|----------|---------|---------|
| 1 | ... | ... | ... | ... | ... | ... | ... |
...
### Decomposition Notes
[Any workflows that couldn't be cleanly split — flag for follow-up]
[Sub-functions if the original scope was too broad]
### Suggested next step
Run a six-dimension investment scorer on these workflows to classify each into automate / build / buy / hire / wait.
Rules
- Never merge a judgment-heavy workflow with a routine-execution workflow — they score differently on every investment dimension
- If frequency is unknown, mark as
unknown— do not assume. Frequency is the single most important scoring input - Output must be a clean table — no prose paragraphs for individual workflows
- Do not recommend actions yet — this phase is decomposition only
Source
Nate's Newsletter, 2026-05-17 — "Executive Briefing: Stop asking if AI can do this. Start asking what shape the work is." Reframes AI investment decisions as workflow classification before capital allocation.