PTENES
MODULE 1.2

🧭 Method — How to Decide

From “I should automate this” to “this is what I’m going to build” — a 6-step framework for making automation decisions clearly and without waste.

6
Topics
40
Minutes
🟢
Beginner
🧭
Method
EAD PIPELINE E Eliminate first A Automate if there's anything left over D Delegate complex/judgment Don’t automate waste AUTONOMY SPECTRUM L0 Manual L1 Suggested L2 Drafted L3 Supervised L4 Autonomous Default = the lowest level that works · Increase only after proving the previous level

Detailed content

1

Find the Constraint

Where every intelligent automation decision starts

Before building anything, you need to know where the pain is. Don’t automate what’s on the wrong path. Find the real bottleneck — or the real opportunity — using two questions that force your brain out of reactive mode.

⚡ TWO POWER QUESTIONS
❶
"If 500 new customers arrived tomorrow, what would break first?"
→ Points to the operational bottleneck. This is where automation has the highest ROI.
❷
"What would get you 500 customers tomorrow?"
→ Points to the blocked growth opportunity. It may be more valuable than fixing the bottleneck.
BOTTLENECK
What breaks as you grow
OPPORTUNITY
What enables growth
PRIORITY
Constraint before optimization
FOCUS
One constraint at a time
2

EAD — Eliminate, Automate, Delegate

The sequence that separates smart operators from those who automate waste

The order matters. Many people jump straight to "automate" without asking whether that task should exist. The result: you spend time building a machine that does something useless more efficiently.

✓ Correct sequence
  • ✓Eliminate first: "What if we simply stopped doing this?"
  • ✓If no one would notice → kill the task
  • ✓Automate later what was left
  • ✓Finally, Delegate what’s complex or requires human judgment
✗ Common pitfalls
  • ✗Automating without asking whether the task should exist
  • ✗Delegate to AI what should be eliminated
  • ✗Automating workflows that change every week
  • ✗Delegate critical judgment to an AI without supervision
💡 Golden Rule

Don’t automate waste. An unnecessary task done with AI is just an unnecessary task done faster. The value lies in eliminating it first.

ELIMINATE
Always the first question
AUTOMATE
What’s left and repeatable
DELEGATE
Complex, variable, judgment-based
SEQUENCE
The order is the method
3

60/30/10 Golden Rule

The realistic distribution of any mature AI operation

Full automation is rarely the real goal — and anyone promising 100% is selling you something. Smart operations have three distinct tracks that coexist, and deciding which task goes in which track is the most important decision before you build.

60/30/10 RULE — OPERATIONAL DISTRIBUTION
60%
Fully Automated
Runs on its own, without human input. Trigger → Output. Examples: daily reports, FAQ responses, standardized order processing.
30%
AI-Assisted (human reviews before it goes out)
AI drafts, a human approves. Draft + Review. Examples: high-value proposals, sensitive communications, decisions with external impact.
10%
Manual — Intentionally
Unique human judgment, strategic relationships, original creativity. Not worth automating or cannot be safely automated.
💡 Practical Calibration

Map your 10 most frequent tasks and classify each into one of the 3 tiers. If more than 50% fall into “manual” without a clear reason, you have a level 1 opportunity. If more than 80% are in the fully automated tier, review whether you have enough supervision.

60% AUTO
Runs without input
30% ASSISTED
Draft + Review
10% MANUAL
Human judgment
BALANCE
100% auto = false promise
4

Map the Process (5 elements)

If you can’t explain it to a person, you can’t explain it to AI

Before opening any automation tool, you need to be able to describe the process from start to finish using 5 elements. This map is the minimum spec for any automation that will work.

T
Trigger — What starts it?
The event that triggers the process. An email arrives, a form is submitted, a specific time, an external system webhook. No clear trigger = process can’t be automated.
D
Data Sources — Where does the data come from?
Spreadsheet, CRM, database, API, email. Each source needs to be accessible to the automation. Data trapped in PDFs or in people’s heads creates a bottleneck here.
X
Data Transformations — What happens to the data?
Calculations, formatting, enrichment, classification, AI text generation. Each transformation is a discrete logical step. If you can't list them, the process is still too vague.
D
Decision Points — Where are the choices?
If X, then Y. If the amount is over R$5k, escalate to a human. Complex or poorly defined decision points are where automations fail silently.
F
Destination — Where does the output go?
CRM updated, email sent, file created, notification triggered, database entry added. The destination determines whether the automation succeeded.
💡 Mapping Test

Explain the process out loud to someone who doesn't understand AI. If you get stuck, there's a vague point. Move on only when you can narrate the 5 elements without pausing.

TRIGGER
What it starts
DATA SRC
Where it comes from
TRANSFORM
What changes
DECISION
Where you choose
DEST
Where it goes
5

Autonomy Spectrum (L0→L4)

Workflows beat agents — move up a level only after proving the previous one

There is a temptation to jump straight to autonomous agents (L4) because it seems more impressive. That choice is almost always wrong. The correct default is the lowest level that solves the problem. Move up only after proving that the level below worked and showed its limits.

AUTONOMY SPECTRUM — L0 → L4
L0
Manual
You do everything
No automation. Documented process, executed by a human. Mandatory baseline before any automation.
L1
Suggested
AI suggests
The system proposes an action; a human decides. Examples: an email classified as spam (you accept it or not), a suggested reply in the CRM.
L2
Drafted
AI drafts
AI produces a complete draft; a human reviews and approves it before sending. Most business LLM use cases fit here.
L3
Supervised
AI executes, human monitors
Automation runs, and a human can intervene. Workflow with exception alerts. Requires a well-mapped process and clear escalation criteria.
L4
Autonomous
Agent decides and acts
The agent makes decisions and takes action without human intervention. High risk if not preceded by validated L0→L3. Last resort, not the default.
✓ Right approach
  • ✓Default = the lowest level that works
  • ✓Prove L2 before moving up to L3
  • ✓Workflows beat agents (prefer deterministic)
  • ✓Document escalation criteria at L3
✗ Frequent errors
  • ✗Starting at L4 "because it seems more powerful"
  • ✗Skip steps without validating the previous level
  • ✗Using an agent where a workflow is enough
  • ✗L4 without defined intervention criteria
DEFAULT = MINIMUM
Smallest level that works
WORKFLOWS FIRST
Deterministic > agent
SCALE UP SLOWLY
Prove it before scaling
L4 = LAST
Not the starting point
6

Tie It to a KPI

If your automation doesn’t move a number, why build it?

Every automation must justify its existence with a real number. "Because it’s cool" isn’t a business case. Without a KPI, you don’t know if it worked, can’t prioritize the next one, and can’t defend the time invested.

📊 The 3 KPI Buckets
📈
Bucket 1 — More Customers
• Lead conversion rate
• Response speed to new contacts
• Number of proposals sent
• Customer acquisition cost
💎
Bucket 2 — More Value per Customer
• Average project value
• NPS / satisfaction
• Upsell rate
• Time to deliver to the client
⚙️
Bucket 3 — Lower Costs
• Manual hours per week
• Process error rate
• Processing time
• Cost per unit delivered
✓ VALID KPI CHECKLIST
□ Which of the 3 buckets does it belong in? (clients / value / cost)
□ Is there a baseline number today? (e.g., "currently takes 4h/week")
□ Is there a specific target? (e.g., "reduce to 30min/week")
□ Is there a deadline for measuring? (e.g., "in 30 days")
□ Can you measure it without extra effort?
💡 The Question to Ask Before Building

"If I build this and it works perfectly, what specific number will change?" If you can’t answer in one sentence, take a step back and reframe the question. Building without a KPI is work without direction.

BUCKET 1
More customers
BUCKET 2
More value per customer
BUCKET 3
Lower cost
"IT’S COOL" ≠ KPI
Move one number

🧭 Module Summary

✓
Find the Constraint — 2 power questions reveal where the pain and opportunity are. Start with the real constraint.
✓
EAD in the right order — Eliminate → Automate → Delegate. Never automate waste.
✓
60/30/10 Rule — Mature operations live in all three ranges. 100% automation is a false promise.
✓
Map the Process (5 elements) — Trigger · Data Sources · Transformations · Decision Points · Destination. Define a minimum spec before building.
✓
Autonomy Spectrum L0→L4 — Default = the lowest level that works. Move up only after proving it. Workflows beat agents.
✓
Tie It to a KPI — Choose the bucket (clients / value / cost) and a specific metric. No KPI, no build.

Next Module:

1.3 — Machine (How to Build and Operate) 🛠️ — From decisions to actual implementation: stacks, tools, and the continuous operations cycle.

← Back to Track Next Module →