🔁 Repetitive, standardizable tasks
AI creates real value where there is repetition and patterns. The more a task follows the same step-by-step process, with predictable inputs and outputs, the more naturally it lends itself to automation. Without a pattern, there's nothing you can reliably teach the machine.
🧱 The 3 signs of a task that can be standardized
- •Predictable input: input data always has the same format.
- •Stable rules: the "how to do it" doesn't change with each run.
- •Clear output: you can describe the expected result in one sentence.
✓ Standardizable
- ✓Classify emails by subject.
- ✓Extract data from invoices.
- ✓Generate meeting summaries.
✗ Cannot be standardized
- ✗Negotiating a sensitive contract.
- ✗Decide on a career strategy.
- ✗Resolve a team conflict.
The same every time.
Stable input and output.
Happens often.
Fits in one sentence.
💵 Automation ROI: time × frequency × cost
ROI is the filter that separates opportunity from hobby. The calculation is simple and powerful: the time time spent per run, multiplied by the frequency, multiplied by the cost per hour. The result tells you whether it’s worth automating—and how long it will take for the setup to pay for itself.
🧮 The ROI formula
value = time_per_run × annual_frequency × hourly_cost
# example
0.5 h × 250 times/year × R$ 60/h = R$ 7.500/year
# payback (if setup costs 8 h ≈ R$ 480)
480 / 7500 ≈ ~3 weeks to pay for itself
📊 How to read the number
- High frequency is what boosts returns the most — prioritize what happens often.
- Short payback period (weeks) = automate now. Long payback (years) = rethink it.
- Hourly cost includes your time: it isn’t free.
🎯 Practical tip
Create a simple spreadsheet with three columns—time, frequency, cost—and one row per task. Start with the automation that has the highest annual savings and the shortest payback period.
Hours per run.
Times per period.
Value per hour spent.
When the setup pays for itself.
🧩 Jagged intelligence: where AI excels and where it fails
AI’s intelligence isn’t uniform—it’s jagged (jagged). In some tasks, it outperforms a human by far; in others, surprisingly simple ones, it fails badly. Understanding the shape of this sawtooth is what keeps you from trusting AI with something it does poorly.
✓ Where AI shines
- ✓Summarize, rewrite, and translate text.
- ✓Classify and extract information.
- ✓Recognize patterns in language.
- ✓Generate drafts from examples.
✗ Where AI fails
- ✗Count and perform exact calculations.
- ✗Following rigid logic without deviation.
- ✗Knowing facts it wasn’t given.
- ✗Ensure it will never make things up.
🚨 Attention: the silent failure
The danger of AI isn’t getting something wrong — it’s getting it wrong with confidence. It delivers a wrong result with the same certainty as a right one. So, for “high-stakes” tasks (calculations, critical facts), always add a check before relying on it.
Language and standard.
Counting and rigid logic.
Fails with confidence.
Check before you trust.
🚫 What NOT to automate: judgment, exceptions, risk
Knowing when to say “not this” is as valuable as knowing what to automate. Three categories should stay out of full automation: tasks involving judgment, full of exceptions unpredictable, or with risk irreversible error.
Judgment
Decisions that depend on values, context, and good judgment
Firing someone, setting strategic priorities, making a diagnosis. AI can help organize the information, but the final decision is human.
Unpredictable exceptions
Rare cases that never follow the pattern
If half the cases are “exceptions,” there’s no pattern to automate. Automate the common case and route unusual ones to a human.
Irreversible risk
Actions you can’t undo
Transfer money, delete data, send public communications. At a minimum, require human approval before taking action here.
🎯 Practical tip
Ask the regret question: "if this goes wrong on its own, how much damage will it cause?". Major, irreversible damage = don’t automate without a human in the loop.
The decision stays with the human.
No standards, no automation.
Irreversible actions require approval.
What to leave out.
🧑⚖️ Human-in-the-loop vs. fully autonomous
Not everything is “automate everything” or “do it by hand.” Between those two extremes is the human-in-the-loop: AI prepares the work, but a human approves it before the sensitive action. Knowing where to put this approval point gives you the confidence to automate much more.
✓ Human-in-the-loop
- ✓AI drafts, human reviews and approves.
- ✓Ideal for sensitive or public-facing actions.
- ✓Builds trust gradually.
⚙ Fully autonomous
- •Runs from start to finish without stopping.
- •Ideal for low-risk, high-volume tasks.
- •Requires verification and solid logs.
🎚️ Autonomy is a dial, not a switch
Start with more supervision and ease off as the automation proves it’s reliable:
- •Level 1: the AI only suggests; you do everything.
- •Level 2: the AI executes, but asks for approval.
- •Level 3: the AI acts on its own in low-risk cases.
Approval before the action.
Runs nonstop.
Release as confidence grows.
Chooses the right level.
🧮 Criteria for prioritizing opportunities
You’ll have more automation ideas than time to build them. Prioritizing well ensures your first automation delivers visible results—and funds the next ones. Four criteria can organize any list: return, effort, risk, and frequency.
📐 The return × effort matrix
Cross return (vertical axis) with effort (horizontal axis) and tackle the quadrants in this order:
- 1.High return · low effort: the quick wins. Start here.
- 2.High return · high effort: large projects, do this after the first win.
- 3.Low return · low effort: only if there's time left.
- 4.Low return · high effort: ignore.
🏆 Why start with the quick win
The first automation needs to prove its value quickly. A quick win delivers results in days, builds confidence, and frees up time to tackle bigger projects. Starting with a long project is a recipe for giving up before you see a return.
🎯 Practical tip
Score each opportunity from 1 to 5 for return, effort (inverted), risk (inverted), and frequency. Add them up. The highest-scoring one is the next to build—and that’s exactly the task Trail 2 will teach you to audit.
A gain that generates.
How hard it is to build.
What could go wrong.
Quick result first.
✅ Module Summary
Next Module:
1.3 — The 5-step map: identify, map, design, build, commercialize.