6.1: When Humans Need to Review

Welcome to Part 6! So far, you've built automations that run completely on their own. But what about when you need a human to check something before it happens?

In this section, you'll learn when and why to add human review to your workflows—and how it actually makes your automations better, not slower.

---

6.1.1: The Power of Human + AI

Here's the thing: Complete automation isn't always the goal.

💡 Think of It Like This

Imagine a sous chef in a restaurant:

• The sous chef preps all the ingredients (fast, reliable)
• The head chef does final plating and taste-checks (expertise, judgment)
• Together, they serve amazing food quickly

That's what human-in-the-loop automation looks like!

AI handles the repetitive, time-consuming work. You focus on the decisions that really need human judgment.

You get:

• ⚡ Speed: AI does 95% of the work instantly
• 🛡️ Safety: You review the 5% that matters
• 🎯 Quality: Best of both worlds

Let's see how this works in practice.

---

6.1.2: Why Human Review Matters

Not convinced you need review? Let's look at three big reasons:

Why Human Review Matters

Combining automation with human oversight gives you the best of both worlds

AI Is Powerful, Not Perfect

AI can make mistakes, especially with edge cases, sarcasm, or unusual requests.

Example: Customer asks "Can you cancel my subscription... NOT!" - AI might miss the sarcasm.

High-Stakes Decisions Need Judgment

Some actions have serious consequences if done incorrectly.

Example: Processing refunds, legal agreements, or financial transactions.

Learn From Human Corrections

When humans review and correct AI outputs, the system improves over time.

Example: First 100 responses reviewed → patterns emerge → improve AI instructions.

The Golden Rule

Automate the repetitive work, keep humans for judgment calls.

Let AI handle 95% of routine tasks automatically, so you can focus your energy on the 5% that really needs your expertise.

💡 The Bottom Line

AI is your super-powered assistant, not your replacement.

Use AI for speed and scale. Use humans for judgment and accountability.

---

6.1.3: When Should You Add Review?

Not every automation needs human review. Here are the most common scenarios where you definitely should add it:

Common Review Points

When should you add human review to your workflows?

Before Sending Important Emails

Always review

Examples:

• •Customer-facing messages
• •Marketing campaigns
• •Legal or compliance emails
• •Anything with your brand name on it

Before Making Purchases

Always review

Examples:

• •Processing payments
• •Creating invoices
• •Issuing refunds
• •Any financial transaction

When AI Is Uncertain

Flag for review

Examples:

• •Low confidence score (<80%)
• •Unusual customer requests
• •Multiple possible answers
• •Complex edge cases

First 100 Responses (Training Period)

Review initially

Examples:

• •New automations need tuning
• •Learn what works and what doesn't
• •Improve AI instructions
• •Build confidence before full automation

💡 Pro Tip: Start with more review, then gradually reduce as your automations prove reliable. You can always add it back if needed!

A Simple Rule of Thumb

Ask yourself: "If this goes wrong, how bad would it be?"

• ✅ Really bad? → Add review
• ✅ A little embarrassing? → Add review for now, remove later
• ✅ No big deal? → Let it run automatically

💡 Start Careful, Get Faster

When you first build an automation, it's smart to review everything. After you see it working correctly 50-100 times, you can start removing review steps.

You can always add review back if something changes!

---

6.1.4: Types of Reviews You Can Add

There are several ways humans can oversee automation. Pick the one that fits your needs:

Types of Reviews

Different ways humans can oversee automation

Approve/Reject

Simple yes or no decision

How It Works

AI generates output → You check a box → Action happens (or doesn't)

Best For

When AI output is usually good and needs quick approval

Example use cases: Email drafts, social media posts, data categorization

Edit Before Sending

AI provides a starting point, you polish it

How It Works

AI creates draft → You edit and improve → Edited version is used

Best For

When AI gets close but needs human touch

Example use cases: Customer responses, marketing copy, reports

Flag for Exception Handling

Mark unusual cases for special attention

How It Works

AI tries to handle → Flags if uncertain → Human handles exceptions

Best For

Complex workflows where most cases are routine

Example use cases: Support ticket routing, fraud detection, order validation

Provide Feedback

Rate AI performance to improve over time

How It Works

AI performs task → You rate quality → System learns from feedback

Best For

Training period or ongoing quality improvement

Example use cases: Response quality ratings, categorization accuracy, sentiment analysis

✨ Mix and match: You can combine multiple review types in the same workflow. For example, use "Edit Before Sending" for important emails AND "Provide Feedback" to improve over time!

💡 Which Type Should You Use?

Not sure which review type to start with?

• Need speed? → Approve/Reject (fastest)
• Need quality? → Edit Before Sending (most control)
• Complex workflow? → Flag for Exception Handling (scales best)
• Just learning? → Provide Feedback (improves over time)

You can always change or combine them later!

---

6.1.5: How Review Workflows Work

Let's see the basic pattern you'll use in the next sections:

How Review Workflows Work

Automation stops at checkpoints for your approval

AI Generates

Automation runs

⚡ Auto

→

You Review

Check the output

👤 Manual

→

You Approve

Give the OK

👤 Manual

→

Action Happens

Final step runs

⚡ Auto

Speed of Automation

• ✓ AI handles repetitive work instantly
• ✓ Drafts ready in seconds
• ✓ No manual data entry

Safety of Human Review

• ✓ Catch mistakes before they go out
• ✓ Add personal touch
• ✓ Full control and accountability

📧 Real Example: Email Campaign

AI writes personalized emails for 100 leads (saves you hours) → You review and approve in 15 minutes → Only approved emails send → Perfect balance of speed and safety!

The Steps Explained

Here's what happens in a typical review workflow:

1. AI Generates (Automatic)

• New row appears (from trigger, form, or manual entry)
• AI columns run automatically
• Draft or recommendation is created
• ✅ No human needed yet

2. You Review (Manual)

• You open the table and see the draft
• Read through what AI generated
• Decide if it's good enough
• ✅ This is where your judgment matters

3. You Approve (Manual)

• If it looks good, check an "Approved" box
• If it needs work, you can edit or skip it
• Only approved items move forward
• ✅ You're in control

4. Action Happens (Automatic)

• Once approved, the final action runs
• Email sends, record creates, etc.
• Everything tracked in the table
• ✅ Automation finishes the job

💡 See How This Helps?

Imagine you have 50 customer emails to respond to:

Without AI:

• Write 50 emails from scratch = 3-4 hours
• ❌ Exhausting and slow

With Full Automation:

• AI writes and sends 50 emails = 5 minutes
• ❌ Fast but risky (what if AI makes mistakes?)

With Review Workflow:

• AI writes 50 emails = 5 minutes
• You review and approve = 20 minutes
• ✅ Total time: 25 minutes with full control!

You just saved 3 hours while staying safe.

---

6.1.6: Real-World Example: Email Approval

Let's make this concrete with a real scenario you might face:

The Scenario

You run customer support and get 30-50 emails per day. You want to:

• ✅ Respond quickly (customers expect fast replies)
• ✅ Keep quality high (can't send bad responses)
• ✅ Save time (writing 50 emails is exhausting)

The Solution: Review Workflow

Here's what you build:

Human-in-the-Loop Pattern

Automation stops for human review before continuing

Research

Auto

Automatic

→

Draft Email

Auto

Automatic

→

Human Review

Manual

→

Send Email

Manual

How it works:

1. 1. System automatically generates email draft
2. 2. You review the draft and make any edits
3. 3. You check an "Approved" checkbox when ready
4. 4. Send Email column only runs if approved
5. 5. You manually click "Send" button for final safety

✅ Advantages

• • Safe (human oversight)
• • Can edit before sending
• • Accountability tracked
• • Prevents mistakes

⚠️ Considerations

• • Slower (waits for human)
• • Requires attention
• • Need review process
• • Not fully automated

🛡️ Best for: Sending emails, making purchases, or any action that can't be undone

1. Email arrives → Creates new row automatically
2. AI reads email → Understands the question
3. AI searches knowledge base → Finds the right answer
4. AI drafts response → Creates professional reply
5. STOP! You review → Check if the response is good
6. You approve → Check the "Approved" box
7. Email sends → Only if you approved it

The Results

Before (Manual):

• 50 emails × 5 minutes each = 4+ hours per day
• 😫 Exhausting

After (Review Workflow):

• AI drafts 50 emails in 5 minutes
• You review 50 emails in 30 minutes
• ✨ Total: 35 minutes per day

You saved 3.5 hours every day while keeping full quality control!

💡 And It Gets Better...

After a month of reviewing, you notice patterns:

• 90% of password reset emails: AI is perfect, no edits needed
• 80% of shipping questions: AI is great
• 50% of refund requests: AI needs help

You can now adjust:

• Let password resets send automatically (no review)
• Keep reviewing refund requests (needs judgment)

Your workflow gets smarter as you learn!

---

6.1.7: The Training Period Approach

One of the smartest ways to use review is during a training period:

How It Works

First 100 responses:

• ✓ Review every single one
• ✓ Note what works and what doesn't
• ✓ Improve AI instructions
• ✓ Build confidence

After 100 responses:

• ✓ Remove review for types that work well
• ✓ Keep review for edge cases
• ✓ Automation gets faster and smarter

Why This Works

Think of it like training a new employee:

Week 1-2: Check everything they do Week 3-4: Check important stuff only Month 2+: They work independently

Your AI assistant is the same way!

💡 Pro Tip: Track Your Success Rate

Keep a simple tally:

• How many AI drafts do you approve unchanged?
• How many need small edits?
• How many need major changes?

If 90%+ are approved unchanged, you might not need review anymore!

---

What You've Learned

Congratulations! You now understand:

• ✅ Why human review matters (AI is powerful but not perfect)
• ✅ When to add review (high-stakes decisions, training periods)
• ✅ Types of review (approve/reject, edit, flag, feedback)
• ✅ How review workflows work (AI generates → you approve → action happens)
• ✅ The training period approach (start careful, get faster)

What's Next?

In the next section, you'll get hands-on and actually build your first approval workflow. You'll see exactly how to:

• Add an approval checkbox
• Make actions wait for approval
• Set up notifications
• Test the complete flow

Ready to build something safe and powerful? Let's go!

---

Previous: 5.3 More Sub-Table ExamplesNext: 6.2 Setting Up Approval Workflows