1. Supervised Learning

What it is:
You train the model using historical data where the correct answer is already known.

Common use cases:

• Predicting prices, demand, or churn

• Fraud detection

• Credit scoring or risk prediction

Examples:
Linear Regression, Logistic Regression, and Decision Trees.

When PMs should use it:
Use supervised learning when you have past data with labels (e.g., customer bought/didn’t buy, policy lapsed/active) and want to predict a clear outcome.

2. Unsupervised Learning

What it is:
The model finds patterns in data without predefined labels.

Common use cases:

• Customer segmentation

• Identifying hidden patterns in usage behavior

• Grouping similar products or users

Examples:
Clustering (K-Means), Principal Component Analysis (PCA)

When PMs should use it:
Perfect when you’re in exploration mode—trying to understand users better or discover segments you didn’t know existed.

3. Semi-Supervised Learning

What it is:
A mix of labeled and unlabeled data.

Common use cases:

• When labeling data is expensive or slow

• Image or document classification

When PMs should use it:
Use this when you have limited labeled data but lots of raw data and still want meaningful predictions.

4. Reinforcement Learning

What it is:
The model learns by trial and error, using rewards and penalties.

Common use cases:

• Recommendation systems

• Dynamic pricing

• Personalization engines

When PMs should use it:
Best for continuous optimization problems where the system learns over time based on user actions.

Final PM Takeaway

Don’t start with algorithms, start with the product question:

• Predict something → Supervised

• Discover patterns → Unsupervised

• Optimize continuously → Reinforcement

Machine learning works best when it solves a real product problem, not when it’s added just because it sounds cool.

Let’s understand this using a simple real-world example.

The Customer Message

“My payment failed twice yesterday. Amount got debited but order didn’t go through. Support hasn’t replied in 24 hours. This is very frustrating.”

The AI Output

{
  "summary": "Customer reports payment failure with amount debited and no response from support.",
  "priority": "High"
}

Now let’s see how this output is evaluated using four common evaluation approaches.

1. Code-Based Evaluations (Heuristic / Deterministic Evals)

This is the fastest and cheapest way to evaluate AI outputs.

Instead of humans checking every response, we write code that verifies whether the output meets predefined rules.

Typical checks include:

• Is the output valid JSON?

• Does it contain the required keys like summary and priority?

• Is priority one of {High, Medium, Low}?

• Is the summary short (e.g., under 30 words)?

• Does it avoid banned or risky phrases?

In simple terms:
Code-based evals answer one question: “Is this output valid as per our rules?”

2. LLM-as-a-Judge Evaluation

Here, another AI model evaluates the output.

It scores the response (usually 0–5) on criteria like:

• Accuracy of the summary.

• Appropriateness of priority.

• Missed critical signals.

This method adds contextual judgment that rules alone can’t capture.

3. Human Evaluation (Expert Judgment)

A human reviewer, like a support lead, looks deeper.

In this case:

• Summary is technically correct ✅

• “Amount debited” signals financial risk ✅

• 24-hour silence violates SLA ❌

• Customer frustration isn’t explicitly flagged ❌

Verdict: Acceptable, but needs improvement
Feedback: Add urgency markers like “customer distressed” and “SLA breach.”

Humans bring accountability and nuance that AI still struggles with.

4. User Evaluation (Real-World Impact)

Finally, the most important test:
Did the customer feel helped?
Was the issue resolved quickly?

If not, even a “perfect” AI output has failed.

Final Takeaway

Each evaluation type answers a different question:

• Code evals: Is it valid?

• LLM-as-judge: Is it reasonable?

• Human evals: Is it truly correct?

• User evals: Did it actually work?

Strong AI systems use all four together.

As a product manager, this should sound familiar. Garbage in → garbage out.
Prompt engineering is simply clear thinking expressed in structured instructions. And today, it’s a core PM skill.

What is Prompt Engineering (in PM terms)?

Prompt engineering is the ability to tell AI exactly what problem you’re solving, for whom, and under what constraints, so the output is actually usable.

Think of AI as a super-smart but extremely literal junior PM. If your brief is unclear, the output will be generic.

That’s where ROCCO comes in.

The ROCCO Mental Model

ROCCO gives structure to your prompts so AI behaves like a domain expert instead of a chatbot.

ROCCO stands for:

1. Role – Who should the AI act as?

2. Objective – What is the goal?

3. Context – Business, user, or market background

4. Constraints – Tone, depth, format, limitations

5. Output – How you want the answer structured

Pro Tip: Add a Q at the end → ROCCO-Q, where Q = Quality Check.

Bad Prompt (What Most Beginners Do)

“Suggest features for my app.”

You’ll get a generic feature list that sounds good—but isn’t actionable.

Good Prompt Using ROCCO (PM Example)

Role: Act as a senior Product Manager at a B2C fintech company.
Objective: Suggest features to improve user retention in a payments app.
Context: Indian users, Tier 1 & 2 cities, high drop-off after first transaction.
Constraints: Simple language, focus on quick wins, avoid deep tech details.
Output:

• Top 5 features

• Why each feature matters

• Expected impact on retention

Final Prompt

Act as a senior Product Manager at a B2C fintech company.
Suggest features to improve user retention in a payments app.
your target audience is indian users from Tier 1 & 2 cities with high drop-off after first transaction.

Use the simple language with focus on quick wins and avoid deep tech details.

Generate Output in the below format:

Top 5 features
Why each feature matters
Expected impact on retention

Pro tip: Add one more Q for Quality Check: Re-check if features are realistic for a 3-month roadmap.

👉 Now the AI responds like a thinking PM, not a feature vending machine.

Prompt engineering isn’t about “talking to AI.”
It’s about thinking clearly.

ROCCO forces clarity. And clarity is a product manager’s real superpower.