The world’s most iconic brands don’t scale by accident.

In this Built to Scale session, where HR leaders share the behind the scenes that fuels their enterprises, Andrea Rickey reveals how she’s modernizing talent strategy across a 19,000-employee workforce and turning HR into a true business driver.

We’ll go deep into what high-leverage HR looks like and what people teams of all sizes can take from it.

You’ll learn how her team is reshaping high-volume hiring through AI and skills-based recruiting, building leadership pipelines, and aligning people decisions directly to business impact — all inside one of the world’s most recognizable companies.

Join us for a candid conversation on the future of talent, organizational design, and what high-leverage HR looks like at scale.

Save your spot

Look, most prompting advice is garbage. "Be nice to the AI!" "Just ask clearly!" Thanks for nothing.

But here's the thing: researchers at Mohamed bin Zayed University tested 26 specific techniques across models from LLaMA-7B to GPT-4. The results? Up to 67% better outputs when you know what you're doing.

The best part? You don't need all 26. Pick 3-5 that fit your task, and you'll see 30-50% improvement. That's the difference between "meh" and "holy shit, this actually works."

These principles break into five groups. Think of them as your toolkit—grab what you need for the job at hand.

Principle 1: Drop the Politeness

Stop saying "please" and "thank you." The model doesn't care, and you're wasting tokens.

❌ "Could you please explain quantum computing if you don't mind?"
✅ "Explain quantum computing in simple terms."

Principle 2: Name Your Audience

Who's reading this? A 5th grader? A PhD? Be explicit.

"Explain machine learning to someone with zero technical background" hits different than "Explain to a data scientist."

Principle 3: Say What TO Do, Not What NOT to Do

Your brain works better with positive instructions. So does the model.

❌ "Don't include irrelevant information."
✅ "Include only relevant information."

Research shows negative prompts underperform by 20-30%. Wild, right?

Principle 4: Prime with Leading Words

Start with phrases like "Let's think step by step" or "Work through this carefully." It triggers deeper reasoning.

Try: "Let's work through this step by step. What's 345 × 67 ÷ 23?"

Watch the quality jump.

Principle 5: Start the Response Yourself

End your prompt with the beginning of what you want back.

"Generate JSON:
{
"name": "

The model picks up where you left off, matching your format perfectly.

Principle 6: Use Delimiters

Separate your prompt sections with ###, ---, or XML tags. Makes everything clearer.

Principle 7: Use "Your Task Is" and "You MUST"

Emphatic language works. It signals importance and boosts compliance by 15-20%.

"Your task is to create a marketing email. You MUST include a clear call-to-action and avoid spam language."

Principle 8: Add Consequences

"You will be penalized if your response contains factual errors."

Sounds harsh? It reduces hallucinations and increases accuracy. Use it for critical stuff.

Principle 9: Offer a Tip

Yeah, seriously. Studies tested amounts from $0.10 to $1,000,000. The sweet spot? $10-$100 for 10-57% better quality.

"I'm going to tip $100 for a solution that addresses all three requirements."

Tiny amounts ($0.10) can backfire. Moderate amounts crush it.

Principle 10: Request Natural Language

"Answer in a natural, human-like manner" kills the robotic corporate-speak.

Compare:
❌ "We are pleased to inform you of our new product offering."
✅ "We just launched something you're going to love."

Principle 11: Show Examples (Few-Shot)

Give 2-5 examples of what you want. This beats long explanations every time.

Principle 12: Simplify the Language

"Explain like I'm 11" forces relatable language and concrete examples.

"Explain cryptocurrency as if I'm 11 years old, using analogies I'd understand."

Principle 13: Kill the Bias

Models reproduce stereotypes from training data unless you stop them.

"Suggest careers for physics graduates. Ensure suggestions avoid stereotypes about who succeeds in these fields."

Principle 14: Let It Ask Questions

"From now on, ask me questions until you have enough information to recommend the best website architecture for my needs."

This eliminates guesswork and reduces back-and-forth.

Principle 15: Test Your Understanding

"Teach me the Pythagorean theorem with 3 practice problems (no answers). Grade my solutions when I provide them."

Great for learning or validating you actually get it.

Principle 16: Assign a Role

"Act as a senior data scientist with 15 years of experience."

This taps into domain-specific knowledge and terminology.

Principle 17: Repeat Key Words

"Write a concise, concise summary. Keep it concise—maximum 100 words. This concise format is critical."

Repetition increases adherence by 10-15%.

Principle 18: Use Consistent Structure

Create templates with ###Instruction###, ###Example###, ###Question###.

Makes your prompts reusable and clearer for complex tasks.

Principle 19: Combine Chain-of-Thought with Examples

Show the reasoning AND the answer.

This combo improves accuracy by 20-40% for reasoning tasks.

Principle 20: Break It Down

One massive prompt? Bad idea. Sequential smaller prompts? Much better.

Instead of cramming everything into one shot:

Principle 21: Specify the Format

"Generate a JSON array of 5 products with name, price, rating (1-5), and description (max 50 words), sorted by rating."

No ambiguity. No reformatting later.

Principle 22: Set Hard Constraints

Clear boundaries = predictable output.

Principle 23: Show Good AND Bad Examples

Principle 24: Define the Tone

"Write a friendly, enthusiastic product announcement for social media, not corporate jargon."

Tone matters more than you think.

Principle 25: Give Context

Background information dramatically improves relevance.

Principle 26: Iterate and Refine

Test → analyze → adjust one thing → retest → document what works.

Prompting is a skill. You get better with practice.

Content Writing?
Principles 2, 3, 10, 16, 20, 24

Data Extraction?
Principles 1, 11, 18, 21, 22

Problem-Solving?
Principles 4, 7, 12, 19, 20

Creative Work?
Principles 6, 10, 16, 24, 25

Learning?
Principles 2, 12, 14, 15, 19

Critical Tasks?
Principles 7, 8, 13, 21, 22

Single principle: 5-15% improvement
3-5 principles: 30-50% improvement
All 26 combined: 50-67% improvement

GPT-4 benefits most consistently. Smaller models show more variation but still significant gains.

Here's the kicker: combining principles creates synergy. The whole is greater than the sum of its parts.

You don't need to memorize 26 principles. Pick the ones that fit your task. Combine 3-5 strategically. Test what works.

The research is solid (VILA Lab, Mohamed bin Zayed University, tested across LLaMA and GPT models). The techniques are proven.

Stop guessing. Start prompting like you know what you're doing.

Because now you do.