Skip to main content

Getting Better Results

Learn how to consistently get high-quality, useful outputs from Schatzi AI.

When the AI "Doesn't Understand You"​

Problem: Vague or Off-Target Responses​

The AI is powerful but not psychic. Unclear prompts lead to generic or irrelevant responses.

Solution: Add Context and Clarify Your Goal​

Instead of:

Help me with this document.

Try:

Review this 10-page sales proposal for a Swiss manufacturing client.
Focus on:
1. Clarity of value proposition
2. Pricing structure
3. Terms and conditions

Provide specific suggestions for improvement.

Framework to Follow​

Use the 5-part prompt structure:

  1. ROLE: "You are a Swiss business consultant"
  2. TASK: "Review this proposal"
  3. CONTEXT: "For a manufacturing client, conservative approach"
  4. FORMAT: "3 strengths + 3 improvements + priority ranking"
  5. CONSTRAINTS: "Keep suggestions actionable and specific"

When Results Are "Too Generic"​

Problem: Superficial or Boilerplate Responses​

The AI gives you template-like content that doesn't address your specific situation.

Solution: Provide Concrete Examples, Samples, and Constraints​

Technique 1: Show Examples​

Generic prompt:

Write a professional email.

Specific with example:

Write a follow-up email to a client.

Here's an example of our typical tone:
"Thank you for meeting with us yesterday. We appreciated 
discussing your growth plans for 2025. As promised, I'm 
attaching our proposal with the revised timeline."

Use this tone to write a similar follow-up for a different 
client where we discussed digital transformation strategy.

Technique 2: Add Constraints​

Generic:

Summarize this report.

With constraints:

Summarize this quarterly report in exactly 5 bullet points.

Requirements:
- Each bullet must start with a number (revenue/costs/growth %)
- Focus only on financial performance
- Avoid industry context or background
- Flag any concerning trends with [CONCERN] marker
- Maximum 20 words per bullet

Technique 3: Provide Your Own Sample​

If you want content in a specific style, paste an example of what "good" looks like:

I need 3 product descriptions. Here's the style I want:

Example:
"Swiss Precision Watch – Where heritage meets innovation. 
Handcrafted in Zürich since 1952. Each timepiece carries 
a 50-year warranty. [SHOP NOW]"

Now create similar descriptions for:
1. Leather briefcase
2. Fountain pen
3. Desk organizer

Match the tone: premium, heritage-focused, concise.

When Cost Is Too High​

Problem: Using Too Many Tokens​

Your token usage is higher than expected, increasing costs.

Solution: Strategic Model Selection and Efficient Prompting​

Strategy 1: Use Smaller Models for Drafts​

Workflow:

  1. Draft with Mini or Standard model (cheap, fast)
  2. Review the output
  3. Refine specific sections with Thinking model if needed
  4. Finalize manually

Example:

Step 1 (Standard model): "Draft a 500-word blog post about 
Swiss data privacy"

[Review draft]

Step 2 (Thinking model): "Expand the section on FADP 
compliance with specific technical requirements"

Strategy 2: Reserve Large Models for Key Tasks​

Use Standard/Mini for:

  • Email drafting
  • Simple translations
  • Quick summaries
  • Formatting and editing
  • Brainstorming ideas

Use Thinking for:

  • Strategic analysis
  • Complex problem-solving
  • Multi-document comparison
  • Critical decision support
  • Technical deep-dives

Strategy 3: Optimize Prompt Efficiency​

Before:

[Uploads 100-page annual report]

"Please read this entire annual report carefully and 
provide a complete analysis of everything including 
all the financial data, operational highlights, risk 
factors, governance information, and any other relevant 
information you can find. I need to understand every 
aspect of this company's performance."

(Cost: ~50,000 tokens)

After:

[Uploads pages 15-20: Financial Summary section]

"Extract key financial metrics from this section:
1. Revenue growth (YoY %)
2. Operating margin
3. Cash flow
4. Debt ratio

Format as a 4-line table."

(Cost: ~2,000 tokens)

Savings: 96%

Iterative Refinement Approach​

The Power of Conversation​

Don't try to get everything perfect in one shot. Build your result through conversation:

Round 1: Get the Structure​

Create an outline for a quarterly business review presentation.
Include: Executive summary, Financial highlights, Key 
initiatives, Q4 outlook.

Round 2: Develop Each Section​

Now expand the "Key initiatives" section. For each initiative,
include: Status, Owner, Metrics, Risks.

Round 3: Refine Specifics​

The Q4 outlook section needs more specific numbers. Add:
- Revenue target (range)
- Expected headcount changes
- Major milestones

Round 4: Final Polish​

Make the executive summary more concise - maximum 100 words.

Why this works:

  • Easier to guide the AI step-by-step
  • You maintain control at each stage
  • Can switch models between rounds
  • Lower cost than trying to get perfection in one prompt

Quality Control Checklist​

Before accepting any AI output, verify:

Accuracy​

✅ Are facts correct?
✅ Are numbers accurate?
✅ Are dates and names right?
✅ Are industry-specific terms used correctly?

Relevance​

✅ Does it address your actual question?
✅ Is the level of detail appropriate?
✅ Is the tone right for the audience?
✅ Does it match your constraints?

Completeness​

✅ Are all requested elements included?
✅ Is the format what you asked for?
✅ Are examples or data points provided?
✅ Is anything important missing?

Usability​

✅ Can you use this output immediately?
✅ Or does it need significant editing?
✅ Is it structured logically?
✅ Is it the right length?

Advanced Techniques for Better Results​

1. Prime with Examples (Few-Shot Learning)​

Show the AI exactly what you want:

I need customer service responses. Here are 2 examples:

Example 1:
Customer: "Where is my order?"
Response: "Thank you for contacting us. I've checked your 
order #1234. It shipped yesterday via Swiss Post and should 
arrive by Friday. Tracking: [link]. Is there anything else 
I can help with?"

Example 2:
Customer: "The product is defective."
Response: "I'm sorry to hear that. Let's get this resolved 
immediately. Please send a photo to returns@company.ch with 
your order number. We'll send a replacement right away and 
arrange pickup of the defective item."

Now write a response for:
Customer: "Can I change my delivery address?"

2. Use Negative Prompting​

Tell the AI what NOT to do:

Write a professional bio for our website.

Do include:
- 15 years of experience in Swiss banking
- Specialization in wealth management
- French and German fluency

Do NOT include:
- Flowery or exaggerated language
- Claims we can't verify
- Personal information
- Marketing jargon like "passionate" or "innovative"

Tone: Understated, factual, Swiss professional.

3. Request Step-by-Step Reasoning​

For complex analysis:

Analyze whether we should expand to Geneva.

Work through this step-by-step:
1. First, what are the main cost factors?
2. Then, what are the revenue opportunities?
3. Next, what are the key risks?
4. Finally, what's your recommendation with reasoning?

Show your thinking for each step.

4. Specify Output Structure​

Be precise about format:

Compare these 3 suppliers.

Format:
| Supplier | Price | Delivery Time | Quality Rating | Risk Factor |
|----------|-------|---------------|----------------|-------------|
| [Name]   | [CHF] | [Days]        | [1-5]         | [H/M/L]     |

Then write 2-3 sentences explaining which you recommend and why.

Common Improvements by Task Type​

For Analysis Tasks​

  • Specify which aspects to analyze
  • Define success criteria
  • Request specific metrics or KPIs
  • Ask for risk assessment

For Writing Tasks​

  • Provide tone examples
  • Specify audience
  • Define length constraints
  • Request specific structure

For Data Tasks​

  • Clarify output format (table, bullets, prose)
  • Specify which metrics matter
  • Define threshold for "significant"
  • Request visual descriptions if helpful

For Creative Tasks​

  • Provide style examples
  • Define brand guidelines
  • Specify what to avoid
  • Request multiple options

Measuring Success​

Track your prompt effectiveness:

Keep a Prompt Library​

Save prompts that work well:

  • Organize by task type
  • Note which model works best
  • Track approximate token costs
  • Update based on results

Learn from Failures​

When results aren't good:

  • Identify what was unclear
  • Note what context was missing
  • Try alternative phrasing
  • Document what worked better

Iterate and Improve​

  • Start with basic prompts
  • Refine based on results
  • Build complexity gradually
  • Share successful prompts with team

Need More Help?​

Learn More​

Get Support​

  • Contact Support
  • Share example prompts and results
  • Get personalized optimization advice
Master the Fundamentals

The most effective way to get better results is to master clear, specific prompts. Invest time in learning prompt engineering and you'll save time and money on every AI interaction.

Quick Reference: Improving Any Result​

If the result is...Try this...
Too vagueAdd specific constraints and examples
Off-topicClarify the goal and add context
Too longSpecify exact length or structure
Too shortAsk for more detail on specific aspects
Wrong toneProvide tone examples or describe audience
Factually wrongVerify source material, try more specific model
GenericProvide examples of desired style
IncompleteList required elements explicitly
Poorly formattedSpecify exact format (table, bullets, etc.)
Too expensiveUse smaller model, split into steps, be more concise