5 Making Confident Decisions
6 From Data to Decisions
You’ve connected to the SalesVP model. You can see your products, volumes, trends, and market shares. But now comes the critical question every brand manager faces:
6.0.1 🤔 The Brand Manager’s Dilemma
“My Primitivo is down 3% versus last year, while my competitor’s Shiraz is up 5%. Should I be worried? Is this a real trend or just random variation? Should I change my strategy or wait it out?”
This chapter teaches you how to move from observation to confident decision-making using statistical analysis.
6.1 The Two-Server Workflow
Get the data
Is this significant?
Test the hypothesis
Act with confidence
Claude seamlessly switches between Power BI (for data) and Statistics (for analysis) in a single conversation.
7 Real Scenarios for Brand Managers
7.1 Scenario 1: Is My Sales Decline Real?
7.1.1 The Situation
Your flagship wine dropped from 125,000 liters last year to 118,000 liters this year — a 5.6% decline. Your manager wants to know: Should we panic?
“My product sold 125,000 liters last year and 118,000 liters this year. Given typical monthly variation in wine sales, is this decline statistically significant or could it be normal fluctuation?”
Claude will:
- Ask clarifying questions about your data (monthly breakdown, variance)
- Perform a t-test or similar analysis
- Give you a p-value and confidence interval
- Explain in plain language whether you should worry
7.1.2 Understanding the Answer
- p-value < 0.05: The decline is likely real — investigate causes
- p-value > 0.05: Could be random variation — don’t overreact
- Confidence interval: Shows the range of “normal” fluctuation
7.1.3 Follow-Up Questions
“Break down my sales by month and test if any specific months drove this decline”
“Compare my decline to the overall category. Is my product underperforming the market?”
7.2 Scenario 2: Product A vs Product B
7.2.1 The Situation
You manage two Cabernet Sauvignons at different price points. The premium one averages 8,500 liters/month, the budget one averages 9,200 liters/month. Marketing wants to discontinue the “weaker” premium product.
“I have two products: Premium Cab averages 8,500 L/month and Budget Cab averages 9,200 L/month over the past 24 months. Is the Budget Cab actually selling more, or is this difference within normal variation? Include the confidence level of your conclusion.”
7.2.2 What You’ll Learn
- Whether the 700 L/month difference is statistically significant
- The effect size — is it a meaningful business difference?
- Your confidence level in the conclusion (95%? 99%?)
7.2.3 ⚠️ Statistical vs Practical Significance
A difference can be statistically significant but not practically meaningful. If you’re 99% confident the difference is 50 liters/month, that’s real but probably not worth acting on.
Always ask Claude: “Is this difference large enough to matter for my business?”
7.3 Scenario 3: Seasonal Patterns
7.3.1 The Situation
You believe rosé wines have a summer peak and want to plan inventory accordingly. But is the pattern consistent enough to rely on?
“Analyze the monthly sales pattern for rosé wines over the past 3 years. Is there a statistically significant seasonal pattern? If so, which months should I expect peak demand?”
Claude will:
- Pull monthly data from Power BI
- Perform seasonal decomposition
- Test if the seasonal component is significant
- Quantify the expected peak/trough months
7.3.2 Planning Question
“Based on the seasonal pattern, how much extra inventory should I plan for June compared to February? Give me a range with 90% confidence.”
7.4 Scenario 4: Did My Campaign Work?
7.4.1 The Situation
You ran a promotion in September. October sales jumped 15%. Your CEO wants to know if the campaign worked or if it was just random timing.
“My product sales increased 15% in October after a September campaign. Looking at the historical month-to-month variation for this product, is this increase statistically significant? Could October typically be a strong month anyway?”
The analysis will:
- Compare October performance to historical Octobers
- Account for typical month-to-month variation
- Determine if the increase exceeds normal fluctuation
- Consider alternative explanations (seasonality, market trends)
8 The Statistical Toolkit
Here are the key statistical tests Claude can run for brand managers:
| Question | Statistical Test | What It Tells You |
|---|---|---|
| “Is this change real?” | t-test | If difference exceeds random variation |
| “Are these products different?” | Two-sample t-test | If two groups are truly different |
| “Is there a pattern?” | ANOVA / Seasonal decomposition | If systematic variation exists |
| “Are these related?” | Correlation | Strength of relationship between variables |
| “What will happen next?” | Regression | Predicted values with confidence intervals |
| “Is this trend significant?” | Trend analysis | If direction is reliable |
9 Best Practices for Statistical Questions
9.1 1. Provide Context
9.1.1 ✅ Good
“My premium wine (150 NOK) sold 8,500 L/month over 24 months with σ=1,200. Is that significantly different from my budget wine (99 NOK) at 9,200 L/month with σ=1,400?”
9.1.2 ❌ Vague
“Is my product doing better?”
9.2 2. Specify Your Risk Tolerance
Different decisions require different confidence levels:
| Decision Type | Suggested Confidence | Why |
|---|---|---|
| Major product discontinuation | 99% | High cost of being wrong |
| Marketing campaign evaluation | 95% | Standard business decision |
| Inventory planning | 90% | Moderate reversibility |
| Exploratory analysis | 80% | Just looking for signals |
“…I need to be 95% confident before recommending this change to leadership.”
9.3 3. Ask About Assumptions
Statistical tests have assumptions. Ask Claude to verify them:
“Before running this test, check if my data meets the normality assumption. If not, what alternative test should we use?”
9.4 4. Get Business Interpretation
Always end with practical implications:
“Based on this analysis, what’s your recommendation? Should I act on this or wait for more data?”
10 Hands-On Exercise
Pick one of your products and complete this analysis:
Get the data: “Show me monthly sales for [product] over the past 2 years”
Check the trend: “Is there a statistically significant upward or downward trend?”
Compare to category: “How does my product’s performance compare to the overall category? Is my trend better or worse than market?”
Make a recommendation: “Based on this statistical analysis, summarize the key findings and recommend whether we should take action”
11 Summary
- Observation ≠ Significance: A 5% change might be noise
- Use the right test: t-tests for comparisons, correlation for relationships
- Specify confidence levels: Higher stakes = higher confidence needed
- Statistical + Practical: Both must align for action
- Claude bridges the tools: Seamlessly goes from Power BI to Statistics
12 What’s Next
You now have the skills to:
- Query data from Power BI
- Validate findings with statistical tests
- Make confident, data-backed decisions
In the next chapter, we’ll explore advanced integrations: combining multiple MCP servers for complex analysis, visualizing relationships, and creating automated reports.