Build a Variance Commentary System with Claude

Before building prompts, clarify what your demand review requires. Good variance commentary should:

• State the overall performance (total actual vs. plan, the %)
• Explain the top 2–3 drivers (what caused the variance — not just that it happened)
• Distinguish between structural shifts (lost customer, new competitor) vs. temporary factors (weather, timing)
• Set up the forward-looking implication (does this variance continue?)
• Be specific enough that someone who wasn't in the analysis can follow the logic

Open claude.ai. Test this foundation prompt with last month's data:

You are helping me write forecast variance commentary for a demand planning review. My audience is supply chain and commercial leadership — they are numerate but want clear business language, not statistical jargon.

Format requirements:
- Start with the headline number (total actual vs. plan, % variance)
- Explain top 3 drivers of the variance in order of impact
- Label each driver as: Structural (ongoing impact), Timing (will reverse), or One-Time
- End with a one-sentence forward look (does this variance continue?)
- Length: 150-200 words

Here is my data:
Category: [name]
Total Actual: [X units/$ ]
Total Plan: [Y units/$]
Variance: [+/-Z%]

Top variance drivers:
1. [SKU/customer/channel name] — [+/-amount] — [reason]
2. [SKU/customer/channel name] — [+/-amount] — [reason]
3. [SKU/customer/channel name] — [+/-amount] — [reason]

What you should see: A clean, professional commentary that captures your data in readable narrative form.

Run this prompt with 2–3 different sets of variance data. Note where it falls short:

• Does it get the tone right for your audience?
• Does the length match what's expected in your reviews?
• Does it handle both positive and negative variances well?

Adjust the prompt with specific refinements:

• "Make it more concise — our leadership prefers bullet points over paragraphs"
• "Add a recommended action for each driver"
• "The forward-look should reference the next 4 weeks, not general outlook"

You'll need slightly different formats for different situations. Build variants for:

Big overperformance commentary:

Adapt the commentary format for a situation where actuals significantly exceeded plan. The tone should be positive but analytical — explain why the upside happened without sounding like you expected it.

Supply-driven variance:

Adapt the commentary for a case where the variance was caused by supply constraints (not demand changes). Clearly distinguish between demand signal and supply execution in the explanation.

Mixed variance (some up, some down):

Adapt the format for a portfolio where some categories exceeded plan and others missed — with a net result close to plan but significant offsets. Show both the positive and negative drivers clearly.

Create a Google Doc titled "Demand Commentary Prompt Library." Paste your working prompts (baseline + variants) with examples. This is your starting point each cycle.