You:
Should ThreadCo switch from Royal Mail to DPD for UK deliveries?

Claude (without CoT):
Yes, DPD is generally faster and offers better tracking.
— Shallow. Missing cost analysis, volume considerations, customer preferences.

You:
Should ThreadCo switch from Royal Mail to DPD for UK deliveries?
Think through this step by step. Consider: cost per parcel at our volume
(500/week), delivery speed, tracking quality, customer complaints about
each carrier, weekend delivery options, and environmental impact.

Claude (with CoT):
Let me work through each factor:

1. COST: At 500 parcels/week, Royal Mail 2nd class is ~£3.20/parcel.
   DPD standard is ~£4.50/parcel. Annual difference: £33,800.

2. SPEED: Royal Mail 2nd class: 2-3 days. DPD: next-day.
   But only 12% of ThreadCo's customer complaints mention speed.

3. TRACKING: This is the key issue. 34% of complaints are "where is
   my order?" Royal Mail tracking is basic. DPD provides real-time
   GPS tracking with 1-hour delivery windows...

[continues through all factors, then gives a reasoned recommendation]

import anthropic
client = anthropic.Anthropic()

# Extended thinking is enabled by setting a thinking budget
resp = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=16000,
    thinking={
        "type": "enabled",
        "budget_tokens": 10000  # Claude can use up to 10K tokens for thinking
    },
    messages=[{
        "role": "user",
        "content": """ThreadCo is considering three pricing strategies for their
new premium line. Analyse each strategy considering:
- Impact on existing customer base (2,000 active buyers)
- Margin implications (current margin: 42%)
- Brand perception in the sustainable fashion market
- Competitor pricing (£25-£45 range)

Strategy A: Premium pricing at £55 with free shipping
Strategy B: Value pricing at £35 with £3.50 shipping
Strategy C: Tiered pricing — £40 standard, £55 for organic-certified"""
    }]
)

# Access the thinking and response separately
for block in resp.content:
    if block.type == "thinking":
        print("THINKING:", block.thinking[:200], "...")
    elif block.type == "text":
        print("RESPONSE:", block.text)

You:
ThreadCo sells 500 tees/week at £28 each. We're considering a 10% price
increase. Historical data suggests a 10% price increase causes a 6% volume
drop for our market segment.

Write a Python script that calculates:
1. Current weekly revenue
2. Projected weekly revenue after the increase
3. Break-even volume drop (at what % volume drop do we lose money?)
4. Annual impact assuming the volume drop stabilises after 4 weeks

Claude writes and runs the code, giving exact numbers rather than
approximate mental arithmetic.

Turn 1 — Decompose:
ThreadCo wants to expand from UK-only to EU shipping. Break this
problem down into all the sub-problems we need to solve. Don't
solve anything yet — just list the sub-problems and their dependencies.

Claude produces a dependency graph of 8-12 sub-problems.

Turn 2 — Prioritise:
Which of these sub-problems are blockers (must be solved first)?
Which are independent and can be done in parallel?
Create a phased implementation plan.

Turn 3 — Solve each sub-problem:
Let's tackle "VAT compliance for EU cross-border sales" first.
@returns-policy.md @pricing-model.ts
What are the specific requirements and how should we implement them?

Turn 4 — Integrate:
Now that we've solved the first three sub-problems, are there any
interactions or conflicts between the solutions? What changes when
we put them together?

# shopmate/reviews/monthly_report.py -- 200 reviews in, 3-minute report out
import anthropic, json
client = anthropic.Anthropic()

def generate_review_report(reviews: list[dict]) -> dict:
    """Analyse all ThreadCo reviews for the month and extract actionable insights."""
    reviews_text = "
".join(
        f"[{r['rating']}/5] {r['product']}: {r['text']}"
        for r in reviews
    )
    resp = client.messages.create(
        model="claude-sonnet-4-6", max_tokens=1500,
        system="""You are analysing customer reviews for ThreadCo, a T-shirt brand.
Produce an actionable monthly report for the founder.
Focus on patterns, not individual reviews.
Return ONLY valid JSON.""",
        messages=[{"role":"user","content":
            f"Analyse these {len(reviews)} reviews:

{reviews_text}

"
            'Return JSON: {"top_complaints":["..."],"top_praises":["..."],'
            '"sizing_issues":["..."],"product_suggestions":["..."],'
            '"average_rating":0.0,"summary":"2-3 sentences for the founder"}'
        }]
    )
    return json.loads(resp.content[0].text)

# Sample reviews
sample_reviews = [
    {"rating":5,"product":"Sunset Gradient Tee","text":"Softest tee I own, the colour is exactly as shown"},
    {"rating":2,"product":"Midnight Pocket Tee","text":"Runs small, ordered my usual size and it was tight"},
    {"rating":4,"product":"Wave Print Crop Tee","text":"Love the print but shipping took 2 weeks"},
    {"rating":3,"product":"Midnight Pocket Tee","text":"Nice quality but definitely size up"},
]
report = generate_review_report(sample_reviews)
print(json.dumps(report, indent=2))