	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Anthropic Claude API > Claude Summarization for Long Documents

Claude Summarization for Long Documents

Author: Venkata Sudhakar

Summarization reduces large volumes of text into actionable insights, saving hours of reading time. For ShopMax India, weekly sales reports run to 50 pages, supplier agreements to 100 pages, and customer feedback to thousands of review entries. Claude can distill these into executive summaries in seconds. The challenge with long documents is fitting them within Claude's context window - the solution is chunked summarization, where the document is split into sections, each summarized independently, and then the section summaries are combined into a final summary.

Three summarization patterns cover most use cases: direct summarization (document fits in context, single call), chunked summarization (document exceeds context, split and summarize each chunk), and hierarchical summarization (split into sections, summarize sections, then summarize the summaries). For ShopMax India business documents, always specify the audience and purpose in the prompt - a summary for a warehouse manager should emphasize stock levels, while the same report summarized for the CFO should highlight revenue and margins.

The following example shows ShopMax India summarizing a long weekly sales report using the chunked map-reduce pattern, producing both a detailed section summary and a final executive summary:

import anthropic
import os

client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))

SALES_REPORT = """
Weekly Sales Report - ShopMax India - Week 18, 2026

Section 1: TV Category Performance
Samsung QLED 55-inch sold 234 units across all cities. Mumbai led with 89 units.
LG NanoCell 50-inch sold 187 units. Price drop of Rs 2000 drove 40% sales spike.
Sony Bravia 43-inch underperformed at 92 units. Stock at 15-day supply.
Total TV revenue this week: Rs 2.34 crore. Up 18% versus last week.

Section 2: Air Conditioner Sales
Daikin 1.5T Inverter was top seller with 312 units as summer season starts.
Voltas 1T Window sold 198 units, strong in Delhi and Hyderabad markets.
Blue Star 2T Split slow at 67 units due to Rs 44990 price point resistance.
Total AC revenue: Rs 1.87 crore. Up 67% versus last week due to season.

Section 3: Washing Machine Category
IFB 6.5kg Front Load remains category leader at 156 units.
Samsung 7kg Top Load steady at 134 units with strong Bangalore demand.
LG 8kg Front Load premium segment showing 22 units, below target of 35.
Total washing machine revenue: Rs 0.82 crore. Flat versus last week.

Section 4: Delivery Performance
On-time delivery rate: 91.3% across all cities.
Mumbai: 94.1% on time. Bangalore: 92.7%. Delhi: 89.4% (logistics issues).
Average delivery time: 2.4 days. Customer satisfaction score: 4.3 out of 5.
28 escalations this week, 19 resolved within SLA.
"""

def summarize_section(section: str, audience: str) -> str:
    response = client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=150,
        messages=[{"role": "user", "content":
            "Summarize this section for " + audience + ". "
            "2-3 sentences, highlight key numbers and trends.\n\n" + section
        }]
    )
    return response.content[0].text

def executive_summary(section_summaries: list, audience: str) -> str:
    combined = "\n\n".join(section_summaries)
    response = client.messages.create(
        model="claude-sonnet-4-5",
        max_tokens=200,
        messages=[{"role": "user", "content":
            "Create a 3-sentence executive summary for " + audience + " "
            "from these section summaries:\n\n" + combined
        }]
    )
    return response.content[0].text

sections = [s.strip() for s in SALES_REPORT.split("Section") if s.strip() and s[0].isdigit()]
sections = ["Section " + s for s in sections]
audience = "ShopMax India CEO"
section_summaries = [summarize_section(s, audience) for s in sections]
print("Section Summaries:")
for i, s in enumerate(section_summaries, 1):
    print(str(i) + ".", s)
    print()
print("Executive Summary:")
print(executive_summary(section_summaries, audience))

It gives the following output,

Section Summaries:
1. TV category had a strong week with Rs 2.34 crore revenue, up 18%. Samsung QLED
led with 234 units in Mumbai. Sony Bravia needs attention with stock at 15-day supply.

2. AC category surged 67% to Rs 1.87 crore as summer season kicks in. Daikin led
with 312 units. Blue Star premium segment faces price resistance at Rs 44,990.

3. Washing machines flat at Rs 0.82 crore. IFB front load maintains category lead
but LG 8kg premium underperformed at 22 units vs 35 target.

4. Delivery performance solid at 91.3% on-time with 4.3/5 satisfaction. Delhi at
89.4% needs logistics attention. 28 escalations with 19 resolved within SLA.

Executive Summary:
ShopMax India delivered strong Week 18 performance with total revenue of Rs 5.03 crore,
driven by a 67% AC surge as summer starts and steady TV growth of 18%. Delivery
performance remains healthy at 91.3% on-time nationwide, though Delhi logistics needs
immediate attention. Priority actions: address Delhi delivery issues, monitor Blue Star
AC pricing, and investigate LG 8kg washing machine underperformance.

For ShopMax India, automate weekly report summarization by scheduling the summarization pipeline to run every Monday morning and email summaries to relevant stakeholders before their first meeting. Use different prompts for different audiences - warehouse managers need stock alerts, city heads need city-specific performance, and the CEO needs overall trends and exceptions. For very large documents (quarterly reports, audit files), the hierarchical approach scales well: summarize each page, then each chapter, then produce the final executive summary in three passes.

Send your comments, suggestions or queries regarding this site to [email protected].