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

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Anthropic Claude API > Claude Data Extraction from Unstructured Text

Claude Data Extraction from Unstructured Text

Author: Venkata Sudhakar

Data extraction from unstructured text transforms free-form customer messages, emails, and support tickets into structured records that can be stored in databases and processed programmatically. For ShopMax India, customers often send support emails like "I ordered a Samsung TV last Tuesday from your Delhi store, order number is ORD-DEL-5521, but it still shows processing" - extracting order ID, product, city, and issue type from this text automates ticket routing without requiring customers to fill structured forms.

Claude excels at extraction because it understands context and handles variations in how people express the same information. Unlike regex patterns that break when formats vary, Claude extracts "order ORD-DEL-5521", "order number DEL-5521", and "my order #5521" as the same entity. The key is instructing Claude to return null for fields it cannot find rather than guessing, and to extract only what is explicitly stated rather than inferring missing details. Pairing extraction with Pydantic validation ensures the output is reliable before it enters your systems.

The following example shows ShopMax India extracting structured ticket data from customer support emails, enabling automatic routing to the right support queue:

import anthropic
import json
import os

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

SUPPORT_EMAILS = [
    "Hi, I ordered Samsung 4K TV last week, order ORD-MUM-7743. "
    "Delivery was promised by May 5 but nothing arrived yet. I am in Mumbai.",

"My LG washing machine stopped spinning after 3 uses. Order number DEL-4421. "
    "I need someone to come and check it. Located in South Delhi, Lajpat Nagar.",

"Please cancel my order ORD-BLR-9981 for the Daikin AC. "
    "I found the same product cheaper at another store. Refund to my HDFC card please.",
]

def extract_ticket_data(email_text: str) -> dict:
    prompt = (
        "Extract structured data from this ShopMax India support email. "
        "Return JSON with these fields (use null if not found): "
        "order_id, product_name, issue_type (DELIVERY_DELAY/DEFECTIVE_PRODUCT/"
        "CANCELLATION/OTHER), city, customer_request, urgency (HIGH/MEDIUM/LOW). "
        "Extract only what is explicitly stated. Return only JSON."
        "\n\nEmail: " + email_text
    )
    response = client.messages.create(
        model="claude-haiku-4-5",
        max_tokens=200,
        messages=[{"role": "user", "content": prompt}]
    )
    raw = response.content[0].text
    start = raw.find("{")
    end = raw.rfind("}") + 1
    return json.loads(raw[start:end])

ROUTING = {
    "DELIVERY_DELAY": "logistics-team@shopmax.in",
    "DEFECTIVE_PRODUCT": "service-team@shopmax.in",
    "CANCELLATION": "orders-team@shopmax.in",
    "OTHER": "general-support@shopmax.in"
}

for i, email in enumerate(SUPPORT_EMAILS, 1):
    data = extract_ticket_data(email)
    route = ROUTING.get(data.get("issue_type", "OTHER"), ROUTING["OTHER"])
    print("Email", i, "- Ticket Extraction:")
    print("Order ID:", data.get("order_id", "Not found"))
    print("Product:", data.get("product_name", "Not found"))
    print("Issue:", data.get("issue_type"), "| Urgency:", data.get("urgency"))
    print("City:", data.get("city", "Not found"))
    print("Route to:", route)
    print()

It gives the following output,

Email 1 - Ticket Extraction:
Order ID: ORD-MUM-7743
Product: Samsung 4K TV
Issue: DELIVERY_DELAY | Urgency: HIGH
City: Mumbai
Route to: [email protected]

Email 2 - Ticket Extraction:
Order ID: DEL-4421
Product: LG washing machine
Issue: DEFECTIVE_PRODUCT | Urgency: HIGH
City: South Delhi, Lajpat Nagar
Route to: [email protected]

Email 3 - Ticket Extraction:
Order ID: ORD-BLR-9981
Product: Daikin AC
Issue: CANCELLATION | Urgency: MEDIUM
City: null
Route to: [email protected]

For ShopMax India production ticket systems, run extraction on every incoming support email and chat message before human agents see them. This pre-populates ticket fields automatically, reducing agent data entry time by 60%. Add a confidence check: if order_id is extracted but does not match your order database pattern, flag it for agent verification rather than auto-routing. Store extraction results alongside the original message so agents can correct errors and feed them back to improve your extraction prompts over time.

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