Introduction: Why Supply Chains Need Stories
How LLMs Transform Data into Stories
LLMs act astranslators
From Raw Data to Prompt
A successful data storytelling prompt follows a straightforward structure:
- Goal:What is the purpose of the story?
- Audience:Who will read it?
- Format:Summary, presentation, or email?
- Data:Include a JSON or CSV excerpt.
- Tone:Formal, persuasive, or operational?
Here’s an example prompt for a generic LLM:
Generate a natural language executive summary for the logistics operations team. Use the performance data from the last 30 days (average transit time: 4.2 days, error rates: 1.8%, delays: 12%). Adopt an analytical tone, highlight emerging trends, and suggest a priority action.Generating Stories with LLMs
LLMs synthesize facts into coherent sentences, add context (such as seasonality or capacity constraints), and maintain a logical flow. When integrated with platforms likeClaude Code, the entire workflow can be automated: data acquisition, prompt generation, response quality evaluation, and final story publication.
Practical Example: Turning Supply Chain Data into a Story
CloudNC integrated a custom LLM into its supply chain optimization engine. Sensors send a JSON file every hour. Below is a Python snippet that demonstrates how to trigger story generation:
import json, requests
from datetime import datetime, timedelta
# Retrieve data from the last 24 hours
payload = {"query": {"range": {"timestamp": {"gte": (datetime.now() - timedelta(hours=24)).isoformat()}}}}
response = requests.post('https://api.cloudnc.ai/supply-chain-data', json=payload)
data = response.json()
# Prompt for the LLM
prompt = f"""
Analyze logistics data for the period {datetime.now().strftime('%Y-%m-%d')}.
Provide:
1. Performance overview (average transit times, error rates).
2. Key trends (increase/decrease in speed).
3. A recommended action for the operations team.
Data: {json.dumps(data)}
"""
# Call the LLM model
llm_resp = requests.post('https://api.anthropic.com/v1/complete', headers={"Authorization": "Bearer YOUR_KEY"}, json={"prompt": prompt, "model": "claude-3-opus"})
print(llm_resp.json()['text'])Using Claude Code Plugin Evaluations for Automated Assessment Reports
In September 2026, Anthropic introduced plugin evaluations for Claude Code, which include six types of evaluators and a CI gate. Companies can now route raw data to an evaluation plugin that:
- Verifies factual correctness.
- Checks relevance for the intended audience.
- Applies style controls.
The automated workflow looks like this:
This minimizes human error and ensures every story passes an LLM-based quality check.
The Fly Language Model: A Metaphor for Complex Stories
The 2026 Fly Language Model (FLM), which simulates the brain of a housefly, demonstrates that even high-resolution neural networks can generate coherent narratives. While not intended for the general public, the FLM teaches a key lesson: the depth of contexts can be managed when the model has a clear prompt directive and memory constraints.
When applied to supply chain scenarios, LLMs can handle multiple dimensions of data (shipping delays, customs constraints, warehouse availability) without losing coherence.
Best Practices for Effective Prompting in Data Storytelling
Prompt Structure: The Formula
- Define the objective
- Profile the audience
- Select the format
- Insert data
- Tone/style constraints
Customizing Tone and Audience
Tailor the language:"The logistics team should..." for internal audiences;"External stakeholders should know..." for executive reports.
Validating and Iterating Results
Always use a second LLM (or an evaluation plugin) to check logic and verify factual claims. Record successful prompts for future reuse.
Key Takeaways: Transform Data into Impact
- Faster, evidence-based analysis.
- Improved communication between teams.
- Consistent quality reports through automated control.
- A competitive advantage in the era of digital logistics.
Start with a single use case, apply prompt engineering principles, and watch your data come to life.
Immediate Actions
- Define a concrete supply chain use case (e.g., transit time monitoring).
- Create a basic prompt using the template above.
- Integrate a Claude Code evaluation plugin for automated quality control.
- Iterate based on feedback and measure stakeholder adoption speed.