1. Chat-Based Analysis: The Zero-Code Entry Point

The simplest way to start: upload a CSV or paste a data sample into ChatGPT, Claude, or any major AI chat interface and ask questions in plain English.

"Here's a CSV of our Q2 sales data. Which product category grew fastest month-over-month? Show me the numbers."

The model reads the data, writes analysis code on the fly, executes it, and returns a summary. This works well for one-off questions and small datasets (under a few thousand rows). The limitation is scale — most chat interfaces cap file sizes and context windows, and you're trusting a cloud service with your data.

For Singapore businesses subject to PDPA obligations, think twice before uploading customer-identifiable data to a US-based AI service. Consider the local approach instead.

2. Programmatic Analysis with LLM APIs: The Sweet Spot

This is where the real productivity gains live. Instead of chatting interactively, you write a short Python script that sends your data and question to an LLM API, then processes the response. Same idea, but automated, repeatable, and capable of handling larger datasets.

import pandas as pd
from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com/v1",
    api_key="your-api-key"
)

df = pd.read_csv("sales_q2.csv")
summary = df.describe().to_string()

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a data analyst. Respond with clear findings."},
        {"role": "user", "content": f"Analyze this quarterly sales data:\n{summary}\n\nIdentify top 3 growth categories and any anomalies."}
    ]
)

print(response.choices[0].message.content)

This costs fractions of a cent per query. DeepSeek's API, for example, charges roughly US$0.14 per million input tokens — meaning a typical analysis run costs less than one Singapore cent. Run it daily on updated data and you've built a lightweight analytics pipeline for pocket change.

3. Local LLM Analysis: Full Privacy, Zero Per-Query Cost

For businesses handling sensitive data — medical records, financial statements, customer PII — sending data to a cloud API is a non-starter. The solution: run an open-source LLM on your own hardware.

Tools like Ollama make this dead simple. Install it, pull a model, and point your Python script at localhost instead of an API endpoint.

# Same analysis, zero data leaves your machine
from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # placeholder, not used
)

response = client.chat.completions.create(
    model="llama3.1:8b",
    messages=[
        {"role": "user", "content": f"Summarize key trends in this data:\n{df.head(50).to_string()}"}
    ]
)

An 8-billion-parameter model like Llama 3.1 runs comfortably on a consumer GPU with 16 GB of VRAM — something like an NVIDIA RTX 4060 Ti or RTX PRO 2000. For Singapore home-office or SME setups, this is entirely achievable. The hardware pays for itself within months compared to per-token API costs at volume.