AI in Data Analytics

Introduction

Data analytics turns raw records into evidence for decisions. Analysts clean data, explore patterns, build models and communicate findings. The bottleneck is often access: many questions wait in a queue for someone who knows the data.

How AI applies

Machine learning finds patterns and makes predictions. Language models translate questions into queries, explain results and draft summaries. The numbers themselves should come from code and databases, not from a model's memory.

Applications and use cases

• Natural-language querying: ask a database questions in plain language.
• Forecasting: sales, demand and resource needs.
• Classification and clustering: customer segments, churn risk.
• Data cleaning: detect duplicates and inconsistencies.
• Insight summaries: describe trends in words for non-specialists.

Technologies involved

SQL databases and warehouses, Python with pandas and scikit-learn, notebook environments, time-series forecasting libraries, and language models for query generation and explanation, connected through read-only credentials.

Implementation guide

1. Document your tables and columns in plain language; models need this context.
2. Create a read-only database user limited to approved tables.
3. Let the model draft SQL, then validate it before running.
4. Show the generated query alongside every answer.
5. Test on questions with known answers before wider release.

Technical resources

A guarded natural-language-to-SQL pattern: the model writes the query, but code decides whether it may run.

import sqlite3, re

def run_safe(sql, db="sales.db"):
    if not re.match(r"^\s*select\b", sql, re.I) or ";" in sql.strip().rstrip(";"):
        raise ValueError("Only a single SELECT statement is allowed")
    con = sqlite3.connect(f"file:{db}?mode=ro", uri=True)   # read-only connection
    try:
        return con.execute(sql).fetchmany(100)              # cap the rows returned
    finally:
        con.close()

In production, also restrict tables, enforce time limits and use a database account with read-only permissions.

Business benefits

Organisations aim for quicker answers, less dependence on a few specialists and earlier detection of trends. Value depends on data quality and shared definitions.

Challenges and risks

• Wrong queries: a plausible-looking query can still answer the wrong question, so show the SQL.
• Data quality: models cannot fix missing or inconsistent data.
• Access control: users should only see data they are permitted to see.
• Overfitting and bias: validate models on unseen data and check fairness.
• Privacy: personal data needs a lawful basis and minimisation.

Examples

Illustrative scenarios: a sales manager asks which regions fell short last quarter and sees the underlying query; a retailer forecasts weekly demand per product; an operations team clusters support tickets to find recurring causes.