Why data quality decides AI quality

The fastest way to a better model is usually better data. This module shows you what 'good data' means and why AI makes it matter more.

1.1 Garbage in, garbage out

A model can only be as good as the data it learns from and the data it's given. Better algorithms rarely rescue bad data, but better data often rescues an ordinary algorithm. This idea — improving the data rather than endlessly tuning the model — is sometimes called data-centric AI.

Demand for these skills is growing fast: Coursera's 2026 job skills research reports data quality and data cleansing among the fastest-growing data skills, each roughly doubling in enrolments.

1.2 Six dimensions of data quality

1.3 Why it matters more for AI

• Models learn patterns — including errors. Mislabelled examples teach the wrong lesson.
• Problems are invisible in outputs. A model trained on skewed data still produces confident predictions.
• Errors scale. One bad rule in a spreadsheet affects one report; a model trained on it affects every decision it makes.
• Unrepresentative data creates unfair outcomes for groups that are under-represented (Module 5).
• Law requires it for high-risk AI. The EU AI Act's Article 10 sets data governance requirements for training, validation and testing data of high-risk systems (Module 6).

1.4 The data preparation workflow

1. Understand the purpose and where the data comes from.
2. Profile it — measure what's there (Module 2).
3. Clean it — fix, remove or flag problems (Module 3).
4. Label it, if the task needs labels (Module 4).
5. Check representativeness and split it for training and testing (Module 5).
6. Document and govern it (Module 6).

1.5 The practice dataset

Throughout this course you'll use a small, deliberately messy customer table. Paste this into a notebook (e.g. Google Colab) and keep it handy. Basic Python and pandas are assumed (see Python for AI).

Python · the practice dataset used in this courseimport pandas as pd
import numpy as np

customers = pd.DataFrame({
    "customer_id": [101, 102, 103, 103, 104, 105, 106, 107, 108, 109],
    "signup_date": ["2025-01-04", "04/02/2025", "2025-02-10", "2025-02-10", "2025-03-15",
                    "2025-13-01", "2025-04-02", None, "2025-05-20", "2025-06-01"],
    "country": ["UK", "United Kingdom", "uk", "uk", "Ireland", "UK", "IE", "UK", "U.K.", "UK"],
    "age": [34, 29, None, None, 41, 230, 38, 27, -1, 52],
    "monthly_spend": [42.5, 18.0, 63.2, 63.2, None, 25.0, 12000.0, 31.4, 22.0, 47.9],
    "churned": ["no", "yes", "No", "No", "no", "yes", "no", "YES", "no", "no"],
})

Key takeaways

• Better data often beats a better algorithm.
• Accuracy, completeness, consistency, validity, uniqueness and timeliness.
• Models learn errors and scale them.
• Understand → profile → clean → label → split → document.