Data insights, and now AI systems, are only as good as the quality of the data they’re built on. Inaccuracies and biases in your data can lead to costly mistakes, misinformed decisions, and unreliable models. In this talk, I highlight the typical lifecycle of data and the phases where bad data sneaks in. I also cover practical ways to prevent, diagnose, and fix issues before they spread across dashboards or machine learning pipelines.

Businesses rely on good data to make thoughtful decisions. When that underlying data is of poor quality, it can lead to unexpected and expensive outcomes. Preventing bad data starts with understanding how it gets created in the first place. In this talk, I’ll walk through the lifecycle of data, show where bad data gets introduced, and share examples of how to catch it early in the pipeline. You’ll learn how to diagnose common data problems, trace them to their root cause, and take actionable steps to fix them.

As AI becomes more deeply embedded in business processes, these lessons matter more than ever. The talk provides a practical blueprint for anyone who works with data to be proactive and intentional about ensuring data quality.

Key takeaways:
1. What is bad data and why should we care about fixing it - especially in the age of AI?
2. How can we prevent bad data from occurring?
3. How can we diagnose conditions that resemble poor data quality?
4. How can we cure and fix bad data before it impacts decisions or models?

I have seen data at scale and worked on products with 400M+ users. Thus, I’ve seen the impact of what bad data can do. Over the years, I’ve developed a robust framework to narrow down data issues quickly so that they can be fixed. I’ve also developed policies around checking for bias in data, that will help businesses be more thoughtful in their data solutions.

I have given plenty of variations of this talk, and just in the last year 2 of those sessions were billed as keynotes. I am able to adapt this talk to a more technical overview, a more business overview, or a mix of both depending on the audience.