Scott Ambler

One of the fundamental challenges for machine learning (ML) teams is data quality, or more accurately the lack of data quality. Your ML solution is only as good as the data that you train it on, and therein lies the rub: Is your data of sufficient quality to train a trustworthy system? If not, can you improve your data so that it is? You need a collection of data quality “best practices”, but what is “best” depends on the context of the problem that you face. Which of the myriad of strategies are the best ones for you?

This presentation compares over a dozen traditional and agile data quality techniques on five factors: timeliness of action, level of automation, directness, timeliness of benefit, and difficulty to implement. When you understand what data quality techniques are available to you, and understand the context in which they’re applicable, you will be able to identify the collection of data quality techniques that are best for you.

Data technical debt refers to data quality challenges associated with legacy data sources, including both mission-critical sources of record as well as “big data” sources of insight. Data technical debt, sometimes called data debt, impedes the ability of your organization to leverage information effectively for better decision making, increases operational costs, and impedes your ability to react to changes in your environment. Bad data is estimated to cost the United States $3 trillion annually alone, yet few organizations have a realistic strategy in place to address data technical debt.

This presentation describes the types of data technical debt, why each is important, and how to measure them. Most importantly, this presentation works through disciplined strategies for avoiding, removing, and accepting data technical debt. Data is the lifeblood of our organizations, we need to ensure that it is clean if we’re to remain healthy.