The Three C’s of Data Readiness for AI

Do We Even Need to Prepare Data for AI?

I've heard different versions of this question across the world. Some wonder if we're overthinking it. Others suggest we should just focus on understanding AI and the data will somehow get ready on its own. Do we even need data to get ready for AI?

The short answer is yes. But it's more nuanced than that.

AI has broken a lot of barriers. Unlike earlier probabilistic models, we don't always need data to look a certain way to work with it. The front end, the back end, and the data have all come together in new ways. But that doesn't mean preparation is obsolete. Data absolutely needs to be ready. The key is understanding that data readiness is always fit for purpose. We need to start with the use case.

My Framework: The Three C’s

At a very high level, the way I think about data readiness comes down to three C’s. It’s not pulled from anywhere, it’s my own framework. The three C’s are:

Contextuality, Confidence, and Control.

Contextuality

Data readiness begins with contextuality, which is metadata management. It’s about what is the data, who is the owner, where is it coming from, where is it going, and how does it look in a schema.

Is it set up? The labels, the names, the file types. It’s basically the information about the data itself. It’s almost like the address, the phone number, the name of the data.

Confidence

The second piece is confidence in the data. What does that data actually bring? What is the value of that data? What are the volumes of that data? Are those values relevant or not relevant?

I come from a statistics background, so every time I see data, I just do a scatter plot. Every time I look at data, I want to see what’s the min, max, what’s the average, what’s the median. Like a very traditional person, I still look at the mean, mode, and median separately. That’s the way I understand and assimilate data. I look at the distributions.

That’s really the confidence in the data. It’s about data observability. If you find an anomly or outlier in the data, we should go back to the business and validate the reason behind the data and collaboratively make a decision to include or exclude that data.

Control

The third C is control, which sounds kind of simple, but it’s really the governance framework. And this is the trickiest part. It’s not a black or white zone. This is about people, processes, policies, and technology.

Every organization, even every department within an organization, has a very different governance framework. The larger organizations tend to be, the more fragmented these frameworks tend to be. When data moves from business to business, team to team, the governance framework traverses through multiple different types.

The Trust Equation Applied to Data

If you think about trust and the trust equation, it’s about credibility, reliability, self-orientation, and intimacy.

The credibility is the value of what the data is bringing. The reliability is that every single time it does that, it does that consistently, with no variation or no unacceptable variation. The intimacy is really about how well you know the data.

Data confidence can be humanized, but probably not in the way people typically talk about it. Instead of asking “Can we trust this data?” we might ask three better questions:

1. Do you have the context?

2. Do you have the confidence?

3. Do you have the control?

The views expressed by Subhadra Dutta are her own and do not necessarily reflect the views of her employer.

This blog post is based on an original AIIM OnAir podcast. When recording podcasts, AIIM uses AI-enabled transcription in Zoom. We then use that transcription as part of a prompt with Claude Pro, Anthropic’s AI assistant. AIIM staff (aka humans) then edit the output from Claude for accuracy, completeness, and tone. In this way, we use AI to increase the accessibility of our podcast and extend the value of great content.