AI Is Only as Good as Its Data

Last week I attended the TomTom Festival in Charlottesville. It was my first time adventuring at this festival and I really loved how they weave the Innovation Summit through the festival — there are events for the entire city, so not just a conference, a community event. I think that is cool, well done Charlottesville.

I had more than one thought from the event and have been digesting my learnings. The one I wanted to share today is related to AI. As I am sure you can imagine, AI got a lot of screen time at the conference.

So here is my thought, or maybe question: who will provide better results — the "great" and powerful model with "good" data, or the "good" model with data that is well maintained? The way I understand it, the underlying data is the key and wins. I would love to hear from the experts in the room if I am on the right track or if I am talking rubbish.

So what does this mean for me? I am still learning a ton about using these AI tools, but I have been much more intentional about how I am structuring the data I provide and the amount I am uploading. It feels like I am getting better results, but who knows.

The trailing thought here is around market data for wages. Historically very messy with no lack of sources, the rule has been to pull from multiple sources and weigh them accordingly. I am intrigued about how, as a small consultant, I can better utilize the available data sources to get the best market indicators possible. I think scraping job postings for published salary ranges is one, the best-in-class salary survey vendors are another (the best), and the newer entrants that pull data directly from the HRIS of their subscribers are yet another.

As with most business decisions, it comes down to prioritizing resources — when is "good enough" good enough, especially as these data sources have very different costs.