AI Briefing: Why Data, FinOps and the Right Model Make or Break Your AI

Hi folks. Welcome back to another AI briefing. Um, [clears throat] I know it's been a while. Um, so for those of you who do not know me, my name is Tom, and I work for Concept 2 Cloud. Um, and we deal an awful lot with, um, [clears throat] legacy systems engineering, enhancements, improvements to be able to leverage data more effectively inside an AI ecosystem, along with helping enable other companies leverage their systems as effectively as possible.

Since I last did an AI briefing, the world has moved on a reasonable amount still. Um, and so what you've got to think about is how you're going to leverage the data, the systems, and the applications that you currently use in a modern AI-centric environment. And of course, data remains key because without data there isn't really any AI enablement that you can do because there's nothing for it to be able to deal with. You know, even if you just think about a frontier model, it's trained with, you know, a huge corpus of data. And so if you want to be able to leverage that inside of your organization, not the huge corpus of data, but le- leverage AI to better the, uh, discovery of information inside of your organization, then you need to think about things like data integrity, data alignment, data consistency.

Because as ever, even in an AI world, you will get garbage in and garbage out. On top of the garbage out, it will cost you more money for systems that are not designed to, um, enable AI from a token perspective. Now, every time you run a query using AI, ask it a question, whatever, there's a number of tokens that are used, and that's starting to cost money. And so the other thing you need to bear in mind is also like FinOps, trying to understand how much AI is gonna cost you as an organization going forward, so that when you come to budget for it, you can use it in the most cost-effective manner. Which brings me on to point number three, and the final one, is finding the right model for the right job.

Because running Opus Four Point Eight for a small text generation, uh, project is probably not what you're after. Whereas having it understand, um, you know, an awful lot of data engineering, for example, write code, Opus is probably what you're after. And those things cost different amounts of money. So if you're using Haiku versus Opus, the amount that it costs you to write that query and get that response is different. So make sure that whilst you're leveraging AI, you use the right model, you pick the right model.

Do some research.