The most successful firms treat artificial intelligence as a process rather than as just a tool. Cesar Brea, a partner with Bain's Advanced Analytics practice, explains how the five-step AI process ensures efficiency and produces more powerful predictions with each cycle.
Read the Bain Brief: The Cure for AI Fever
Read the transcript below.
CESAR BREA: There's a lot of noise in the press today about AI, and we find a lot of it confuses the means for the ends. People are talking about AI as a thing. The most successful firms actually think about this as a process. In general, an AI process that's successful has five different components.
The first one is obviously getting the question right, making sure that you actually understand very clearly the business problem you're trying to solve, and also link it to the statistical measures that you're optimizing for, that link to that business problem.
The second thing that's really important to do is to make sure that you pay attention to data. Better data is always going to beat a better algorithm. It's just the nature of the beast. When you begin to focus on developing the model, what you really want to do is make sure that you start simple and approach it in iterative sprints that actually always focus on results and improve your performance against the baseline that you set. Most people don't think in terms of a baseline. They just think in terms of the result that they achieve.
Fourth, the successful firms realize that building AI is not simply about building one algorithm. This is actually an end-to-end process that involves a lot of data engineering. It involves thinking about not just performance of models, but about the transparency and scalability of what you're doing. And then finally, being able to integrate that and put it to work.
And finally, the fifth point is really making sure that you understand that it's not just about a better model or model platform. You actually have to get people or processes to use this. And that means thinking in concert about the engineering of the interfaces that people will use, as well as the management structures, practices, organizations and other factors that actually influence their human ability to interact with the output of the AI. So we'd say in the end, basically: Don't think tool, think process.