Ask HN: Bottom Up ML Development at FAANG?

I’ve worked as a Data Science/ML Engineer for 12 years. I’ve worked for 2 “boring” Enterprise Software companies most people haven’t heard of, and 2 non-tech commercial hardware product (e.g. HVAC systems) companies. In all my jobs, my work has been defined by the “business.” It’s a typical top-down approach, with the business or stakeholders proposing all the use cases and the business value. Our job is to explore feasibility of these use cases and build something.

I recently interviewed for an SWE-Machine Learning role at a FAANG company. I asked the interviewer how they worked with the business to come up with use cases, and they said they (the SWEs) came up with the use cases, not a business stakeholder. SWEs typically came up with ideas, wrote a proposal (design document??), and things went from there, all to way to finding the people to build it and test it. It was a very bottom-up approach. When I asked them how they worked with the business side of things to determine the value of what they were building, they seemed a bit perplexed at my question. I asked for more details on examples of projects that were done this way, but the answers were very vague. They would not specify which products they worked on either. The interview fell apart at that point.

Could somebody who has worked in this type of way at a large tech company help me understand how this works? What’s an example of a project that was done this way? How do you propose a project and get buy-in? How are people allocated to the project, how to you evaluate the success of it, etc? Is it really a completely bottom-up approach? Are there any blog posts, books, etc that talk about this process in more detail?

I’ve been thinking about the products where this type of approach could work, and I can definitely see how this could work for a social media product. You could build an ML feature and run experiments to validate if it drives time on the site by directly measuring how people are using the site. But surely FAANG has a ton of other use cases (e.g. sales pipeline) where domain experts are driving the use cases?

Any insight people can provide will be helpful. A bottom-up approach is really different from all the jobs I've had my whole life. So I'm very curious to learn more. Thanks!