Katie Bauer!

I got asked what this "standardization" looks like. I wish I knew, and I'd just say it out right if I did! But I still have some thoughts based on what I've seen work on both eng and DS teamshttps://twitter.com/imightbemary/status/1404471774297812998 …

When I think about standardization, two things come to mind: ensuring folks can contribute without having to hold a bunch of background knowledge in their head, and reducing unproductive variance in operations.

For collaboration without background knowledge, a good analogy is unit tests or CI. Both make it possible for engineers to contribute to the same body of work (a production code base) with guardrails that stop them from making obvious or catastrophic mistakes

In the data world, IMO that mostly means everyone is working with the same data models and using the same data transformations. This ensures folks are working with the abstractions, which means it's easier for their work to connect and build off prior art.

In practice, this often seems to manifest as opinionated data warehousing. It allows you to enforce the relevant assumptions and business context without anyone knowing you've done so.

For reducing variance in operations, I immediately think of experimentation. There are commonly agreed upon ways to run experiments, and the consistency reduces a lot of the mental and organizational overhead that makes cross-team collaboration hard.

The procedure of experimentation is a way of ensuring quality and (at least in theory) will make comparison of results across experiments meaningful.

Another source of unproductive variance in operations is people coding in subtly different environments. In a worst case scenario, different versions of software packages can generate different results.