I'm so excited to have had a small hand in this evolution of Google Search. It was some of the most fun & challenging research I've worked on, and it was an honor to team up with the best people out there. So much more to come!

A designer is a good prior over a large product design space. Sure you can use experiments as a substitute for relying on priors, but why would you subject your users to that?

The "Pizza by Alfredo" paradox strikes again

It's high planning season, which means it's again time to decide how the hell we measure data teams' impact. While this has always been a source of handwringing for data leaders, this year I've got two categories of metrics that I not only like--I'm actually planning to use them

Thankfully they clarified that they simply like the scene & don't think I'm a trash fire manager a la Michael Scott And they kindly recorded the call so that my beautiful "holding back tears" face would be documented for all of eternity

Today was my 2nd-to-last day @slice. Was halfway into a 1:1 when my entire team jumped on the call as a goodbye surprise. They said they were going to "leave it to the professionals to give a proper goodbye," then played this https://t.co/yP30epKg3u Will miss this group a ton 😭

This is a really excellent talk. @harterrt puts a lot of structure around how to frame & communicate your invisible, technical work to stakeholders (and to your manager!)

This is a really great video from the @WSJ on rapid Covid tests, how to think about them and how to use them specifically for travel. Not all tests are the same. Worth watching all the way through! https://t.co/T2UWm86Foc

If done well, these lead to analysts taking only the "time that’s necessary to give someone the confidence to make a decision, and no more."

2. We have to understand the problem *from the perspective of the stakeholder* Even when we fully understand the research q, we can't make an impact if we can't convince functional leads to act. We have to let stakeholders' perspectives inform the approach & how we frame recos

1. We *must* have a good grasp of the problem we've been asked to solve We can't make an impact if we don't address the right question. And understanding the problem helps identify which threads are worth following.

When analysts "judge their work by how quickly people make decisions with it," it helps incentivize us to focus only on the rabbit holes that will inform the decision in question. I think @bennstancil calls out two important prerequisites for this too:

Even the simplest analyses can branch into a million different directions. There's always a pull to investigate each branch, but having the right mindset from the start can help resist that pull.

It's especially hard for less experienced analysts to know when their work is "done," and it's common for them to go down far too many rabbit holes during their analysis

"Prolific, a tool for scientists conducting behavioral research, had no screening tools in place to make sure that it delivered representative population samples to each study." 🙃🙃🙃 "The company has now re-organized, putting a team in charge of demographic balancing"

To boot, this "teenager on TikTok" literally just improved the quality of all future research performed through the tool. They should be thanking them

Everyday I love @kelsa_ a little bit more brb heading to the container store to blow $100 on things I didn't know existed but that we absolutely need

This stands far above all other automated analysis tools. The output looks useful & is responsibly framed. Refreshing to see this. "[It] follows the same approach as a human analyst. It should have a specific recommendation, be easy to understand, and start from a real question"

vendors are **calling my cell phone** now 😐 why