Garrett Anderson

In my experience as a hiring manager, I appreciate a thank you note, however it's not going to be a differentiating factor unless there's something memorable you're following up on. If you send a generic "thanks for your time, looking forward to working with you" email, it's less impactful than a note that mentions something we discussed in the interview, reinforces a skill we talked about, or follows up on an open question from the interview. Treat it as a bayesian analysis. You're trying to update my prior assumptions with new data in your favor.

An honest contemplation and thorough meta-analysis of the analysis that failed can be force multiplier if approached correctly. It's important to set ego aside and dig into each step in the process to find weak points of failures. While it can be tempting to start from an assumption that failure was due to externalities or that every step but one was correct, breaking down the work into individual elements can both help identify failure points and also help improve your work flow for the next analysis.

Often overlooked, the expected business result can make or break an analysis. Looking purely at statistical power can tell you, for example, which treatment cab improve your onboarding funnel the most. However, only effect size can tell you whether the ROI for either treatment is worth the cost to implement a process change.

One- size- fits-all tools will rarely give the level of granular detail and custom aggregation that will make the best use of any analysis. It's tough to find a balance between ease of use and customization power, but it's an important area to focus when evaluating. Tools that allow code access in addition to a GUI often bridge the gaps between a casual report generator and an in- depth power user.

Tools such as pyWhy can help speed up the process of splitting statistical methods to the data to arrive at conclusions faster, or to give more focused areas to investigate further

Expert input is crucial during a data cleaning task. Approaching it by rote or with statistical methods without understanding the underlying data leads to wrong methods, wrong techniques, and reduces the quality of your input data rather than improving it