Training and Support

  • don’t just send people to a class. Need an ecosystem that can handle different types of users. For example, a report script the use once a month but I’ll need to tweak 2 to 3 times a year

Are adopting and adapting agile to power users circumstances – easy, what is it mean to be agile when business analysts may be part of a team but usually work on an individual project

Helping folks become comfortable with the discipline that data science requires

  • But you need an approach that doesn’t dump all of the discipline of computer programming and database development on top of all these folks all at once, and you need something that respects the fact that their time is limited
  • Agile Data Governance: if you try to do it all at once it’ll be overwhelming. Not everything, but backlog list focused on the areas where the differences are most critical (not for things like security, where obviously you need to figure out before people get access)
  • standards, processes, templates; training and mentoring to nurture this capacity; figuring out how to overcome resistance

Strong data foundation that isn’t too complicated (baby SQL-friendly temporary data structures/tables)

Evangelism Cross department team to evangelize, support, make some types of decisions Start with the core folks, ideally people were a little more skilled and are easy to work with and have supportive departments Create multiple on ramps


Organizational culture, including how much departments cooperate versus are silos

Executive buy-in: don’t need to start with it, but you do need to get it to succeed Manage expectations – both hype and skepticism Making sure people get that data wrangling is key – e.g., 80% of ML project is consolidating and cleaning up data Make sure ok with iterative approach May want 2 sets of rules. For example, on the trail vs off the trail – folks who have more latitude but don’t get support if they get into trouble Watching out for demands on folks’ time

Build a diverse, inclusive data science

Identify and develop natural leaders where possible

Having a bunch of buddies who can help you out

Smoothing the Learning Curve

Setting Expectations so power users don’t end up promising more than they can deliver are getting pushed into doing something that’s 2 high-stakes that they’re not ready to handle

Encourage play

Advantages of combining across ecosystems

  • sharing trainings, including thinking of trainings in a way that would facilitate this easy, building trainings that are like playlists so it’s easy to mix-and-match parts to fix a particular audience
  • Nudging support groups to be more friendly and inviting
  • Foster informal connections that can lead to jobs, business opportunities
  • Provide a bridge for thinking through the connections between learning, support, and work
  • Smoothing the Learning Curve

For example, using auto ML today is like microwaving a hot pocket: users don’t understand the assumptions and implicit decisions underlying the tools they are using. As auto ML becomes important to organizations’ success, this is not sustainable.