Mikhail Zhilkin works as senior data scientist for Stockholm studio of King, taking care of Candy Crush franchise, and has overseen the launch of Candy Crush Soda and Candy Crush Jelly.
Lightning talk; Data-driven vs Reality-driven
“Data-driven innovation” sounds good, but for data to be useful two things must happen:
- data adequately represent reality;
- data are used correctly.
I will focus on #2:
- Data analysis is a complex process, with numerous opportunities to make a mistake. One mistake may be enough to render the analysis useless if not harmful.
- It’s hard to learn from mistakes, as there isn’t the same feedback loop as, for example, there’s in software development. Bugs go unnoticed.
- Peer review doesn’t solve the problem completely, and comes with its own problems.
- What else to try: automate and standardise as much as you can; create a culture of self-doubt.