What: Antonio Fitas from Kolibri Games walks us in this talk through 5 years of data at the company he works at. Most of their data come from their most famous game Idle Miners Tycoon with something like 100 million events each day and 50 TBs of data inside their data pots.
My perspective: I think Kolibri Games’ journey can be summarized in one sentence they used at the end.
“Choose the right [technologies -> stack & processes & team structures] at the right time for the right amount of data”
Ok I did some replacing there, but they had the word “technologies” in, I simply added the processes and team structures which they displayed at length in this talk.
I won’t comment too much on the talk, instead, I’ll highlight things I found particularly great. The overall theme of Kolibri Games’ data journey in my eyes is summarized in the top quote and really means they got one thing right:
They aligned their data journey directly with their company strategy! But that means it changes all the time! One day they need it fast & lean, on another, they need a much heavier toolchain.
Let’s take a look…
Antonio explains how in 2016, they used just 3rd party tools, making them extremely lean. Great start, but lots of challenges.
BUT the next step was not to fix the challenges, it was to align with the company strategy to increase the reach of the game. The company strategy says “we will introduce performance marketing”, so the data strategy became to support Marketing & Sales with that.
So they added a 5th tool to the already troubled stack! I love how they deliberately increased their data problems, to stay in line with the company strategy. In my opinion, this is exactly the right choice.
Of course then in 2018, they planned to centralize data. For their journey this was the perfect moment because they were now extracting much more value from the data, it wasn’t just used for exploration & deep diving anymore but also for performance marketing.
So they started to build a centralized data pot, deeply coupled to Azure. They realized this enabled them to run the whole chain just with 1–2 people!
From there on, they made a few very interesting choices. They had a decentralized data people organization from day one, the first hire being inside the marketing team, making their central data team a typical “platform team”.
I also stumbled over how he talks about “the loud minority feedback” which led them to make decisions on a customer feedback basis, which failed because they weren’t backed up by data. Data is important, a good point ;)
Another interesting piece is their data manifesto from 2020:
- 90% of all decisions (on Idle Miner Tycoon) need to be backed by data
- 90% of all data questions need to be answered in 1h or less
- 90% of all changes need to be validated with analytics.
Their current “datacision” cycle is pretty interesting. They built an agnostic pipeline. Meaning a product (or project) manager sits together with his analyst and they together model the events they need for their next decision, they get that data built-in and right through the event hub into the data lake/ data warehouse where the analyst & the product manager then can access it.
It means, for Kolibri, on every new feature it is the responsibility of the product manager to ensure tracking of the outcome of the decision. It means, product managers actually can do that easily, because the underlying tooling landscape is designed for that.