One hundred days of solitude

One hundred days of solitude

I regularly meet data scientists outside King, and, although it might sound outrageous, it is actually true. Moreover, sometimes I have an opportunity to speak to very highly positioned data scientists within a range of very prominent organisations. Read it again: I used the word ‘very’ twice in the same sentence – it should speak for itself. Such meetings, sometimes around a pint of beer or a cup of coffee, might generate some food for thought worth sharing. So once upon a time, a highly positioned data scientist within a prominent company, let’s call him for brevity Mr. Scientist, was talking about new starters and was mostly bemoaning the fact that it takes quite a long time for new starters in his organisation to start to perform. The topic seemed interesting to me and I continued the discussion not just because I wanted to look like a polite conversant.

One hundred days of solitude

-“So how long does it take?” I asked.

-“Six to eight months before we see any coherent results” replied Mr Scientist.

 I left speechless and the most I could do at that moment was to express a shadow of concern on my face and that is instead of saying something like, “No f-word way!”

Don’t get me wrong, I am not a condescending person and definitely not judgemental, but let’s try to analyse this carefully. The company where Mr Scientist is working is big and prominent as I said. Its work in the area of Data Science isn’t much different to what we are doing here at King, so that gives us an opportunity to have a common denominator and freely compare things.

However, let us start from a different point – let’s start from academia. The MIT Centre for Academic Excellence suggests: “On average freshmen will usually spend 18-20 hours in either lecture or recitation per week and an additional 34 hours preparing for a class.” That is for four modules per semester. This load is quite comparable with a working week of eight to nine hours a day. Now please note that when you come to your new working place you are not a freshman, but quite often M.Sc., or even Ph.D., who already knows how to study and definitely should perform better in terms of studies than a freshman even at MIT. I am on purpose not taking previous working experience into account. So looking at the numbers provided by Mr. Scientist, it means that a candidate they usually hire faces a task tantamount of learning four to five modules of something he has never been exposed to and the time it takes is equal to an average academic year.

Go to a kitchen, make a coffee, and meanwhile try to comprehend these numbers. Back already? Let’s continue then.

What are the most important topics for a data scientist? As I said above, the organisation where Mr Scientist is working isn’t very different to ours. The minimal requirements for a data scientist at King, and probably  across the whole globe, are good working knowledge of Python, Statistics, Inference and Modelling, SQL, and learning abilities. All the rest, such as Java, Hadoop, Spark, and Machine Learning, are definitely good bonuses, but not always compulsory. Certainly, in addition to it, one would have to learn a context in which the company operates. At King, it would mostly be mobile gaming, social networking, and in some departments direct marketing. Hence, it means that in that prominent organisation a starting period is comparable with an academic year in MIT for a freshman. Think six to eight months and four to five modules. I am not going to discuss if it’s valid or not, that is definitely up to each reader and Mr Scientist’s organisation to do so. Instead, I can try to compare it with how we look at it at King.

Probation period in any, even the most friendly company on the planet Earth – and King is definitely among the most friendly ones – might be a stressful period of time. It is the time when your new hires want to make an impact and show potential. They set out on a mission of delivering, but at the same time you as a manager or a peer set out on the mission of evaluating. Certainly there is no need to be super zealous and ‘look under the fingernails,’ but we definitely want to know if we are able to work with this person in future. It might be a very lonely period for a starter, sitting in front of a computer and quite often not daring to ask a question in order not to make a wrong impression. By the way, one should certainly pay attention to it, and if your starter doesn’t ask questions then something is wrong either with your welcome process or possibly with the new starter.

“Lost in the solitude of his immense power, he began to lose direction.”
– Gabriel García Márquez, One Hundred Years of Solitude

So what do we do here at King and especially in Marketing Data Science in order to get results and provide one with the experience of learning simultaneously? First of all, it can’t be any better than offering your new starter a clearly defined project, which she is able to accomplish during the probation period. Please note that I am talking here about regular probation period of three months. The first project is super important and it will set expectations for the future for both sides. Define clearly what you want in terms of work from your new hire in the coming three months. That means before you hire a person try to ask yourself the question: “Do I know what she is going to do when she arrives?” The generic answer – she would be doing work – is invalid. Set up a project where your starter will be able to have a clear impact and you will be able to make a proper evaluation. It can be any project, with the following condition: it must be manageable to finish it in the first three months or less of probation. Go to the dark corners of your backlog, there is always a project or two there that would have provided plenty of learning, but you are not starting them because there are a lot of routine short tasks and this particular project would require more than just an iteration. Building a new econometric model instead of an old one, re-specifying feature vectors for a machine-learning system, analysing the statistical distribution of the log time of your user’s anything that you always believed would be great to have, but couldn’t do it due to lack of resources or different priorities. Found it? Hooray, this is your moment and this might be a great project for your new starter. Lots of learning and a clear impact – these are the two key things. Write down a proper specification, identify areas of learning, set proper expectations, and here you go. So think about it: three months, accomplished project, context learning, a clear understanding of future work as opposite to an academic year of learning without actually delivering any work. Are you still thinking?

Join the Kingdom -

Share this article!

Michael Leznik

About Michael Leznik

As the Head of Marketing Data Science, Michael oversees all scientific analysis of marketing related activities at King. The team Michael leads is concentrated on the top view of the network of King players. Most of the analyses are econometric in their nature, while some work is concentrated on graph and classification-related algorithms. By closely following the guidelines of scientific research the Marketing Data Science team is providing King with an intelligent inside on network through the prism of economic theory with help of numerical algorithms and database related techniques. Michael lives in UK, but also worked in several different countries throughout his career. He enjoys reading, watching films, and British ales.