I’ve been in the data game for a while now and this is what I’ve learnt about data scientists and what they actually do.

Let’s start with the fountain of infallibility, Wikipedia, on their definition:

A data scientist is a professional who creates programming code and combines it with statistical knowledge to create insights from data.

That is pretty broad - my personal way of describing it is more like:

One who works with machine learning models and predictive statistics.

My current opinion is that there are three subspecies of data scientist out there, undoubtedly with some overlap but defined by the majority of work you do.

The cutting edge creator

My first archetype is the scientist who is creating new types of models and researching new statistics to explain the world with. Creators like Jeff Dean or Tianqi Chen would easily be in this category along with many others from OpenAI, Deepmind, MIT, but also individual contributors from around the world improving the source code to models. These are extraordinarily impressive people and organisations without whom many ‘data scientists’ would not have a job. I estimate that less than 1% of data scientists fit into this category.

The applied modeller

My second archetype is the data scientist who applies a predictive model to their business problems using existing model libraries. They can be extremely valuable, able to build revenue forecasting models or support centre demand models or product recomendation models or whatever the business needs of them. When most businesses look to hire a data scientist this is who they hire. I’d put myself in this category, and would guess that this makes up perhaps 60% of data scientists.

The descriptive analyst

My third archetype is the data scientist who doesn’t build models but instead is tasked with what is more accurately described as analytical work. They may create visualisations, undertake ad hoc descriptive analysis, help wash customer data against the business, build data pipelines or a number of other things but they’ll often do it with Python or R giving an image that what they are building is the same as the above two data scientists. This doesn’t really meet my personal definition of data scientist but does meet the Wikipedia definition. In general, a title of data engineer or data analyst is more appropriate here but many businesses know the scientist title is sexy so will use that instead. I’d guess about 40% of data scientist roles out there are best described as this.

Why this matters

Employers want the right data specialist for their business and employees want the right role for their skillsets and aspirations. if your business is in its early stages of data maturity there is no way you want an applied modeller or worse a cutting-edge creator, and immature firms can be the biggest culprits for this mishiring. For an immature firm, they want the descriptive analyst, a role perhaps more efficiently found by advertising for a data analyst than a data scientist. Otherwise, a newly hired modeller’s talents are wasted and they will need to do the analysis and engineering work anyway and just get disheartened and likely quit.

Now, I don’t want to come across as a gatekeeper on the position title. Titles are social constructs which often fail to capture what someone does anyway. It’s just a few words someone made up after all. I would love for a greater respect for the analyst title and am content with more widespread usage of the scientist title. I guess we could do worse than having more self-described scientists out there.