Adding value as Data Scientists

I’ve been thinking and discussing with various people lately – ‘career path for data science’. Someone said to me recently:

• Go become a research scientist and specialise in specific machine learning models say NLP at a specific company such as Google, Amazon, etc.
• Become a data scientist at a startup or growth company and accept that you will spend most of your time doing software engineering.

You will notice that I don’t say ‘Well I can’t get into Jeff Dean’s team at Google – I’ll go do that job at a startup’. That never works

The reason it never works is that organisations largely need to solve problems. So that is either adding value by presentations or presenting analysis – this is often not aligned with the business needs of early startups or growth companies – particularly those that need to build systems. 

So that means realistically – a la – you need to be a software engineer first. This means being capable of passing an engineering bar for your employer. This also means that spending time learning more ‘full stack’ skills is a good use of your time.

So if you want to be a valued and successful and productive data scientist (or what I’m actually describing is perhaps a machine learning engineer) spend some of your time learning good practices in terms of code review, testing, logging.

To tell one anecdote – I consulted once with a company who had spent 3 months with a consultant building ‘machine learning models’. The code was unmaintainable, it took months to convert it to production ready code. And in some sense – and this is sad – the company wasted money. They got models but no infrastructure to deploy it/ use it in a product.

Therefore – in your first 3 – 6 months at a job – remember to focus on providing value end-to-end by building something and that means picking a non-risky project. Not spending 3 months on a deep learning model when a logistic regression will do.

Provide end-to-end value – whether that’s a dashboard, a data pipeline, a model in production or a presentation and interfacing with other teams to get something changed in a product page

Your job as a data scientist is to add value. If you want to work for deep mind – go work for deep mind.