Why Data Scientists Must Focus on Developing Product Sense

Data Scientists should focus on developing product sense to move fast and systematically, create models that are relevant and to able to know when to stop.

By Nav Kesher, Facebook


Throughout my analytics career, people have asked me on what makes a good Data Scientist. You can see it in their eyes that they are expecting an answer in the realm of machine learning, neural networks or some coding paradigm; but then when I say “Product Sense”, they are surprised. Product Sense is one of the most under appreciated skill/quality that Data Scientists focus on. While everyone is busy churning models, its equally important to make sure that as a Data Scientist, you are delivering impact and giving actionable analytics that will move the product in the right direction. Product sense helps Data Scientists get better at the art of storytelling and as we all know stories are known to provoke thoughts and bring out insights that were previously not explained or understood.

Here are three reasons why I feel developing product sense is important for Data Scientists:

  1. It helps to move fast “systematically”: Data Scientists are expected to be efficient, decisive, and ready to learn and adapt. Understanding the product and the people problem it solves helps the Data Scientist set the goals for analysis and prevent scope creep in the future. Having a deep understanding of your product will also help you bootstrap models and improve feature engineering (from selecting indicator variables to understanding the relation between variables to feature representation and error analysis post-modeling)
  2. Models without product impact are of no use: As Data Scientists, we are expected to influence the direction of the product with data and actionable insights. But all our analysis and insights will be useless if we can’t convince our cross functional partners to act on our insights. In the end, no one cares how many features we engineered or how deep our neural nets go unless our work solves real people problems and improves our product in tangible ways that can be measured
  3. Know when to stop: Throughout our schooling years, we are trained to make sure everything is a good as it can be. Working on a product on the other hand expects to have something useful sooner than later (of course, there are many use cases, where regulatory or other relevant clauses will require you to be 100% accurate before shipping). As a data scientist, you may have to be okay not answering every last question you can with data or shipping a model that may can have marginal (and diminishing) returns with additional tuning. You can of course wait for 99% accuracy, however, in many cases by the time you get to such high accuracy, the milepost you were aiming to get to may itself have moved ahead

Now, that we have established that product sense is indeed important for Data Scientist, how do we go about building and improving product sense. Here are a few ways that have helped me improve product sense throughout my career

  1. Be involved in qualitative research and talk to your users: This is the best way to build empathy for your users and be sure that you are actually building products that solves their problems
  2. Don't discount your intuition and yet at the same time be data driven: I don't think anything will ever replace human intuition and in a product setting, it can be a handy tool for a Data Scientist as they are framing their analysis and prioritizing the questions they want to answer. A good product intuition combined with data driven logic is one of the most efficient ways to make an impact on the product
  3. Know your competition: It goes without saying that understanding the ways your competition is approaching similar people problem will go a long way in improving your own product. This also helps to understand and solve loosely defined problems by observing others and come up with actionable insights to guide product development decisions
  4. Eat your own dog food: The best way IMO, to build empathy for the end-users as well as iron out glitches and evaluate if your models are working in the way you intended them to. Most machine learning solutions of late have turned into black box models, and "dogfooding" your products is a great way (in addition to analytical approaches) to understand the impact of such models.

In today's world, it’s not enough to just play around with data and create models. We as Data Scientists are equal stakeholders in the product that is being shipped. I think that good Data Scientists are like quant product managers who understand how their work influences product decisions and solves people problems, which of course comes by developing great product sense.