Search
  • Kendall Fortney

3 Important Product Management Pain Points with Data Science



It is hard to ignore the growth of Data Science and the explosion of the phrase “powered by AI” in company websites. While I do not have an advanced degree, any company that has a Data Science team should also ensure that their Product Managers understand the basics at the very least. My experience has shown that thinking a little ahead can have an exponential impact later on when you want to pivot a product. Below are 3 important points to help you on that journey:

Understand How Data will be Used

Working in any software company will teach you quickly that the greatest pain is that there is such a thing as “data debt” just like there is “tech debt.” Data Scientists will often spend lots of time cleaning data to get it into a state to be even marginally usable, but thinking about some of the potential problems when working on feature time can save time down the road.

Example

Adding a comment field to a post seems straightforward, store the comment ID, comment content, author and created date in the RDS and call it a day. But in reality that is just the beginning of questions you should ask yourself:

  • Should the author be stored as an unique ID or the string that makes up the author’s name?

  • Is it important to set the created date to the sec, or an even finer timestamp if there is a chance that comments could be submitted at a quick pace automatically?

  • Should the timestamp be UTC or the timezone of the author?

  • What kinds of characters are stored for the content and does that mean you might lose some emoticons or other special characters?

As Product Managers we are always serving as the voice of the customer, but sometimes the voice we also have to channel is the Data Scientist trying to understand the data three months down the road. They will thank you later.

Design for a Holistic Data Model

When looking at designing a new feature it is useful to ask how the resulting data could be used for enhancement of the product overall. This goes beyond your basic engagement but instead to thinking about how data from this feature could be combined with already existing data for better prediction.

Example

Adding “Contact for Price” to an ecommerce listing could be designed to drive lead generation and could have success measured on clicks. A good Product Manager would also consider how clicking on that button could indicate a difference in that buyer compared to other cohorts and the way a Data Science team may need to access that data for research and incorporation as a feature in other models.

Data models are inherently complicated, but it is important for a Product Manager to understand what are some of the categories of data and how they might interact. Sometimes it is a conversation with the Data Science and Engineering team, but often it is just stepping back and thinking about all the places a customer may interact with your platform.

Data Science is not Magic, it is Training Data

Like all things there are real limitations (often in surprising ways) that can crop up if you do not understand the quality and quantity of your training data. The powerhouse of Google and Facebook is not their money but the amount of data they can access.

Example

Classification of a user as a churn risk is a common business problem but there are many reasons someone may churn. Splitting out the data by features (like small, medium, enterprise accounts) can make the data sample even smaller. In the end it may be difficult to get any statistically significant result if you get too specific.

Being realistic about what can be delivered is not as fun as promising a perfect result, but that margin of error is important to have communicated prior to when, not if, the model gets it wrong. Understanding the size, data quality and diversity of the training data can be much more informative of scope for a feature than a corporate wishlist.

Conclusion

Product Management in the time of Data Science requires an evolution methodology to include the team on feature planning and development. Remember to understand how the data will be use, design for a holistic data model and thinking about training data will help your products launch with the right feature functionality!

0 views0 comments