Amongst the various rock stars that a product manager collaborates with on a daily basis, ‘data scientists’ are now considered crème de la crème. They have an armoury of tools that gives them perspectives that mere mortals can never imagine on their own.
As a product manager, you are continuously looking for improvements even in mature products that work reasonably well. Just imagine how tough it is to improve a product like Gmail, which is a product that works really well.
One very pragmatic approach to improve an existing product is to:
- Identify obstacles in task completion for the user and remove themOr
- Implement ideas to improve already efficient task completion
In this context, it is always wise to understand the possibilities that machine learning & data sciences offer, especially for internet products where data collection is relatively easy.
The contribution of data sciences start from simple tests for statistical significance of A/B test results when you compare different solutions to pick the best among them.
Filling a job application form for a user can be made simpler with text parsing algorithms that can read from a doc resume and fill forms.
More complicated machine learning algorithms categorise emails into spam, promotions and so on in Gmail, helping users to focus on emails that matter the most. If you have noticed closely, while composing an email, Gmail offers suggestions for the next email address to be added in the To: list – to make it easier for the user to complete the to list.
A very successful product discovery feature-‘recommendations’ used across e-commerce and classified portals is the contribution of a variant of collaborative filtering or a similar advanced algorithm.
It is interesting to notice that many of these algorithms along with the expertise to customise or develop them are accessible to many companies at a reasonable cost. Companies that are shy of hiring talent can even access these algorithms in pay-per-use model from vendors.
However what is crucial is identifying the right problem to be solved with these tools.
One approach is to identify the features that worked well for other industries and see if they can be modified to suit your industry. ‘Product recommendation’ is a very successful example which has worked across industries. Remember the ‘people who viewed this also viewed’ recommendations ?
Another approach is to convert tacit knowledge and biases of of domain experts that help them solve a problem in their day to day work into algorithms to make the process scalable and error resilient. For example senior customer service managers will be able to sense fraudulent classified listings from the information that is provided in the listing. Interviews with few such managers can reveal good insights that help in feature selection for building a model for flagging fraudulent classified listings.
This is just the tip of ‘data sciences’ ice-berg.