Sunday, November 27, 2016

Data Science Can Help You Find A Good Doctor

In 2015, I led a team that built a data science driven product called Action. Action predicted people who might need care in the near future and helped them understand their benefits and their conditions better. Tens of leading companies in the US have Action now. Many more are in the process of implementing it.

The second data science driven project that I led at Castlight health was to address a seemingly simple problem that is very difficult to solve. This is the problem. How do you find a good doctor? How do you know if that doctor has treated many people for similar conditions before? Has the doctor kept up with advances in medicine. Does the doctor provide quality service?  While these sound like simple questions, the answers are not easy to get. Information about quality of health care providers such as doctors are very hard to find. You can find reviews on sites such as Yelp. While those reviews are helpful, they may not provide the necessary information to make an informed decision. For example, you may not get information about the number of times a doctor has performed a procedure. You may not get information about the quality of outcomes. This is compounded by the fact that many doctors work at multiple hospitals and healthcare facilities.

The good news is such information is available from multiple sources and we have been adding such information to  providers' profiles in Castlight Health for a long time. The bad news is such information is hard to decipher and verify. The data comes from varied sources in various intervals with varying levels of data quality. This is where data science comes in.  Data from multiple sources can be combined and matched with healthcare providers such as physicians, using rules generated based on machine learning models. Data scientists and product managers train the models to match data using a training data set and then apply the model to vast quantities of data.

After spending a few months on this problem, our product and data science teams made the first release recently. We made significant improvements to coverage and accuracy.  We will continue to make meaningful improvements every month to help people find a good doctor. It is important to note that there is nothing like clean data. Experienced data scientists and product managers in the data domain know that there is only less dirty data.

I am very pleased with the mission of the second project. I know that when the millions of people who use Castlight Health choose a doctor, there is a higher likelihood that they will pick a better doctor and as a result get a better outcome. When my daughter asks me what I do for a living, I plan to tell her that I help mommies and daddies find a good doctor for their children so that their children can be healthier.  I am pretty sure she is going to be impressed.

If you do not have Castlight Health at work, ask you benefits leader. It is a useful product to understand your benefits and make the best of it. 

Book - Ego Is The Enemy

I recently read the book "Ego Is The Enemy" by Ryan Holiday.  It is an interesting book that points out an important fact. In life and at work, if you take your ego out of the equation, work become a lot simpler and life becomes a lot more productive. It is hard to do but it worked for me a few times. Interestingly, my ego became a problem when my projects succeeded. That is the time when I needed something to ground myself. This book helped. Worth a read.

Monday, November 21, 2016

Book - Reinventing American Healthcare

The book, Reinventing American Health Care is a very good book for technology product managers who want to understand the history of the American Health Care industry, its current status and a glimpse into the near future. I moved from human capital management technology to health benefits management and predictive analytics technology at Castlight Health. Among the few books I read to understand American Health Care industry, this one is the best.

If you are working in the human capital management technology industry, health benefits management is one of the growth areas. In the coming years it is inevitable for large human capital management technology companies to expand into health benefits management and wellness management. In my opinion, health benefits and wellness management is where the next hundreds of millions in cloud software and services revenue is going to come from. This book is a good read for all the HCM product managers and HCM consultants who want to understand the fundamental problems of this new market. It is not an easy market. But good opportunities are never easy.

Saturday, August 06, 2016

How Can Product Managers Add Value In A Data Centric Enterprise Software Company

All enterprise software companies that I have worked in and have had the opportunity to interact with have mainly four groups of experts managing different aspects of their data. There are data driven products. These products are enabled by data intelligence products. Data intelligence products are enabled by an Enterprise Data Warehouse that organized and stores the data. The enterprise data ware house is fed by a data management team that monitors, received, prepares and audits data.




The Role of Product Managers In A Data Centric Enterprise Software Provider
My workplace is a data centric organization. The value and quality of data we have determine the kind of innovation we can build and deliver to our customers. At my workplace my colleagues and I have designed and applied product management tools such as design thinking and agile development to data driven products development and data intelligence products. We are not there yet on applying product management tools, practices and techniques to data management.

My hypothesis is that data management has become a critical foundational capability for enterprise software companies building data driven products. Today, it is managed like a information technology (IT) support service that is called upon when needed or yelled upon when something goes wrong. It is treated like a utility that is supposed to work.  Instead  it should be treated like a product organization whose capabilities could become a unique advantage for a data centric organization. It should be supported with product management skills, tools and techniques.

Product managers can add value to data management organizations by doing the following.


  • Identifying the capabilities that will add value to the business and build a barrier to competition.
  • Define and document the capabilities a data management team is building including acceptance criteria.
  • Convey the value created by a data capability for the business to executives.  For example, making a new data set available might enable the creation of new innovation. Making a data set available sooner and more frequently might be a market differentiator. It might keep competition away or may enable the business to charge more for current services. Building a file monitoring and audit capability might make the business more reliable, scalable and might help the business expand into new markets that were otherwise not financially feasible to operate in.


There are some hurdles and risks. 
Data management teams traditionally have operated as an information technology organization that did a few projects with longer term milestones. They may not understand or appreciate operating like a product development organization. However every product is going to be data driven and to succeed, every company needs to manage their data management organization like they manage a product organization. My belief is that those who master this will succeed. Those who don't will fail. If you are a product manager or a data analyst, I recommend that you join an organization that understands and appreciate this insight. Ask questions about how an organization operates before joining them. Interview data teams in a company and understand their acceptance or resistance to such a direction. Such due diligence will increase you chances of success in your job. It will also increase the success of the business.

If you are playing the role of data product manager and are working with a data management team, please share you experiences. I would love to know your thoughts. If all this conversation got you excited, consider joining me at Castlight. We are hiring for the Director, Data Intelligence product management role you see above. 

Friday, July 15, 2016

How can a product manager help a data science team be successful

I have been working with data scientists to build data products for over a year now. The first product we build is Castlight Action. Castlight Action is now being used by several large US employers. During this time I learned a lot about the role product managers can play while working with data scientists building a data or data intelligence product. Since the successful release of the Action product, we are taking some of the lessons learned in Action and applying them to other problems and products. These are my observations from both the experiences.

The Pre-requisite
Product managers can prepare themselves by getting some formal education on data science. I did this a couple of years back by getting certified on courses meant for data scientists. However, I believe that may not be necessary and may not be feasible for most product managers.

Last year, John Hopkins University released a certification in data science for executives. This is a good course for product managers to take. I took it this year and found it very valuable. Since then, I tried out some of the concepts in the real world. The concepts taught in this course work in the real world. I highly recommend this course for all product managers planning to or aspiring to work with data scientists.

The Key Responsibilities
I want to outline some of the key responsibilities of a product manager working on a data science product. This may not be a comprehensive list. I am listing the things I have observed so far. I may continue to update this list.

1. Define the purpose of the data science project
The product managers needs to define the capability your company will have once the data science project is complete. I can give you two examples from the work of my teams at Castlight Health. a) We wanted to identify segments of people who are similar and predict their healthcare needs for a certain period in the future. b) We wanted to look at two doctors with the same name in a directory of doctors and determine if the doctors are the same person or not. This is important to create a reliable directory of doctors for our users. Keep the capability description to one page. There is no need to write a long requirements document.

2. Break down the milestones and deliverables for the data scientists
Because any data science project is a research project, it is harder to breakdown work that falls into the cadence of a scrum team. This is an area where a product manager can play a useful role. The product manager need not tell the data scientist what to do. Most product managers will not have the skill to do so. How ever a product manager can specify what needs to be accomplished.  To do this a product manager needs to understand how data scientist works to solve a problem. The course on Data Science for executives from Johns Hopkins can help you with that. Don't just browse through the course. Get the certification.

3. Temper the expectations of your colleagues and business leaders
A product manager needs to explain a data science project and the expected deliverables in simple language to executives, who may or may not be educated about how data scientists work. It is important to explain that the results of a data science project are not always predictable. After a month of work the data science team may come to the conclusion that a particular model does not work as expected and may have to go down  a different path. It is the product managers responsibility to set reasonable expectations and communicate results.

4. Product Managers can perform tasks such as creating test data sets
Data product managers can even perform tasks in a data science project. For example, they can create a test data set to evaluate the efficacy of a machine learning model. Product managers do not need engineering backgrounds or have knowledge of programming to do so. I created a test data set recently for our doctor directory matching project using the tools, my data science colleagues created. They are command line tools. So some familiarity with command line tools and some curiosity about data are pre-requisites. Participating in such tasks will help product managers understand the data and the business problem intimately. It will also help you build credibility with the data scientists.

I plan to write more about the role of product managers in a data or data intelligence product. If you have played the role of a product manager for a product involving data scientists, please share your thoughts. You may have noticed that I use the terms data products and data intelligence products. I believe that data product are different from data intelligence products. More about that later.





Related Posts Plugin for WordPress, Blogger...