By Anne Fischer, Senior Director, Advanced Analytics, Truven

Big business is officially obsessed with big data. According to a new survey of Fortune 1000 C-suite executives conducted by NewVantage Partners, 63 percent said they now have at least one big data project in production, nearly double the level reported in 2013.1 Moreover, 27 percent of respondents said they’d spend over $50 million on big data by 2017, up from just 5 percent of respondents in 2014.

In healthcare, where the trend toward big data adoption is off the charts, organizations of every type – providers, employers, health plans, life sciences firms, and even federal and local governments – are using this data for everything from decision support to point-of-care treatment. Along the way, the various professionals involved have had to come to grips with the huge communications challenges that emerge when you put a bunch of data scientists and business executives in a room and ask them to work together toward a common goal.

The result is often the stuff of Dilbert comic strips, where a simple inter-office miscommunication results in a cascade of bad decisions, and chaos ensues. Data scientists and business executives come from very different places and, though they are increasingly learning from one another, we are currently at an awkward phase in the adolescence of big data development where both sides know just enough about the other’s role to be dangerous.

It is within this realm of semi-understanding of one another’s languages that some of the biggest disconnects occur. Consider the following not-so-imaginary scenario with a request by a healthcare executive to one of her data scientists:

“I need you to build a population stratification model that will identify high-risk patients using ensemble methods and incorporate socio-economic data to support population health managers in identifying actionable intervention opportunities while encouraging value-based care. By next Tuesday.”

This request exemplifies several common mistakes that business-oriented groups often make in working with data scientists. As a data scientist, a few of the immediate questions
I would have for my boss are:

  • Patients at risk of what, exactly? Death? Hospitalization? High cost?
  • Exactly what kind of data, socio-economic or otherwise, will be available?
  • Why do you think an ensemble technique is appropriate?
  • Who are population health managers, and what kind if interventions do they provide?
  • How is value-based care measured?
  • Are we identifying individual patients or whole groups of people “at risk”?

When the data scientist is given the opportunity to clarify these and other points, a successful outcome is far more likely. With that in mind, here are four tips to help business-oriented colleagues communicate more effectively with data scientists.

1. Be specific about the business objective

When defining a project for a data scientist, it’s important to be specific about the desired outcome of the development methodology. Critical questions to address include:

  • Who will be using this model, and what is their role?
  • What business problem are they trying to solve?
  • In what context and how frequently will the model be applied?
  • What action will be taken based on the results?
  • What data sources/elements are available?
  • What level of accuracy is required?

2. Avoid excessive industry jargon/provide a translator

While you may operate in the world of marketing and customer relationships, most data scientists do not. In fact, many are not even domain experts in the particular industry in which they work. Just as your eyes may glaze over when they talk about positive predictive ratios or neural networks, theirs probably glaze over when you talk about value-based care and population health initiatives.

Communication between groups always involves a certain amount of translation so that everyone understands exactly what is being said. An often-shared cartoon in software development circles uses an image of a swing to indicate how initial business requirements can morph into a completely different solution as we go through the phases of software development. To help mitigate that, a business analyst or product designer may play the translator role within software development. Don’t underestimate the importance of that same translation role in working with data scientists.

3. Stop expecting miracles

While the market hype would lead one to believe that applying predictive analytics and machine learning to big data will solve the world’s problems overnight, the truth is much more complicated. Big data, data mining, machine learning – all of these are tools that can enable better solutions, but real value requires real investment, both in the resources utilizing these tools and in the amount of time dedicated to finding the right solution. Hiring experts and adequately preparing them to succeed is paramount. Developing the appropriate solution is time consuming, even if the tools allow for a quicker move from raw data to “analysis.” A statistical model can be created very quickly, but a true solution to a business problem takes more time.

4. Let the data scientist do the data science

Finally, allow the data scientist to make decisions related to statistical techniques, important data elements, and appropriate application of the resulting method. This is where their training and expertise come into play. Telling a data scientist what statistical method to use to solve a problem is like telling a software engineer what programming language to use. No one likes to be told how to do their job, especially by someone who is less of an expert. This is their specialty; allow them to shine.

While these techniques don’t guarantee you a Dilbert-free zone, they will go a long way to improving the end result of the models and methodologies created by your data scientists.


  1. http://newvantage.com/wp-content/uploads/2016/01/Big-Data-Executive-Survey-2016-Findings-FINAL.pdf