150 Data Scientists and still no business value?

Harvard Business Review just published an article entitled Why You’re Not Getting Value from Your Data Science.

It says data scientists are overwhelmed by the complexity and quantity of data. Business experts, for their part, are underwhelmed by the output of those data scientists.

Why this failure to get business value?

The data scientists were obsessed with fine tuning complex models rather than formulating fresh predictive questions, losing site of the main purpose of their work: generating business value. In fact, when the author asked a room filled with 150 data scientists who had ever generated proven business value, no one raised their hand.


Me too. I would have expected about 5 of the 150 hands to go up.

Still, the point remains that the data science industry has indeed reached a point where a glut of unprepared and poorly led analysts has created a very high potential for failure.

From my experience leading and evaluating analytics programs over the years, I would call out three key problems commonly seen within data science programs:

  1. A failure to bridge the gap between business goals and analytic efforts

Math-y people focus on math-y work, typically producing a succession of faster, more accurate math-y methods. The ‘good but irrelevant’ data science tools are carefully crafted into ‘even better but still irrelevant’ tools.

2. A failure to work in a lean manner

Data science projects typically are highly exploratory, with uncertain outcomes. Focusing on a few high-risk efforts means a high likelihood of failure. It’s called ‘putting all of your eggs in one basket’, and it’s a bad idea.

3. Lack of discipline in measuring results

I’d bet that most of the 150 data scientists surveyed in this study did produce measurable business value, despite not knowing it. I’d also bet that no one had the discipline to measure the results.

Four tips to start getting business value from your data science efforts:

Here is my advice for data science programs

  1. Live and breathe business intuition. There are certainly people in the company with extensive experience with the customer, the product, the market. Stalk them. Go back to them every few days and show off your data and initial results. They’ll laugh at you when they see you doing something blatantly wrong. You’ll start doing things right pretty quick.
  2. Measure Results. Don’t start a data science project unless you know why you’re doing it and what it looks like when it succeeds. As they say, ‘if you’re not keeping score, you’re just practicing’
  3. Work in an agile manner. Agile project management is key to making an analytics program effective, and it ties in closely with the first point above. Business intuition will fuel the feedback loops and provide input for specifying many of the project deliverables. Work in terms of minimum viable products and short delivery cycles.
  4. Appoint a leader with understanding of both business goals and analytic possibilities. Ideally, one person can match business value with analytic potential and lead a team of data scientists in attaining high value results. Finding such a person can be difficult (see my article Recruiting a Chief Data Scientist).

As with most things that are worth doing, making a data science program effective can take substantial effort and will require several iterations before the program is structured in an effective way. Don’t give up if the initial program seems to be failing. As Mckinsey said in their recent report, The age of analytics: Competing in a data-driven world, leading firms which have already developed strong analytics programs are not only winning in their own fields, but are actively looking for ways to disrupt adjacent industries. Don’t get disrupted by them.

  1. https://hbr.org/2016/12/why-youre-not-getting-value-from-your-data-science
  2. http://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/The-age-of-analytics-Competing-in-a-data-driven-world

The full version of this article appears on my data science blog.

I’d love to hear peoples’ thoughts and experiences in the comments below.