The Last Mile
A marathon. A triathlon. A PhD or parenthood. A data science project. Anyone who has worked hard at any endeavor can attest to the amount of work that goes in at the very end, just when you think all the big hurdles have been faced. Our walls here at the Flatiron school repeatedly exhort us to face the final hurdle, the last mile, when fatigue has set in. “Be Scrappy”, our walls whisper in silence.
As a conceptual framework, the last mile has wide usage in the world of business. Originally used by the telecommunications industry to describe the private costs of delivering service from a main communications network to an individual end-user as opposed to the shared costs of the main communications network itself, the term now is used widely in supply-chain management, transportation and bandwidth delivery.¹
Here I’ll be focusing on a specific usage we data scientists should all be paying attention to: obtaining buy-in for the insights you derived from decision-making stakeholders.
Let’s take a look at a standard depiction of the data science pipeline. Let’s suppose you and your trusty team of data scientists plot out your project. You know what question you are trying to answer, the laborious work mining and cleaning the data has been completed. You’ve examined the data, developed your features and predictive models, and finally you’ve created gorgeous visualizations that should properly being hanging opposite the Mona Lisa in the Louvre. You’re done right? Well, not really. Welcome to the last mile.
Harvard Business Review notes the problem in an article published early this year:
“In a question on Kaggle’s 2017 survey of data scientists, to which more than 7,000 people responded, four of the top seven “barriers faced at work” were related to last-mile issues, not technical ones: “lack of management/financial support,” “lack of clear questions to answer,” “results not used by decision makers,” and “explaining data science to others.”²
NewVantage Partners, a big data and business strategy consulting group, lends support to this view. NVP has been conducting a big data executive survey for a number of years now.³ The participating executives primarily came from the financial services industry with significant responses from healthcare executives. Among the challenges to business adoption, 85 % were attributable to cultural and organizational issues, the two largest concerns being a lack of organizational alignment and cultural resistance.
While a lack of organizational alignment can mean failure on the part of data scientists to apprehend the core issues of there company, it is far more likely that management has simply failed to foster that alignment by building data-centric organizations. Failing that, resistance to insights from data scientists is established as a cultural norm.
What is it that companies are looking for in a data scientist? Quite simply, someone who could do it all. A unicorn. Or perhaps more accurately borrowing from major league baseball, a five-tool-player: someone who can hit for average and power, run, field and throw . Paul Depodesta (caricatured by Jonah Hill in the movie “Moneyball), discussed the pervasive quest to find such a player:
‘And everyone wants five-tool players. There’s just very few of them on the planet! But virtually every player has one tool. So we started saying, “Well, let’s investigate each player’s strength. Is there a way to combine all these strengths and cover up some of their deficiencies? We’re not going to be able to do away with their deficiencies. But as a team, can we do it all?”’⁴
Companies are doing the same thing with data scientists. But to paraphrase Moneyball, instead of buying players, they should be buying wins. And to do that means looking at teams of players instead of one player who can do it all . Because even if there is someone who can do it all, isn’t 70- 80% of the data scientist’s working going to be spent on gathering sources and cleaning the data? If that’s the case, how much attention can really be paid on communicating with executives? And whose role is is to build-in technical competencies across an organization so that there is a data-centric culture? Is that the data scientist’s job as well?
To solve the last mile problem in data science will take a company-wide effort at addressing the gaps between what the data science team produces and executive decision making. A cross-functional team approach that builds on strengths and plans for shortfalls will yield better results than single-contributor players . Building-in training for employees who are not data scientists will help to foster that data centric-culture, the emergence of which will be crucial to breakdown the barrier between analytical insights and decision making.
[1] “Last Mile (Transportation).” https://en.wikipedia.org/wiki/Last_mile_(transportation)
[2] “Data Science and the art of Persuasion”, Scott Berinato, Harvard Business Review, January-February 2019, https://hbr.org/2019/01/data-science-and-the-art-of-persuasion
[3] “Big Data and AI Executive Survey 2019”, Thomas Davenport and Randy Bean, NewVantage Partners, 2019, http://newvantage.com/wp-content/uploads/2018/12/Big-Data-Executive-Survey-2019-Findings-Updated-010219-1.pdf
[4] Berger, Kevin. ‘Revisiting “Moneyball” with Paul DePodesta’. Nautilus, 22 August 2013, http://nautil.us/issue/4/the-unlikely/revisiting-moneyball-with-paul-depodesta http://nautil.us/.
