搞管理的与搞数据的如何和谐共处

    比方说，你需要做出一项重要的战略决策，而分析师团队已精心建立了一个复杂的数学模型，它应该会告诉你往哪个方向走。问题是，即使数据科学家按照他们认为简单明了的表达方式阐述了他们统计算法的详细情况，你还是一窍不通。

    不要惊慌。在一本题为《跟上量化分析师：了解与使用数据分析的指南》（Keeping Up with the Quants: Your Guide to Understanding + Using Analytics）的新书中，托马斯•达文波特与合著者金振浩（音译）试图就如何明智使用大数据向企业管理人员提供了建议，其中包括提出哪些问题，如何判断量化分析师是否真正理解他们旨在解决的业务问题。

    麻省理工学院电子商务中心（the MIT Center for Digital Business）研究员、哈佛商学院（Harvard Business School）客座教授达文波特此前还曾出版过两本有关量化分析的著作。他在这本新书中引用了英国著名统计学家乔治•伯克斯下面这句名言：“所有的（数学）模型都是错误的,但其中有些模型是有用的。”更有用的是富有经验的管理者的直觉。 达文波特写道：“很少有企业管理人员在分析和直觉两方面都擅长。因此，我们的目标就是做出善于分析的决策，同时让管理人员的直觉发挥作用。”

    举例而言，卡尔•肯普夫就认同这种看法。肯普夫是一位资深科学家，负责领导英特尔（Intel）的一个决策工程部门，他在公司的绰号是超级量化分析师（UberQuant）和首席数学家。即便如此，肯普夫依然认为，良好的量化决策“并不取决于数学。而是取决于决策人员与分析人员之间的合作关系。”达文波特指出：“如果被称为首席数学家的某个人声明这并不取决于数学，那么我们应该引起注意。”

    《跟上量化分析师》引人入胜地详细叙述了英特尔及其他成功的公司——包括威瑞森移动公司（Verizon Wireless）、多伦多道明银行集团（TD Bank Group）及默克制药公司（Merck）——如何帮助管理人员和数据科学家加深彼此了解，从而实现有效合作。就英特尔而言，肯普夫把负责解决某个问题而资历较浅的“数学分析人员”与不擅长数学的人员一起派遣到海外，让他们聆听、学习及获得一些一般商业知识。

    达文波特写道：“至多，正如新员工那样，分析人员经过培训后可以参与业务流程。肯普夫认为，衡量这方面培训成功的最低标准是数学分析人员自己认为自己理解了业务问题，而最高标准则是企业管理人员认为数学分析人员理解了业务问题。”

    对于企业管理人员而言，他们可能需要复习代数学。达文波特写道：“比如，企业管理人员不必明白双曲型偏微分方程。（大家可以松口气了。）但复习课堂的白板上至少必须有一幅图表，下面列出这样的问题：‘由于A和X呈正相关关系，如果A上升，那么X的变化方向是什么？’” 他补充说：“与任何其他类型的模型一样，几个具体的例子（无论是历史、还是预想的例子）非常有用。”诸如饼分图和条形统计图等视觉辅助工具也是如此，在默克公司担任商业分析部门主管的帕特里克•摩尔最喜欢这类工具。

    Let's say you've got a crucial strategic decision to make, and a team of analysts has painstakingly built a complex mathematical model that's supposed to show you which way to go. The trouble is, even after the data scientists have laid out the details of their statistical algorithm in what they think are simple terms, it's Greek to you.

    Don't panic. In a new book called Keeping Up with the Quants: Your Guide to Understanding + Using Analytics, Thomas H. Davenport and co-author Jinho Kim set out to advise executives on how to make sensible use of big data, including which questions to ask and how to tell whether the quant jocks really understand the business problem they're purporting to solve.

    Davenport, a visiting professor at Harvard Business School, a research fellow at the MIT Center for Digital Business, and the author of two previous books about quantitative analysis, quotes eminent British statistician George Box: "All [mathematical] models are wrong, but some are useful." Even more useful is seasoned managers' intuition. "Few executives are skilled at both analytics and intuition," Davenport writes. "The goal, then, is to make analytical decisions while preserving the role of the executive's gut."

    Karl Kempf, for one, agrees. Kempf is a senior scientist who heads a decision engineering group at Intel (INTC), and whose nicknames around the company are UberQuant and Chief Mathematician. Even so, Kempf believes that good quantitative decisions "are not about the math. They're about the relationships." Notes Davenport, "If someone referred to as the Chief Mathematician declares that it's not about the math, we should pay attention."

    Keeping Up with the Quants goes into fascinating detail about how Intel and other successful companies -- including Verizon Wireless (VZ), TD Bank Group (TD), and Merck (MRK) -- help managers and data scientists understand each other well enough to collaborate effectively. In Intel's case, Kempf sends the "math people" charged with solving a problem on a kind of junior year abroad among non-math types, to listen, learn, and pick up some general business knowledge.

    "At most, the analyst can be trained, as a new hire would be, to participate in the business process," Davenport writes. "Kempf judges the low bar for success as when the math person thinks he or she understands the business problem. The high bar is when the business person thinks the math person understands the business problem."

    For their part, executives may need to brush up on their algebra. "The business person doesn't have to understand, for example, hyperbolic partial differential equations," Davenport writes. (Well, there's a relief.) "But at a minimum there has to be a diagram on the white board setting out such questions as, 'Since A and X are related, if A goes up, in what direction does X go?'" He adds, "As with any other type of model, a few concrete examples -- historical or made up -- are extremely useful." So are visual aids like pie charts and bar graphs, a favorite tool of Patrick Moore, who heads the commercial analytics group at Merck.