大数据并非万能

Clay Christensen tells a good joke about a tour of heaven. “How come there’s no data here?” the Harvard professor asks his celestial guide. “Because data lies,” comes the response. And that is why, Prof Christensen goes on, “whenever anyone says ‘Show me the data’, I just say ‘Go to hell’.”

克雷•克里斯坦森(Clay Christensen)讲了一个有关天堂旅游的有趣笑话。“这里怎么没有数据呢?”这位哈佛教授问他的天堂向导。“因为数据撒谎，”对方回答说。克里斯坦森教授接着讲，所以“每当有人说‘把数据拿给我看’时，我就会说‘下地狱去’”。

The gag got a laugh at last week’s Drucker Forum in Vienna, where fans of the late Peter Drucker’s claim that management is a “liberal art” voiced fears about the way data are wielded to crush human insight and inventiveness.

在近期在维也纳举行的德鲁克论坛(Drucker Forum)上，这个笑话引起了笑声。在论坛上，认同已故彼得•德鲁克(Peter Drucker)的管理属于一门“文科”观点的粉丝们，表达了对数据被用来碾压人类洞察力和创造力的担心。

But there are signs of a backlash against big data even where it has loomed largest. As chief executive of UK supermarket chain J Sainsbury until 2014, Justin King commanded a data set that showed, for instance, that purchases of diet products were the best indication that customers were planning to go on holiday — and that they might therefore be open to some deft direct marketing of suntan lotion.

但目前有迹象表明，即便在大数据运用最广泛的领域，大数据也遭遇了强烈反弹。比如，担任英国连锁超市森宝利(J Sainsbury)首席执行官直至2014年的贾斯廷•金(Justin King)掌握的一个数据集显示，购买减肥食品是顾客打算去度假的最佳信号，因此他们可能很容易接受某些精明的防晒霜直接营销。

He believes retailers should use such information to represent the shopper better in, say, negotiations with suppliers. But at a Financial Times 125 Forum I chaired recently, he said he worried data were now used against customers. He has, for instance, criticised the use of loyalty card data to “game the customer” by offering them vouchers to switch brands.

他认为，零售商应当使用这类数据——比如在与供应商的谈判中——更好地代表顾客。但在不久前我主持的英国《金融时报》125论坛(FT 125 Forum)上，他表示，他担心如今数据的使用是不利于顾客的。例如，他对利用积分卡数据“算计顾客”、通过提供代金券诱使他们转换品牌的做法提出了批评。

It is too soon to declare the triumph of what one ex-colleague used to call “big anecdote” over the ideology of easy-to-measurism that has held boardrooms in thrall for the past few years. For example, the hastily declared failure of pollsters to predict a Donald Trump victory in the US election is more likely to be due to unsound one-on-one surveys than yawning deficiencies in wider data-gathering. The science of data analytics, when combined with cognitive computing and even neuroscientific and behavioural research, is also going to get more sophisticated and precise.

现在要宣称我的一名前同事所称的“重磅轶事”相对于“易于衡量”观念——过去几年企业董事会牢牢奉行这种观念——取得了胜利，还为时尚早。例如，有人仓促宣布民意调查机构未能预测到唐纳德•特朗普(Donald Trump)在美国大选中获胜，但预测失败的原因更有可能是不可靠的一对一调查，而不是宏观数据收集方面的巨大缺点。数据分析科学，跟认知计算、甚至还有神经科学与行为研究结合在一起，也将变得更先进、更精确。

For now, some of the tools measuring customer satisfaction are as blunt as those smiley-face pads you find at airports, asking you to assess your experience. I still wonder how the airline I flew with last summer interpreted the input from the cheerful toddler who was repeatedly stabbing the angry-face icon on the machine at our departure gate.

目前，有些衡量顾客满意度的工具就像你在机场发现的邀请你给旅途体验打分的笑脸打分板一样生硬。我仍在好奇，今年夏季我乘坐飞机的那家航空公司，对于那个开心的学步小童反复去戳登机口旁那台机器上的愤怒脸图标意味着什么如何解释。

Separately, Facebook — whose access to vast user-created troves of information retailers and airlines can only dream about — has got into trouble with its advertising customers after admitting mistakes measuring the time users spend viewing video advertisements and articles.

另外，Facebook在广告客户那里遇到了麻烦，因为Facebook承认，在衡量用户观看视频广告和阅读文章的时间上出了错误。Facebook掌握着零售商和航空公司只能梦想一番的海量用户生成信息。

Too often, computer-generated “facts” come close to overruling common sense. When Pope John Paul II died in 2005, a senior editor noted that the news had surged to the top of the FT website’s most-read stories and ordered me (I was then editing our opinion pages), to commission insights into Vatican policies, Catholic mores and papal history — none of which was a hit. Three days later, Saul Bellow died. His obituary also topped the rankings. There was no corresponding call to deepen our coverage of US novelists and their work.

有太多时候，计算机生成的“事实”几乎碾压常识。当2005年教皇约翰•保罗二世(Pope John Paul II)去世时，一名资深编辑注意到，该消息已猛升至英国《金融时报》网站热门文章首位，然后命令我(当时我是观点版面的编辑)约一些有关梵蒂冈政策、天主教习俗和教皇历史的分析文章，结果这些文章没有一篇受到追捧。三天后，索尔•贝娄(Saul Bellow)去世，他的讣告也登上了榜首，但没人打电话让我们做美国小说家及其作品的深度报道。

Insights from only a few users can still be valuable. Mr King advises against ignoring the shopper who complains she waited 15 minutes at the self-service tills, even if your spreadsheet shows the average wait was two minutes. Her perception that it took much longer may tell you more than whole dashboards of data.

就算只是少数用户的意见，也可能很有价值。金建议，不要忽视抱怨自己在自助收银机那里等待了15分钟的顾客，即使你的电子表格显示平均等待时间是2分钟。她感到等待的时间长得多，这或许能告诉你全部数据以外的东西。

Similarly, asked what Spotify would do with the “customers from hell”, Joakim Sundén, senior tech leader at the music streaming service, told the Drucker Forum that their “deep pain” might be telling you about a problem you had not identified.

同样，当被问到Spotify如何应对“来自地狱的顾客”时，这家音乐流媒体服务公司的资深技术主管若阿基姆•松登(Joakim Sundén)在德鲁克论坛上说，他们的“深度痛苦”或许正在告诉你一个你之前未曾发现的问题。

Remember, too, that there are some situations in which data may never be much help. One is innovation, where the tyranny of the business plan cramps ideas and narrows options, according to experts gathered in Vienna last week. As Rita Gunther McGrath of Columbia Business School puts it: “It’s always easier to go back to the spreadsheet.” Roger Martin, who heads the Rotman management school’s Martin Prosperity Institute, says he would ban the word “proven” from organisations that wish to innovate. “It’s hard to explore possibilities if you have to know the answer before you start,” adds Tim Brown, chief executive of Ideo.

也要记住，在某些情况下，数据或许永远帮不上大忙。德鲁克论坛上的专家认为，一个是创新，专横的商业计划束缚了思想，局限了选项。正如哥伦比亚商学院(Columbia Business School)的丽塔•冈瑟•麦格拉思(Rita Gunther McGrath)所说：“回去看电子表格，总是更容易的。”罗特曼管理学院(Rotman School of Management)马丁繁荣研究所(Martin Prosperity Institute)所长罗杰•马丁(Roger Martin)说，他会禁止希望创新的机构使用“经过验证的”这个词。“如果你必须在开始前知道答案，那就很难探索可能性了，”Ideo首席执行官蒂姆•布朗(Tim Brown)补充说。

Knowing your customer will never be a zero-sum contest between a researcher with a clipboard and IBM’s Watson. Nor should it be. The best insights come from some hard-to-define blend of what you know from listening to individual users, what you can learn from their collective past behaviour and what you intuit they will want in future. The really flawed assumption is that a capsule of data inserted into the analytics machine will always generate the perfect brew.

理解你的客户，永远不是拿着带夹子的写字板的研究人员和IBM的沃森(Watson)之间的零和竞争。也不应该是。最好的理解产生于一种难以定义的混合认知：你倾听单个用户所了解到的东西，你从他们的集体过往行为中学到的东西，以及你从直觉知道他们未来想要的东西。真正错误的假设是，把一些数据输入分析机器，总会生成最佳答案。