公司如何用AI定价
How companies use AI to set prices
公司如何用AI定价
The pricing of products is turning from art into science
产品的定价正从技巧转为科学
FEW AMERICAN business tactics are as peculiar in a freewheeling capitalist society as the manufacturer’s suggested retail price. P.H. Hanes, founder of the textile mill that would eventually become HanesBrands, came up with it in the 1920s. That allowed him to use adverts in publications across America to deter distributors from gouging buyers of his knitted under garments. Even today many American shopkeepers hew to manufacturers’ recommended prices, as much as they would love to raise them to offset the inflationary pressures on their other costs. A growing number, though, resort to more sophisticated pricing techniques.
在美国这个自由放任的资本主义社会，很少有商业策略像“制造商建议零售价”那般古怪突兀。这个定价机制是P.H.哈内斯（P.H. Hanes）上世纪20年代提出的，他创建的纺织厂最终发展成了汉佰百货（HanesBrands）。他在美国各地的出版物上刊登建议零售价的广告，防止经销商对购买他的针织内衣的买家漫天要价。即使在今天，许多美国店家仍然坚守制造商建议的价格，尽管他们也很想抬价以抵消通胀带来的其他成本压力。不过，越来越多的公司正诉诸于更复杂的定价方式。
A seminal study from 2010 by McKinsey, a consultancy, estimated that raising prices by 1% without losing sales can boost operating profits by 8.7%, on average. Getting this right can be tricky. Set prices too high and you risk losing customers; set them too low and you leave money on the table. Retailers have historically used rules of thumb, such as adding a fixed margin on top of costs or matching what competitors charge. As energy, labour and other inputs go through the roof, they can no longer afford to treat pricing as an afterthought.
咨询公司麦肯锡在2010年开展的一项开创性研究估计，在不损失销售额的情况下，每提价1%平均可以让营业利润提升8.7%。要把握好这个度会很难。定价过高，可能会失去顾客；定价太低，就得不到应得的最大利益。零售商历来都采用经验法则，比如在成本之上加上固定的利润额，或者跟上竞争对手的开价。随着能源、劳动力和其他投入的价格飞涨，它们已经承担不起把定价放到后头考虑了。
To gain an edge, shopkeepers have been turning to price-optimisation systems. These predict how customers will respond to different pricing scenarios, and recommend those that maximise sales or profits. At their core are mathematical models that use oodles of transaction data to estimate price elasticities—how much demand increases as the price falls and vice versa—for thousands of products. Price-sensitive items can then be discounted and price-insensitive ones marked up. Merchants can fine-tune the algorithms to prevent undesirable outcomes, such as double-digit price surges or larger packages costing more by unit of weight than smaller ones.
为能驾驭局面，零售商们已经转而使用价格优化系统。这些系统预测顾客对不同定价方案的反应，推荐那些让销售额或利润最大化的方案。系统的核心是利用大量交易数据来评估成千上万种产品价格弹性的数学模型。所谓价格弹性，就是随着价格的升降，需求增加或减少的程度。然后对价格敏感的商品可以打折，对价格不敏感的商品可以加价。商家可以微调算法以避免不合理的后果，比如两位数的大涨价，或者大包装的重量单价反而高于小包装。
These systems are becoming cleverer thanks to advances in artificial intelligence (AI). Whereas older models used historical sales data to estimate price elasticities for individual items, the latest crop of AI-powered ones can spot patterns and relationships between multiple items. Makers of pricing software are incorporating new data sources into their models, from customers’ tweets to online product reviews, says Doug Fuehne of Pricefx, one such firm. The cloud-based platform developed by Eversight, another provider, allows retailers to test how slight increases or decreases in the price of, say, Heinz ketchup at different stores affect sales not just of that specific condiment but across the category. It is used by big manufacturers such as Coca-Cola and Johnson & Johnson, as well as some supermarkets (Raley’s) and clothes-sellers (JCPenney).
随着人工智能（AI）的进步，这些系统正变得愈发聪明。旧的模型用历史销售数据估算单个商品的价格弹性，而最新一代AI模型可以发现多个商品之间的模式和关系。定价软件的开发者正在把顾客的推文和网上产品评价等新的数据源整合进自己的模型中，道格·福涅（Doug Fuehne）表示。他就职的软件公司Pricefx就是其一。另一家软件公司Everight开发了基于云计算的平台，可以让零售商测试一种商品价格的小幅上调或下调——比如对不同商店里的亨氏番茄酱调价——会如何影响该商品的销量，乃至整个品类的销量。使用这一平台的有可口可乐、强生等大型制造商，也有一些超市（Raley's）和服装卖场（JCPenney）。
All this makes pricing systems “much more three-dimensional”, observes Chad Yoes, a former executive at Walmart who oversaw pricing at the retail behemoth. Retail bosses are keen to promote this sophistication to investors, who value firms’ pricing power at a time of high inflation. In February Starbucks, a chain of coffee shops, boasted about its use of analytics and AI to model pricing “on an ongoing basis”. US Foods, a food distributor, has touted its pricing system’s ability to use “over a dozen different inputs” to boost sales and profits.
所有这些都让定价系统变得“立体得多了”，曾在零售巨头沃尔玛负责定价的前高管查德·尤斯（Chad Yoes）表示。零售业的老板们热衷于向投资者宣传这类复杂精密的系统，因为投资者在高通胀时期看重公司的定价能力。今年2月，连锁咖啡店星巴克声称自己利用分析技术和AI为定价“实时”建模。食品分销商美国食品（US Foods）自诩其定价系统能够运用“十几种不同的输入数据”提升销售额和利润。
Price-optimisation may make prices more volatile. “Retailers are pricing faster today than they ever have before,” says Matt Pavich of Revionics, another pricing-software firm. That is especially true in the fast-moving world of e-commerce. But even Walmart reviews the prices of many items in its stores 2-4 times a year, says Mr Yoes, up from once or twice a few years ago.
价格优化可能会让价格更不稳定。另一家定价软件公司Revonics的马特·帕维奇（Matt Pavich）说：“现在零售商定价比以往任何时候都要快。”在瞬息万变的电子商务世界尤其如此。但尤斯表示，就算是沃尔玛，每年也会对店内许多商品的价格做二到四次评估，而几年前只做一两次。
What pricing systems do not do is lead inexorably to higher prices. Mr Pavich calls this misconception “one of the biggest myths” about products like his. Sysco, a big food distributor which rolled out new pricing software last year, is a case in point. The firm says the system allows it to lower prices on “key value items”—as price-sensitive bestsellers are known in the trade—and raise them on other products. It can thus increase profits by expanding sales while maintaining margins. That keeps investors content and shoppers sweet. ■
定价系统并不是让价格只涨不跌。帕维奇称这种误解是对像他的这类产品“最大的谬见之一”。去年应用了新定价软件的大型食品分销商Sysco就是一个很好的例子。该公司表示，该系统让它能够降低“关键价值商品”（业内对价格敏感畅销品的称法）的价格，并提高其他产品的价格。这样一来可以扩大销售而不损害利润空间，从而增加盈利。这既让投资者满意，也讨好了顾客。
