Stitch Fix：利用大数据卖衣服

只用了8年时间，在线零售商Stitch Fix就将业务做得风生水起。每年有超过320万名消费者通过它的服务购买牛仔裤、羊毛衫和手链等服装和饰品。

与传统网购平台不同的是，Stitch Fix的订阅用户会通过快递收到成箱的服装和饰品，收件的频率多高都可以。在注册的时候，用户需要回答一长串的问题，比如他们喜欢的着装风格和体型等等。然后Stitch Fix的计算机算法和造型师们会根据这些信息，选择给用户寄送哪些商品。用户可以留下他们喜欢的，付完钱后再将剩下的商品寄回去。

现在，Stitch Fix的CEO卡特里娜·莱克正在为公司下一阶段的发展奠定基础。她想利用Stitch Fix强大的数据分析能力，更准确地预测消费者想要购买和保留哪些商品，以创造更多的业务。

莱克对《财富》杂志表示：“我们正在研究如何通过个性化推荐，为你的衣橱贡献更多合适的衣服，让你可以在各种不同的场合穿着。”

StitchFix会对每名用户的数据进行分析，然后生成一份个人档案，然后它会利用可视化手段，制作一份“潜在造型地图”。每张“地图”都包含了几百件官方推荐的衣服，所以它对每个用户的推荐都是极其细致的，而不是笼统地分成几个大类。

在华尔街看来，Stitch Fix创造新的收入来源的速度还不够快。截至10月中旬，该公司的股价已经较2019年的最高点下跌了30%。这一方面是由于吸引和保留用户需要更高的成本，另一方面也是由于有竞争对手复制了它高度个性化的电商模式。

零售业目前正在发生的剧变，也给带来了很多挑战。

最近，Stitch Fix公司推出了一项叫做“Shop Your Looks”（意为“选购你的造型”）的新功能。这也是该公司做的一项重要的试验。它会在已经寄送给用户的推荐商品的基础上，再向用户推荐一些用来搭配的单品。比如用户在收到一件夹克衫后，可能会马上又收到一封电子邮件，建议他们再买一副太阳镜，专门来搭配这件夹克。

该公司希望Shop Your Looks能够扮演一个“穿搭小能手”的角色，继续勾起用户的购买欲，并且提高他们访问Stitch Fix的频率。不过莱克也表示，她也意识到这个功能是有风险的，大家很可能会觉得Stitch Fix只不过是另一个用“推荐商品”向消费者狂轰乱炸的普通网购平台罢了。

在这种模式下，消费者最多只能够在线看到30到40件推荐商品——虽然选择也不少了，但是绝对不会像亚马逊或eBay那样显示出无穷无尽的搜索结果。到目前为止，在使用过该功能购买商品的人中，有60%购买了不止一件。

从业务成绩上看，Stitch Fix可以说是喜忧参半。在截至今年8月3日的12个月间，它的营收入较上年同期飙升29%，达到15.8亿美元，实现利润3690万美元。不过它的利润却较上年同期下降了18%，这与Stitch Fix在打造新服务上投入重资不无关系。

Stitch Fix必须向那些紧张的投资者证明，它有能力继续吸引新用户，同时向现有用户卖出更多的商品。与此同时，它还要面临同业者们对库存服装疯狂地打折销售带来的压力。

KeyBanc Capital Markets公司的分析师艾德·伊鲁玛指出，来自亚马逊的压力，也是Stitch Fix面临的一个“长期隐患”。亚马逊自称是个“能买一切”的网购平台，它的服装业务整体上增长很快，今年6月，亚马逊还推出了自家的个性化购物服务，使Stitch Fix直接成了它瞄准的靶子。

除此之外，Stitch Fix还有一个劲敌——诺德斯特龙（Nordstrom）的Trunk Club，这也是一个偏高端的定制购物服务。与此同时，Instagram和Pinterest等社交媒体服务也对各大电商平台越来越友好了，这些都让本已十分复杂的在线零售业增添了新的变数。

这意味着Stitch Fix必须不断提高其技术的准确度。Stitch Fix拥有一支约3000名真人造型师组成的团队，他们会根据计算机算法的分析结果，决定应该往寄给用户的包裹里放入哪些衣服。

为了优化公司的数据分析能力，Stitch Fix去年还推出了一项名为Style Shuffle的新服务，它每次会向用户展示一款有可能上架的新品，然后让用户进行投票。通过该工具收集的信息，有助于Stitch Fix更准确地对用户进行推荐。到目前为止，该功能已经反馈了大约30亿次用户的评价信息。

与此同时，Stitch Fix也在努力扩大对服装的选择。目前，该平台的服装主要来自一些小品牌。而现在，一些大牌服装也已经逐渐登陆了Stitch Fix，比如New Balance和Madewell等等。这些大品牌之所以如此看中Stitch Fix，在一定程度上也是为了分享数据，好知道用户喜欢什么。同时这些信息也有助于Stitch Fix更准确地预测市场对其自营服装品牌的需求。而自营品牌已经日益成为该公司业务中至关重要的一部分。

在莱克看来，关注数据是她的唯一选择。

她表示：“如果一个人没有收到他们喜欢的东西，他们就会不再使用Stitch Fix。能否为人们提供个性化的服务，这对我们来说是一个事关生死存亡的问题，这是我们的生命线。”

In just eight years, online retailer Stitch Fix has created a flourishing business. More than 3.2 million shoppers use its service annually to buy merchandise from jeans to wool sweaters to bracelets.

Unlike with conventional online retailers, customers subscribe to Stitch Fix to receive boxes of apparel and accessories, or “fixes,” as often as they want. When signing up, clients answer a long list of questions about the kind of clothes they like and their body type—information that the company’s algorithms and human stylists use to choose which items to send. Customers keep and pay for what they like, and send the rest back.

Now Stitch Fix CEO Katrina Lake is laying the groundwork for her company’s next chapter. She wants to tap Stitch Fix’s data-crunching prowess to even more accurately predict what shoppers want to buy and keep, and to drum up more business between so-called fixes.

“We are trying to figure out how we can use personalization to deliver more parts of your closet so that you can use those items for all occasions,” Lake tells Fortune.

StitchFix analyzes data to generate an individualized profile for each customer, which it visualizes in a “latent style map”. Each map is comprised of hundreds of suggested pieces of clothing, constructing an extremely nuanced picture of each user—versus pigeonholing into overly general categories.

For Wall Street, Stitch Fix’s push for new revenue sources can’t come soon enough. As of mid-October, its shares were down 30% from their 2019 high, owing both to the rising cost of attracting and retaining customers and to rivals’ copying its personalized approach to e-commerce.

The ongoing upheaval in the retail industry makes Stitch Fix’s latest push that much more challenging.

One crucial test for the company is Shop Your Looks, a feature that suggests additional items to customers to complement what Stitch Fix sends them in their fixes. For example, clients who keep a jacket sent to them may later receive a suggestion via email that they buy a pair of sunglasses to go with it.

The hope is that Shop Your Looks will prompt an impulse buy between boxes and get customers to visit Stitch Fix more often. Lake recognizes the risk of making Stitch Fix just another online retailer that bombards shoppers with an exhausting list of “suggested items.”

That means the e-tailer shows at most 30 to 40 suggested items online—a lot of choice, but not the endless scroll shoppers see in Amazon’s or eBay’s search results. So far, 60% of people who have bought an item using this feature have bought more than one.

In terms of its business, Stitch Fix is getting mixed results. In the 12 months ended Aug. 3, its revenue soared 29% to $1.58 billion compared with the preceding year. During that period, the company had a profit of $36.9 million. But that profit was down 18% from the previous year, as Stitch Fix spent heavily to build out new services.

Stitch Fix must show nervous investors that it can continue to attract new customers and sell more to existing ones, all while grappling with the apparel industry’s rampant discounting of overstocked clothing.

As KeyBanc Capital Markets analyst Ed Yruma points out, Stitch Fix also faces “long-term concerns” related to Amazon. The self-proclaimed Everything Store’s overall apparel business is growing rapidly, and in July it debuted its own personal shopping service—putting Stitch Fix directly in its crosshairs.

As if that’s not enough, Stitch Fix has a serious rival in Nordstrom’s Trunk Club, a slightly higher-end bespoke shopping service. Meanwhile, Instagram and Pinterest have both made their services friendlier to online retailers, adding a new wrinkle to what is already a complex retail environment.

That means Stitch Fix must keep improving the accuracy of its technology. An army of some 3,000 human stylists uses what the algorithm spits out to help decide what to include in customer fixes.

Style Shuffle, a feature added last year that shows customers prospective products one at a time and lets them vote on each, is part of the company’s effort to improve its data crunching. The information collected through the tool—some 3 billion ratings have been submitted—helps make customer suggestions more accurate.

Meanwhile, Stitch Fix is also working on expanding its clothing selection, which is heavy on smaller brands. Big-name clothing makers have gradually come on board, including New Balance and Madewell. Part of the pitch is that Stitch Fix can share data with them about what customers like. That kind of information also helps Stitch Fix more accurately predict demand for its own clothing brands, an increasingly crucial part of its business.

For her part, Lake doesn’t see any choice but to focus on data.

“If somebody is not receiving things that they love, they’re going to stop [using Stitch Fix],” she says. “We live and die by our ability to personalize for people. That is our lifeblood.” 

我的私人购物顾问

衣柜“闹饥荒”，但是不知道该买什么？Stitch Fix的CEO告诉我们，可以让计算机算法和真人造型师来帮你决定。

“你想要一条破洞牛仔裤吗？”Stitch Fix公司的CEO卡特里娜·莱克问道。她的鼠标光标此刻正停在一张褪色的蓝色牛仔裤的图片上。

我从来没买过破洞的牛仔裤，所以我不知道该怎么回答她。好在莱克的MacBook电脑上有一个软件，它已经代表我做出了一个有根据的猜测——我有74%的可能性会喜欢这条裤子。于是我告诉她：“好的”。这位CEO用鼠标点了一下图片，把这条裤子添加到了我的购物箱里（也就是Stitch Fix定期寄送给用户的个性化包裹）。然后我们又接着看起了外套。

“哇，这一件很适合旧金山的天气。”她指着一件黑色的夹克说。很显然，我应该买一件挂在衣柜里——根据Stitch Fix的软件，我有62%的几率会买它。

Stitch Fix为用户挑选衣服不仅靠算法，也靠艺术。该公司的造型师们在为用户挑衣服时也是有发言权的。今天，莱克就让我看到了这个过程的幕后环节，并且用我的Stitch Fix个人档案，为我现场搭配了一个“箱子”。

My Own Personal Shopper

Stitch Fix’s CEO shows what it’s like to let algorithms and human stylists choose your wardrobe. By Michal Lev-Ram

“Do you want a ripped denim?” asks Katrina Lake, CEO of online styling service Stitch Fix, her computer cursor hovering over an image of faded blue jeans.

I’ve never actually owned a pair of pants with premade holes, so I’m unsure how to answer. Lucky for me, the software Lake is running on her MacBook has already spit out an educated guess on my behalf: There’s a 74% chance that I’ll like this particular garment. I tell her yes, and the CEO clicks on the image, adding it to my “fix” (the personalized box of five items Stitch Fix sends to its clients). We move on to outerwear.

“Ooh, this one is good for San Francisco weather,” she says, pointing to a black jacket. Apparently, it belongs in my closet—¬I have a 62% chance of keeping it, according to Stitch Fix’s software.

It’s not just algorithms that pick clothes for customers at Stitch Fix; it’s also art. The company’s human stylists have a say when creating fixes for customers. Today, Lake is giving me a behind-the-scenes look at the process—and using my real-life Stitch Fix profile to put together a real-life fix for me.

它的工作原理是这样的：在每次推荐之前，Stitch fix都会将一名用户与一名造型师进行配对，在这个过程中，它会考虑到地理位置和时尚偏好等变量（我们可以跳过这部分了，因为在这次演示中，莱克亲自担任了我的造型师）。然后，选中的造型师会进入用户的个人账户，对系统算法认为符合客户品味的预选衣物进行评估。

在这个过程中，系统会对大量数据进行分析，包括用户的个人档案（比如我已经告诉Stitch Fix，不要给我发送带有动物图案的衣服）、购买历史（我可能口头上说自己喜欢大胆一点的颜色，但实际上买得最多的还是黑色的）等等。设计师对最终的选择仍然有发言权，并且可以推翻系统的建议。

莱克表示：“这些有助于设计师在深思熟虑后做出正确的选择。”她还表示，如果顾客明确要求，造型师也可以给用户发送一件低评分的商品。

我也亲自看到了这种情况的发生。我让莱克给我找几双靴子。她点击进入了这个类别，但系统显示，即便是评分最高的靴子，被我喜欢的几率也只有4%。莱克说道：“我们已经给你寄了11双鞋了，但你只留下了两双。”（于是我们决定跳过靴子的部分。）

几天后，一个“箱子”被快递员送到了我家门口，里面还有这位CEO的一封信。“这只是为了好玩——这里都是我们根据预测，认为你会喜欢的东西。”事实证明，莱克的眼光和她的公司的算法确实厉害——我留下的三件衣服，恰好是系统认为我最有可能留下的那三件。另外，没错，我现在超喜欢破洞牛仔裤的。（财富中文网）

本文另一版本登载于《财富》杂志2019年11月刊，标题为《Stitch Fix利用算法向你推荐穿搭》。

译者：朴成奎

Here’s how it works: Before a fix is started, Stitch Fix’s technology pairs a customer with a stylist, taking into account variables like location and fashion preferences (we’ve skipped that step because Lake has been designated as my stylist for this demo). Then the selected stylist accesses the client’s account to review a preselected assortment of clothes that the system’s algorithm has deemed to be in line with that shopper’s taste.

A lot of data feeds into this computerized curation, including a customer’s profile (I’ve told Stitch Fix not to send me “critter” prints, for example) and purchase history (I may say that I want bold colors but tend to keep black tops). The stylist still has say over the final selection and can override the system’s suggestions.

“It helps the stylist thoughtfully make the right choices,” Lake says of the technology, adding that stylists can send shoppers an item with a low score if the shopper specifically asks for it.

I see this play out in real time when I ask Lake to find me some boots. When she clicks into the category, though, the highest-ranked boots are listed as having only a 4% likelihood of ending up in my closet. “We’ve sent you 11 pairs of shoes, and you’ve only kept two,” Lake says. (We decide to skip the boots.)

A few days later, my fix arrived on my doorstep, along with a note from the CEO. “Just for fun, here are our predictions on what you’ll like!” wrote Lake, noting for each item the statistical probability that I will. As it turned out, the combination of Lake’s eye and her company’s algorithms was a winner: The three garments I kept all happened to have the highest likelihood of my keeping them—and, yes, I’m now the proud owner of ripped denim.

A version of this article appears in the November 2019 issue of Fortune with the headline “Stitch Fix Thinks Outside the Box.”