BBC News America Amazon: How Bezos built his data machine

By Leo KelionYou might call me an Amazon super-user.I’ve been a customer since 1999, and rely on it for everything from grass seed to birthday gifts.There are Echo speakers dotted throughout my home, Ring cameras inside and out, a Fire TV set-top box in the living room and an ageing Kindle e-reader by my bedside.I…

BBC News America

By Leo KelionYou might call me an Amazon super-user.I’ve been a customer since 1999, and rely on it for everything from grass seed to birthday gifts.There are Echo speakers dotted throughout my home, Ring cameras inside and out, a Fire TV set-top box in the living room and an ageing Kindle e-reader by my bedside.I submitted a subject data access request, asking Amazon to disclose everything it knows about meScanning through the hundreds of files I received in response, the level of detail is, in some cases, mind-bending.One database contains transcriptions of all 31,082 interactions my family has had with the virtual assistant Alexa. Audio clips of the recordings are also provided.The 48 requests to play Let It Go, flag my daughter’s infatuation with Disney’s Frozen.Other late-night music requests to the bedroom Echo, might provide a clue to a more adult activity.Clicking on another file reveals 2,670 product searches I had carried out within its store since 2017. There are more than 60 supplementary columns for each one, containing information such as what device I’d been using, how many items I subsequently clicked on, and a string of numbers that hint at my location.One spreadsheet actually triggers a warning message saying it is too big for my software to handle. It contains details of the 83,657 Kindle interactions I’ve had since 2018, including the exact time of day for each tap. An associated document divides up my reading sessions for each e-book, timing each to the millisecond.And on it goes.The endeavour was timed to coincide with a BBC Panorama documentary I’ve been involved with, which tracks Amazon’s rise through the prism of it being a data-collector.“They happen to sell products, but they are a data company,” says James Thomson, one of the former executives interviewed.“Each opportunity to interact with a customer is another opportunity to collect data.”Founder Jeff Bezos frames it in terms of being a “customer obsession”, saying the firm’s first priority is to “figure out what they want, what’s important to them”.And he qualifies this by saying Amazon mustn’t violate people’s trust in the process.Yet as the company continues to grow, and expand into new activities, there are calls from both inside and outside Amazon to keep its data-feasting obsession in check.Sleeping Lady resort is about a two-hour drive from Seattle.The name comes from the shape of the mountains that tower above its wooden cabins.When Bezos bussed Amazon’s staff there for a brainstorm in January 1997, it lived up to its Icicle Road address. A storm meant some missed the first evening’s events.But they’d all arrived when their leader began the next morning’s session, saying he wanted to create “a culture of metrics”.James Marcus was an in-house book reviewer at the time.“This mania for quantification was in Jeff’s heart,” he says.Marcus was grouped with Bezos at the event, as teams scribbled equations on white boards trying to invent ways to measure customer enjoyment.“Jeff’s algorithm wasn’t that much better than anybody’s algorithm that day,” he says.“But he understood that data was in fact very valuable.“The idea that every mouse click and every twist and turn through the website was itself a commodity, was a new sort of thinking for most of the employees – and for me too.”The next challenge was to decide what to sell beyond books.They picked CDs and DVDs. Over the years, electronics, toys and clothing followed, as did overseas expansion.And all this time, Amazon was building a battalion of data-mining experts.Artificial intelligence expert Andreas Weigend was one of the first.Before joining, he had published more than 100 scientific articles, co-founded one of the first music recommendation systems, and worked on an application to analyse online trades in real-time.Amazon made him its first chief scientist.“I had weekly meetings… with people, whoever wanted to stop by, where we just looked at clickstream histories in the evening, with beer and pizza, to wrap our brains around why would people actually do this, why on Earth would they click here,” he remembers.Clickstreams are the digital breadcrumb trail which Amazon follows to see which sites users come from, how they travel through its own pages and where they go to next.(Amazon’s response to my data request didn’t contain my own clickstream history, although the firm has provided such records to others in the past. A spokesman could not explain why.)A savvy ad-tech specialist was among Weigend’s recruits.David Selinger quickly climbed the ranks to lead the new Customer Behavior Research unit.Bezos, Weigend (centre) and Selinger (right). Weigend: “I don’t know how that picture happened, I know that picture is in my bedroom in my bed at my home and nobody really is clear how Jeff Bezos got in my bed”Bezos, Weigend (centre) and Selinger (right). Weigend: “I don’t know how that picture happened, I know that picture is in my bedroom in my bed at my home and nobody really is clear how Jeff Bezos got in my bed”“Our job was to build a customer-based dataset and then prove that there were opportunities, kind of fissures of gold,” Selinger says.Once a week they too delved into individuals’ behaviour.“We had to make it actionable,” Selinger continues.“So to do that, we would project on the wall this view [of a] single customer and try to understand who she was.“What was unique about the internet and Amazon at the time was that we were able to take each individual customer and then change the experience.”Their work gave rise to personalisation and targeted recommendations, such as a customised front page for each user, and tailored emails in their inboxes.“I was shocked to see how predictable people are,” says Dr Weigend.“We didn’t think of it as exploiting, we thought about helping people make better decisions.”Weigend and Selinger moved on, but Amazon continued to hire talent to find innovative ways to turn data into dollars.Among them was ex-banker James Thomson.“I had worked for other companies where there was a so-called data warehouse,” says the former Amazon Services business chief.“But Amazon’s is literally the largest.“Amazon knows not just your preferences but the million other preferences of customers that look a lot like you.“So Amazon can basically anticipate what you’re going to need next – size up the inventory of which brands they are going to need in three to six months when you are ready to ‘unexpectedly’ buy those products.”It used to be exotic to talk of “big data”. These days the buzz phrase is “artificial intelligence”.However you frame it, Amazon leads the way in finding patterns in the noise of customer behaviour.But while this fuels its profits, it has also prompted concerns about the elevated positions Bezos and his deputies enjoy as a result.“We find ourselves being shot backward into a kind of feudal pattern where it was an elite, a priesthood, that had all the knowledge and all the rest of the people just kind of groped around in the dark,” says Shoshana Zuboff, a Harvard professor and author of The Age of Surveillance Capitalism.“This narrow priesthood of data scientists and their bosses sits at the pinnacle of a new society.“They are not beholden to us as customers, because in the surveillance capital model we are not customers, we are sources of raw material.”Your brand, our customersNot all of Amazon’s big decisions have been based on data.Sometimes Jeff Bezos simply goes with his gut.At the turn of the millennium, he wanted a logo revamp.The old logo showed the website’s address above a simple downturned swoosh.A package design agency pitched the idea of turning the curve upside down to form a smiling arrow pointing from A to Z.Bezos instantly loved it and rejected the need for consumer-testing.“Anyone who doesn’t like this logo doesn’t like puppies,” he said.The logo not only made Amazon seem friendly, but also highlighted its “everything store” ambitions.Trying to stock everything on its own was out of the question. So, instead it went into business with its competition.One after another, Amazon convinced bigger companies to outsource their e-retail operations to it.ToysRUs, Borders, Waterstones, Marks & Spencer and Target were among household names to sign up.The deals boosted everyone’s earnings in the short term, but Amazon was also making a longer-term play – it recognised the value of its partners’ data.John Rossman headed up the initiative for a time. “At that point people didn’t understand the potential for e-commerce and digital business, and they essentially just viewed it as, ‘Hey here’s additional revenue,’” he recalls.“They really gave away the keys to a kingdom.”Under a strategy named Launch and Learn, Amazon first partnered with rivals, then studied their sector’s value chain, and finally expanded onto their patch.It was years before the older firms realised the value of what they had given away.Some collapsed. Others extricated themselves in time but regretted their error.“They learned a tonne on our dime, and we didn’t learn much,” Target’s ex-chief strategy officer Carl Casey later complained.The other side to Amazon’s strategy was to convince smaller third-parties to sell new and used items via Marketplace – a platform which allowed their goods to be listed on the same pages as its own stock.It had a slow start, but became a massive success.“Third-party sellers are kicking our first-party butt,” the firm gleefully disclosed in its last annual report, referring to the fact that since 2015 independent sellers have accounted for the majority of physical goods sold via its site.Part of Marketplace’s success is down to Amazon’s willingness to share increasing amounts of analytics with the sellers. But only Amazon gets complete access.“Whether you’re Target or another big retailer, or whether you’re a small entrepreneur who set up a third-party seller account, in all situations you’re basically renting the Amazon customer,” explains Thomson. “In the end, Amazon collects all that data – and it stays within the Amazon database.”For many merchants this is an acceptable trade-off.Lavinia Davolio says Amazon made it possible to sell her luxury Lavolio sweets across the world, as well as afford a London store.“The main benefit of the Amazon Marketplace is that my product suddenly is more visible to millions and millions of customers,” she says.But when pushed, she acknowledges: “We can’t really communicate directly to them – the customer stays with Amazon.”Most customers are probably happy that sellers can’t bombard them with follow-up messages.However, when things go wrong, merchants can feel like they’ve little recourse.Roland Brana’s mo
Read More