Running Amazon, the gigantic business empire with unprecedented complexity from drastically different businesses, from vastly spanned geographies worldwide and from the massive size and scope of operations, is no doubt a daunting challenge for many.
If you were charged with such important obligation, you would probably be deeply buried to the neck in administering the day-to-day.
To the huge surprise of all, Jeff Bezos, founder and CEO of Amazon, actually rarely spends time on day-to-day considerations. “I try to organize my personal time so that I live mostly about 2 to 3 years out.”1 He also requests that his top leaders do the same. Ram saw similar mindset in former chairman and CEO of GE, Jack Welch.
Is this because Bezos is super hands-off? Absolutely not. On the normal spectrum of hands on vs. hands off, Bezos would perhaps be the one to redefine diving deep into details by magnitude.
If this is true, how to reconcile this perplexing paradox?
Without digitization, one could not do that. Bezos’ secret lies in the Amazon world-class standard-setting AI-powered data and metrics system, in which everything that matters can be tracked, measured and analyzed, with insights generated and routine decisions automated.
Such a system not only liberates Bezos, executives, and frontline employees at Amazon from managing routine daily chores and the inevitable bureaucracy associated, but also enables powerful AI-powered tools-that are fundamental to the Amazon management system.
Jeff Bezos is a man of numbers. This is his unique way of understanding the world, having fun in life and running business at Amazon.
When he was a little boy, Bezos would go on long-distance driving adventures with his grandparents. During those long hours on the road, he would kill time by doing minor arithmetic problems and making estimates. One specific case Bezos mentioned during his speech on 2010 Princeton Commencement was that after hearing an advertisement campaign about smoking which stated that each puff of a cigarette would cut short people’s life span by a few minutes, he proudly told his grandma, “At two minutes per puff, you’ve taken nine years off your life!”
Sounds incredible? It may seem incredible to most of us, but this was simple and straight-forward for Bezos. Building on the estimates of daily consumption of cigarettes, and the number of puffs per cigarette, he quickly did the math in his head. (On that day, Bezos learned a valuable life lesson from his grandpa that it’s harder to be kind than clever.)
In terms of founding and managing Amazon, a fast-expanding giant woven into almost all aspects of the economy and people’s lives, Bezos followed the same line of thinking. At Amazon, when Bezos throws a question at you, there is no wriggle room for obfuscation or buzzwords. If you dare to try, he will rant at you, ruthlessly, “the answer starts with a number!” Everyone at Amazon knows the famous line by W. Edwards Deming, “In God we trust, all others must bring data.”2
For many people working at Amazon, the first thing they do every day is look at the numbers. Armed with smartphones, many of them start this daily ritual even before getting out of bed. They have become masters of cutting through a bevy of numbers to know what is really happening.
Just as Bezos has his unique standards to define the right talent for Amazon, as described in the previous chapter, he also has his unique standards to define what the robust data and metrics are.
Execution is about knowing and delivering details. Amazon’s level of details for data and metrics is magnitudes beyond many people’s wildest imagination. “Shock” is usually the first reaction of many outside Amazon.
If you were asked to pick a location for a new data center, how many factors would you consider? About five, ten, twenty, or dozens? Amazon used a checklist of 282 metrics when choosing its first data center in China, according to Mr. Wan Xinheng, mayor of Zhongwei, a small city located in the west of China, where Amazon built its first data center in the country back in 2015. In an interview in 2016, Mr. Wan said that he was clearly shocked.
If you were to set annual goals for your company, how many items would you list? About five, ten, twenty, or dozens? Amazon nailed down 452 detailed goals for 2010, as stated in Bezos’ 2009 Shareholder Letter. But goals, by themselves, are not enough. Amazon also specified owners, deliverables, and targeted completion dates for each.
If you were in charge of the third-party book category at Amazon, how many metrics would you look at each day? About 5, 10, 20, or dozens? Amazon compiled 25 pages of various metrics, such as:3
• Order defect rate (ODR): the percentage of orders with negative feedback from customers, be it an explicit complaint, a low rating, or a dispute.
• Pre-fulfillment cancellation rate: the percentage of orders cancelled before shipment.
• Late shipment rate: the percentage of orders that arrived later than the committed date.
• Refund rate: the percentage of orders that resulted in refunds for any possible reason.
• Contacts per order: the average number of all human interactions for each order.
• The best-selling books, key words, writers, publishing houses, and third parties.
• The most-searched books, key words, writers, publishing houses, and third parties.
• The time required to load a webpage.
Just imagine such detailed metrics, two to three magnitudes beyond the normal definition of detailed, going on for 25 pages.
Would you find 25 pages too much? As a matter of fact, this is already a scaled-back version from the original 70-plus-page list. Of course, if you want to deep dive into some metrics, you can always log on to Amazon’s internal system to fully indulge yourself in an ocean of data and metrics.
In most traditional companies, data collection is broken down by silo, by layer, and by actual involvement. Each division or function can see data generated and collected only within its own domain of business operations. For example, sales may see sales numbers, marketing may see marketing expenses, production may see production orders and finance may see inventory turns, bottom lines and cash generation. However, it would be extremely hard to link all these data points and figure out, at each SKU level, which ones were the best in generating cash flow and net profits.
In these types of organizations, it is almost impossible to get data from other silos. Information sharing may be thwarted by a number of reasons such as concerns about confidentiality, lack of authorization or reluctance stemming from personal grudges, as well as other obstacles disguised by delay, distortion, or purposeful omission of critical pieces of information. The list of legitimate reasons will be long, and the list of deliberate and delicate excuses will be even longer. Why? Because in many cases, information has become the basis of power.
That’s why when traditional companies embark on the journey of digitization, data transparency is usually one of the first steps. As one chairman repeatedly reminded everyone during an extended executive workshop on digitization, there should be “no more hoarding of data.”
As we mentioned in the chapter on “Building Block 1: Customer-Obsessed Business Model,” data is the new equity in the digital age. In this sense, all data belongs to the entire company, not any individual or any division.
At Amazon, a small team is bestowed with the end-to-end responsibility for one product or one service. How to ensure that this team does a good job? After ensuring the selection of the best people, the next most important crucial enabler is the availability of end-to-end data that is not segregated by silo nor by function. Without such data support, running a business would be as difficult as maneuvering in a pitch-dark mansion.
In this sense, the transparency of end-to-end data is an effective forcing mechanism in dismantling silos and enabling end-to-end accountability.
In many companies, business reviews are held quarterly or monthly, with a delay of ten days or more due to the time needed for accounting proceedings. As a result, it is common for the Q1 reviews to be held around April 10 and the May review around June 10.
In one real-case example from our experience, the quarterly review of a company’s key account business was held on April 15. During the meeting, the VP of key account business reviewed with his team the Q1 performance vs. budget for each of the top 20 accounts. For those with big gaps, he probed the data from the quarterly down to monthly performances and found out that, for one particular account, January sales met the budget, but sales suddenly dropped in February and March. He questioned the person in charge about the potential causes, brainstormed with the team about how to fix them, and made decisions on four action items right on the spot.
What do you think about this VP? Ready to sing his praise for having a nose for details, a bias for action, and the guts to make decisions on the spot? He is a great leader, right? He may be a great leader by the traditional standards, but such a way of running a business is woefully inadequate for winning in the digital age. Any actions starting on April 15 are already too late by two-and-a-half months.
At Amazon, such data is tracked on a real-time basis with no time lag. Relevant people can review the results daily, hourly, or by the second. Armed with real-time data and metrics system, the person in charge of that particular account mentioned above could have probably detected the anomaly by herself as early as within the first few days of February, or even the last few days of January, and could have adjusted by oneself or with one-level approval at most. No need to waste two-and-a-half months. In some cases, your company’s fate can be sealed within two-and-a-half months or even just two-and-a-half days.
This is probably the most unique aspect of Amazon’s data and metrics system.
When setting goals, most companies focus on revenue growth, margins, and net profits. However, among Amazon’s 452 goals for 2010, “The word revenue is used eight times and free cash flow is used only four times . . . .the terms net income, gross profit or margin, and operating profit are not used once.4”
Why? Revenue, growth, margins, and net profits are outputs. Amazon believes that to ensure good outputs, one needs to get to the bottom of the issue and seriously track the inputs.
Why does Amazon track the time required to load a webpage? Because its data analytics shows that “even a minuscule 0.1-second delay in a webpage loading can translate into a 1 percent drop in customer activity.”5
Why does Amazon track the metric of contacts per order? Because each contact, i.e., human interaction with the customers, can reveal a potential system defect and clearly has costs, big or small. In fact, by tracking and then aggressively reducing contacts per order by 90%, Amazon significantly improved its profitability in 2002, i.e., turning positive in operating profits for the first time in the company’s history.6
At Amazon, each claim needs to be supported by data and metrics. Unfounded promises will not fly. For those who get caught, their days at Amazon are numbered.
Bezos clearly embodies Amazon’s Leadership Principle of Dive Deep, which declares that: Leaders operate at all levels, stay connected to the details, audit frequently, and are skeptical when metrics and anecdotes differ. No task is beneath them, and they would invest time and energy to verify personally.
For example, at an executive meeting during the Christmas season in 2000, Bezos asked the head of the Customer Service Department about customers’ wait time. This is a metric of how long customers have to wait before their calls get picked up by a customer service representative. Without offering any supporting evidence, the person replied that it was well under one minute.
How could such a colossal mistake escape Bezos’ eagle eye? Using the speakerphone in the middle of the conference room, Bezos dialed the 800 number of Amazon’s call center. He even took off his watch to track the time.
Guess how long Bezos waited for his call to get through? Not one minute, not even two minutes, but four-and-a-half minutes. That is 270 seconds. You may take a moment to count from 1 to 270, to get a sense of how long this feels. No doubt, for that particular executive, the collective wait with Bezos and the entire executive team must have felt like an eternity.
Why would Bezos invest the precious four-and-a-half precious minutes of the entire executive team on this seemingly “trivial” detail? Two reasons.
First, to Bezos, who is truly obsessed with customers, this was just the opposite of trivial. In fact, it was paramount to the customers’ experience. No customers would call Amazon’s call center just for a friendly chat. Usually it was an unpleasant encounter or a frustrating problem that triggered their call. The long wait would simply exacerbate their growing dissatisfaction and mounting anger.
Second, Bezos used this specific example to vividly demonstrate the Dive Deep principle right on the spot, i.e., no task is beneath them, and they should invest time and energy to verify personally. After these painfully long four-and-a-half minutes, everyone on site and everyone who heard about the anecdote would definitely learn the lesson by heart. This is effective coaching in the moment.
How would Bezos himself make a point and support it with bullet-proof evidence?
To illustrate Amazon’s pricing objective of “not discounting a small number of products for a limited period of time,” but offering “low prices every day and apply them broadly across our entire product range,” in his 2002 Shareholder Letter Bezos quoted the results of a price comparison of 100 best-selling books.
To eliminate bias, when picking the 100 best-selling books, he used the list of Amazon’s major competitor, you know whom. To ensure representativeness, he examined the composition of these 100 books by category and by format, and had people visit four of their superstores in both Seattle and New York City for price points. Based on the collected information, he compared prices by collective cost, by each title, and by number of books being sold at a discount.
The price-comparison exercise generated the following discoveries:
• “At their stores, these 100 bestselling books cost $1,561. At Amazon.com, the same books cost $1,195 for a total savings of $366, or 23%.
• For 72 of the 100 books, our price was cheaper. On 25 of the books, our price was the same. On 3 of the 100, their prices were better (we subsequently reduced our prices on these three books).
• In these physical-world superstores, only 15 of their 100 titles were discounted — they were selling the other 85 at full list price. At Amazon.com, 76 of the 100 were discounted and 24 were sold at list price.”7
At Amazon, the ultimate test for the robustness of a data and metrics system is to simply step into the fire. If you can survive the barrage of questions from Bezos and his executives, usually two to three degrees beyond the normal standards of a deep dive, and provide convincing answers supported by solid numbers, it means you have passed.
To define and continuously refine such ultra-detailed, end-to-end (cross-silo and cross-layer), real-time, and input-heavy mazes of metrics, and to continuously track, measure and analyze the massive volume of data by all these metrics, is no piece of cake. It takes heavy investment of money and, more importantly, people’s time and energy over many years and at all levels of the organization.
Why has Amazon been so committed to this course? Bezos’ strong personal fascination with numbers clearly plays a role here. However, what matters more is the hefty return of such upfront heavy investments.
Armed with the AI-powered data and metrics system, Amazon can liberate all builders at all levels of the organization and at the same time ensure the continuous bar-raising of a forever-Day-1 organization.
In most traditional companies, once the business grows bigger, the number of employees expands rapidly as well.
As prescribed by the traditional management theory span of control, the number of subordinates that a manager can effectively supervise is limited. The optimal number varies by the nature of the work at hand, but the range usually goes from 2-3 to 6-8, and rarely expands beyond 10-15. Therefore, understandably, many big companies tend to have six to seven layers of managers. We know a few giants with 10 or more layers.
Equipped with its AI-powered data and metrics system that can continuously track, measure, and analyze business operations, detect anomalies and automate routine decisions using ultra-detailed, end-to-end, real-time and input-heavy metrics, Amazon has actually defied the span of control theory, the cardinal rule of business organization design. Such a data and metrics system significantly minimize the need for physical supervision.
In fact, Amazon defies the rule in such a fundamental way that Jeff Wilke, CEO of Worldwide Consumer, could personally manage 500 project teams. How is that possible? The credit first goes to the data and metrics system, and then to the internal project management system, a powerful tool built on the data and metrics.
Amazon also has business review meetings, but with two key differences from most traditional companies. One is cadence: Amazon’s review is on a weekly or bi-weekly basis. With the accelerated feedback and adjustment loop, Amazon can identify issues, and make mid-course adjustments much faster and with much more agility than its competitors.
The other is focus. Instead of focusing on historical performance and having each executive or manager do lengthy presentations, Amazon’s review focuses more on how to solve particular customer problems and how to design and implement experimentations to improve, innovate, and invent.
In this sense, Amazon’s data and metrics system liberates the executives from having to bury themselves in the routine day-to-day operations, and frees up more time and energy for them to devote to continuous improvement, innovation and invention, and to live in the future.
This is one of the pivot foundations for Bezos’ vision of building Amazon into an invention machine.
In most traditional companies, once a decision is made by the boss, it is very hard to have it overturned. The inevitable waves of distortion and delays invariably compound the folly of the original decision, which, by the time the consequences are felt by frontline workers, feels completely off the rails.
When the frontline people receive such suboptimal or even “insane” instruction from a certain boss who is already detached and disconnected from current market dynamics, and current preferences of the target customers, what to do? In most cases, their only option is to suck it up and live with it.
Frontline people rarely have an opportunity to voice their views, let alone occupy a seat at the table when key decisions are being made. Even if they are lucky enough to be granted access, their views are invariably crushed by the boss, who behaves as if having more power and more experience somehow confers upon him more wisdom or customer experience.
At Amazon, the frontline people can be liberated from such painful frustrations. When difficult decisions are to be considered, frontline workers are encouraged (in fact required) to take the initiative of pulling all relevant data from the system and running the required analysis on their own. If the results support their views, there is no need to wait or to worry. They are expected to go to the boss immediately and get the flawed decisions in question reversed.
Also, given the transparency of data and metrics, frontline people don’t have to wait for the boss’ probing, questioning, and subsequent instructions weeks or months afterwards. When real-time data send the warning signals, the respective metric owner will take the immediate initiative on their own to identify root causes and develop corrective actions for mid-course adjustments.
Moreover, transparency of such ultra-detailed, end-to-end (cross-silo and cross-layer), real-time and inputs-oriented data and metrics makes the usual uphill battle for cross-functional collaboration much easier. Real-time data is the best persuasion point to get the right help from almost anyone in the company.
This is one of Amazon’s secret ingredients for speed and agility.
Almost all companies espouse a goal of a performance-driven culture. Without the strong support of a superbly robust data and metrics system, executives are often missing critical data they can use to make effective and informed decisions. In the absence of the most informative data, such aspirations do not fulfill their potential.
When Jeff Wilke joined Amazon in 1999 with the mission to fix the company’s operations, one of his first changes he made was to devise “dozens of metrics” and order “his general managers to track them carefully, including how many shipments each fulfillment center received, how many orders were shipped out, and the per-unit cost of packing and shipping each item.”8
Most people may think these are mundane tasks, but they’re essential for customer convenience, operational excellence, and continuous bar-raising.
In fact this rigorous work proved itself to be instrumental to Amazon’s future success in Prime (two-day free shipping for prime members) and FBA (Fulfillment by Amazon). Even today, twenty years later, one Chinese e-commerce mega-giant still struggles with how to get an accurate per-unit cost of packing and shipping for each item, and how to help general managers of fulfillment and dispatch centers improve performance.
That’s why Wilke could promise Bezos “that he would reliably generate cost savings each year just by reducing defects and increasing productivity.”9 Of course, he also delivered. As you know, this is very important at Amazon.
Bezos always aspires to build a forever-Day-1 organization at Amazon. Continuous bar-raising is at the heart of this vision. The data and metrics system are in fact a foundational enabler to nail it down for everyone in every activity at Amazon in a crystal-clear, super-specific and highly measurable way.
Before founding Amazon, Bezos worked for four years at D. E. Shaw and Co, a boutique investment firm on Wall Street, which actually let computers make all trading decisions.
During his weekly brainstorming session with founder David Shaw, Bezos was able to test out some earliest thoughts about the promise of the coming digital economy: he had already envisioned that some of Amazon’s greatest inventions that have become the common practices we take for granted today, such as the ultimate personalization that treats each customer differently.
In the beginning of his 2010 Shareholder Letter, Bezos wrote:
“Random forests, naïve Bayesian estimators, RESTful services, gossip protocols, eventual consistency, data sharding, anti-entropy, Byzantine quorum, erasure coding, vector clocks . . . . walk into certain Amazon meetings, and you may momentarily think you’ve stumbled into a computer science lecture.
“Look inside a current textbook on software architecture, and you’ll find few patterns that we don’t apply at Amazon. We use high-performance transactions systems, complex rendering and object caching, workflow and queuing systems, business intelligence and data analytics, machine learning and pattern recognition, neural networks and probabilistic decision making, and a wide variety of other techniques. And while many of our systems are based on the latest in computer science research, this often hasn’t been sufficient: our architects and engineers have had to advance research in directions that no academic had yet taken. Many of the problems we face have no textbook solutions, and so we — happily — invent new approaches.”
Clearly, Bezos’ personal passion for technology and signature spirit of imagining and inventing has become one of Amazon’s underlying characteristics. In this aspect, Amazon is the front-runner.
This is a key differentiator. How many CEOs and senior executives even have a feel of what these digital tools are and what magic these tools can do? If they don’t have the feel personally, do they have someone who they can trust and who know how to apply these digital tools into business?
The applications of this approach are countless, and inform virtually all key decisions made by Amazon.
How does Amazon pick the location of its next fulfillment center? The answer is Mechanical Sensei, a software system “that simulated all the orders coursing through Amazon’s fulfillment centers and predicted where new FCs would most productively be located.”10
How does Amazon help the hundreds of thousands of third-party sellers who contributed $160Bn gross merchandise sales in 2018? By providing “the very best selling tools we could imagine and build.”11 Such tools help sellers incorporate all factors related to business operations, such as seasonality, historical results, future predictions, competitive offerings, and cash flow considerations, in order to make the best decisions in order, inventory, pricing and promotion, as well as provide the most convenient services in order processing, payment collection, shipment tracking and performance analysis.
How does Amazon manage the large number of third-party sellers, mostly small- and medium-businesses? From the very beginning, Amazon’s third-party platform was developed based on the design principle of self-governance. The data and metrics system can meticulously track the performance of each third-party seller by using a whole set of metrics, and then roll the actual performance results along these metrics into an aggregated index score. For top performers, the system will automatically create various rewards according to the pre-defined rules specified in the algorithms; for the ones with issues, alerts will be sent and in severe cases, the management team will be involved in discussions before removing them from the Amazon platform.
How does Amazon ensure its competitiveness in pricing? Pricing bots. These are “automated programs that crawled the Web, spied on competitors’ prices, and then adjusted Amazon’s prices accordingly, ensuring that Bezos’s adamant demand that the company always match the lowest price anywhere, offline or online, would be met.”12
How does Amazon drive more consumption from each consumer? Personalized recommendations. And who at Amazon decides which items to recommend to which customers? Actually . . . .no one. An algorithm-enabled system fully automates personalized recommendations for each individual consumer.
How does Amazon develop options and decide the fastest and cheapest delivery option for each order? By the early 2000s, an Amazon fulfillment software system could run millions of such decisions every hour. Given Amazon’s relentless pursuit of improvement, its highly sophisticated fulfillment software systems have been on a never-ending treadmill of iterations ever since creation. In 2014 alone, Amazon “rolled out 280 major software improvements across the FC (fulfillment center) network. Our goal is to continue to iterate and improve on the design, layout, technology, and operations in these buildings, ensuring that each new facility we build is better than the last.”13
********
Enabled by the two foundational building blocks, meaning the right people (Building Block 2) and the right data, metrics and AI-powered tools (Building Block 3), Amazon is now well set for a long-thought-to-be-impossible mission: building a continuous and accelerating invention machine aimed generating ground-breaking, game-changing and customer behavior-shaping inventions that create new market spaces and economic opportunities of massive magnitude.
It sounds highly intriguing, but also really seems impossible. We are with you. So how to make invention Amazon’s DNA? How to construct an invention machine up to such high aspirations?
We welcome you to the next chapter.