It’s often said that you don’t really understand something until you can express it as an algorithm. As Richard Feynman said, “What I cannot create, I do not understand.”¹²

—PEDRO DOMINGO, THE MASTER ALGORITHM

In 2004, I attended a senior executive team meeting at Amazon that just so happened to coincide with the Salesforce.com IPO. During the meeting, one of the other executive team members casually commented that Salesforce was the world’s largest customer relationship management (CRM) technology company.

Big mistake.

A senior Amazon leader (guess who) immediately reacted. “We are the world’s largest CRM company!”

The point was clear. Like a CRM company, Amazon is obsessed with managing and analyzing the data around customer interactions to improve its relationships with them. And it is doing it on a much bigger scale than Salesforce. The digital nature of Amazon’s business and its focus on collecting obscene amounts of data actually make it significantly different than the average e-commerce company.

IoT empowers you to collect data about what’s going on in your business at a scale and magnitude never seen before. But unless, like Amazon, you’re willing to take that data and develop models, analytics, and algorithms and “do the math” around what this data means, you’ll miss out on the value of this huge asset.

Principle 4: Using mathematical equations and the Internet of Things, you can track the levers and processes of your business, learn more about specific processes, and gather data that will power and inform those equations, driving improvements and efficiencies.

In this chapter, I’ll take you through the basics of how Amazon and other leading companies use analytics and algorithms to improve their business.

AN EXPLOSION OF DATA

Most organizations primarily collect core transactional data used to run their business—orders processed, inventory tracking, customer interactions, and financial transactions. The Internet, of course, has created a new sphere of data collection that’s richer than that core transactional set, but the Internet of Things is leading to a data explosion.

It is estimated that the “data universe” is doubling every two years. Ninety percent of all stored data has been generated in the past two years. By 2020 there will be forty zettabytes (forty trillion GBs).¹³ The amount of stored data generated by IoT will grow from 2 percent in 2013 to 10 percent by 2020.

One of the biggest risks for a businesses in this environment is that they spend valuable time and resources collecting data, but they aren’t actually able to use that data to successfully drive business outcomes.

You may not be surprised to hear that at Amazon, there is a specific and methodological way of tracking and analyzing data to inform business decisions. It’s an essential step in taking advantage of the power of Internet of Things, as we’ve been discussing it so far.

AMAZON’S CULTURE: IN GOD WE TRUST. ALL OTHERS MUST BRING DATA

Amazon.com has a culture of data-driven decision-making and demands insights and recommendations that are timely, accurate, and actionable. At Amazon, you will be working in one of the world’s largest and most complex data environments.¹⁴

—MULTIPLE AMAZON JOB LISTINGS

On one Wednesday afternoon, I stepped into the office of an Amazon senior finance leader. There on his wall was a poster bearing Edward Deming’s famous quote: “In God We Trust. All Others Must Bring Data.”

It was a clear measure of the culture across Amazon, where as much or more time is spent defining and agreeing on how to measure a new feature, service, or product as designing the feature itself. Teams spend weeks considering the inputs and outputs of an operation and what data might be needed to run that operation and understand its complex internal workings.

The general scope of that approach follows four steps:

Step 1. Define the decision-making logic. Develop a strong understanding of how decisions in a process are currently being made and how they should be made.

Step 2. Build an equation (or equations) to describe that decision-making logic.

Step 3. Improve the data you’re collecting (use connected devices as part of your approach).

Step 4. Use that data to power the formulas. Use the formulas to drive the decision-making logic.

A small subset of teams are focused on bigger, long-term capabilities. They take this focus even further by creating mathematical equations to model key business functions they’re responsible for tracking. These equations are known as fitness functions.

In addition to the insight a fitness function provides, its development builds clarity and agreement around the core values and accountability of the team. Unlike standard team metrics, a completed fitness function has to be approved by Amazon’s executive team.

We used to joke that getting a fitness function approved was more difficult than getting canonized.

Once a team has agreed on its metrics (or, in some cases, its fitness function), the team’s weekly rhythm is defined by a cascading set of metrics meetings. A successful metrics meeting includes clear conversations about what happened during the course of the week and why. The team will identify any needed fixes or adjustments, as well as who will be responsible for the implementation of those fixes. Although these meetings are called “metrics meetings,” really what they are is “accountability” meetings.

The saying “If you can’t measure it, you can’t manage it,” is company religion. To ensure accountability, the ownership of each metric is assigned to an individual. This happens even if, as is often the case, all of the factors driving that metric aren’t in the owner’s reporting structure. As the owner of a specific metric, you are expected to seek its improvement and understand its root cause factors.

Accountability isn’t the only benefit of this approach. The approach also lessens bureaucracy and keeps organizational structure from interfering with progress. A team can operate relatively autonomously once its metrics are approved. It can set its own priorities and identify innovation strategies.

So team metrics keep individual teams in line, but how does Amazon make sure the larger company is progressing on schedule in the right direction?

To keep teams accountable to one another and set high-level standards, Amazon relies on another level of accountability known as a service-level agreement (SLA). SLAs are guarantees of critical performance that can exist not only internally between teams but also externally between Amazon and its customers or other third parties.

An SLA that guarantees on-time order delivery might apply to a range of teams within the company, but it also sets accountability to third-party vendors and customers. On a daily basis, a team’s metrics will be vigorously compared to SLAs.

INSTRUMENTATION: REAL-TIME, FINE-GRAINED DATA

Once a team’s metrics and SLAs are in place, the focus turns to data collection that will inform those metrics and SLAs. At Amazon, there are very specific standards for the quality and type of data a team should collect. Amazon’s executive team refers to those standards as “instrumentation.”

Expectations around data collection—or instrumentation—at Amazon are twofold: First, data should be fine grained in nature. You can always summarize and aggregate data, but you can’t go back and derive more detail from a dataset.

Secondly, that data should be available in real time. You can always batch data up or slow it down, but you can’t always speed it up. Design for no time lags and no batch systems.

There are lots of reasons this is important. Let’s say a grocery company uses a refrigerated storage bin to keep its fruits and vegetables fresh. Suddenly, over the course of a day, all the lettuce goes bad. If the company has been collecting fine-grained data—for example, any changes in temperature or pressure and their time stamps—that grocery company might actually be able to figure out what it was that caused the lettuce to go bad. Otherwise, they’re stuck wondering about the possible variables and causes in that situation.

Of course, it’s not possible to collect fine-grained real-time data in every situation. There are limitations to the nature of the data that you’re able to collect in a specific situation, but at Amazon it’s expected that you will work vigorously to achieve instrumentation.

SCALING DATA COLLECTION AND WORK

In some cases, it doesn’t yet make sense to use connected devices and the Internet of Things to collect data. If you’re conducting a trial or a short-term low-tech experiment, you might actually want to collect some data manually. And some mundane or routine tasks actually can’t be automated.

In those cases, it’s best to sit back and accept that what you’re facing is really what Bezos calls OPW, or other people’s work. These are tasks that you don’t want your team members wasting their time or talent on. In these cases, your goal should be to figure out how to get others to do these mundane, routine jobs. Luckily, Amazon has built a service for exactly this situation. It’s called Mechanical Turk.

Amazon Mechanical Turk is a marketplace for work that requires human intelligence. The Mechanical Turk service gives businesses access to a diverse, on-demand, scalable workforce and gives workers a selection of thousands of tasks to complete whenever it’s convenient.

Amazon Mechanical Turk is based on the idea that there are still many things that human beings can do much more effectively than computers, such as identifying objects in a photo or video, performing data deduplication, transcribing audio recordings, or researching data details. Traditionally, tasks like this have been accomplished by hiring a large temporary workforce (which is time consuming, expensive, and difficult to scale) or have gone undone.¹⁵

When developing an OPW strategy, one option is to turn to Mechanical Turk. Still, even with Mechanical Turk, you’ll face quality problems. Any time a person is involved, you’ll face limits to the data collected, real labor costs, and, in some cases, lag time.

If at all possible, you should automate even basic data collection.

CONVERT ANALOG TO DIGITAL

One of the inherent benefits of a native digital business is the ability to collect much more data than traditional businesses in the course of day-to-day business operations. That extra data is known as “digital exhaust.” When a company’s core business and customer experiences are digital in nature—centered mainly on desktop, mobile, or some other e-commerce model—it’s relatively easy to collect metrics and measures across the business.

While all digital businesses throw off this digital exhaust, only some collect enough of it to create the formulas and algorithms that would improve business outcomes. A company that successfully uses its digital exhaust might combine the browsing and online-shopping history of a customer with buying history of other customers in their segment to make personalized recommendations or offers, either while they are shopping or later over e-mail.

This is in contrast to traditional businesses and processes (think about your average corner grocer or distributor), which don’t have that inherent data advantage. Any data captured is typically the result of required transactions or key process steps. And much of it is collected and input manually.

The nature of these analog processes creates high costs and high barriers to collect lots of data and, just as importantly, makes it very difficult to make real-time adjustments.

IoT is the “analog to digital world” converter.

By installing sensors in everyday processes, an analog business can actually put the plumbing in place to create and collect the same insights and instrumentation as a purely digital process.

Obviously, even with IOT, there are real costs of collecting, transmitting, processing, and storing data. Those costs mean that there needs to be thoughtful rationalization about how much data is enough. Still, relative to the data-starvation diet most existing processes rely on today, the opportunities for data collection among even small analog businesses are huge.

THE SKINNY ON ALGORITHMS

The general perception of algorithms is that they’re extremely complicated and opaque. And they can be, but don’t let this intimidate you. Algorithms are really just formulas or rules that use data to make a prediction, recommendation, or decision. Oftentimes, the best algorithms are fairly simple rule-based systems that help smart humans scale their decision making.

You can use algorithms to make real-time improvements in your business process based on outside or external factors or to make better decisions about the things you need to do.

An algorithm could be used to optimize a warehouse worker’s pick path (the route he or she takes through the picking area of a warehouse while selecting merchandise). It might use historical information to recommend how much inventory to purchase. It might even determine which promotion to offer customers based on their purchase history or browsing data.

How do you start? It all starts with being able to define what question or problem you are trying to answer or optimize and then “do the math” to create the formula and define the variables. (More on that in just a second.)

At some point in the process, you will need to employ true mathematical expertise to turn amateur rules and algorithms into robust and sophisticated applied mathematical models. If that sounds expensive, it’s because it is. But it’s also worth it.

Bezos wrote about this in Amazon’s 2010 shareholder letter. After giving readers a glance into the computer and data science advancements they have used to scale their business, he wrote, “Now, if the eyes of some shareowners dutifully reading this letter are by this point glazing over, I will awaken you by pointing out that, in my opinion, these techniques are not idly pursued—they lead directly to free cash flow.”¹⁶

HOW TO GET STARTED IN YOUR BUSINESS? DO THE MATH

Measurement is the first step that leads to control and eventually to improvement. If you can’t measure something, you can’t understand it. If you can’t understand it, you can’t control it. If you can’t control it, you can’t improve it.

—H. JAMES HARRINGTON

Do you know the formula for your business process? Do you know what the input variables are versus the output variables? Working with my clients, I do a significant amount of process-reengineering and process-improvement work.

When I get started, an early question (and test) I give is to ask three questions:

1. First, do you have a sufficiently deep and accurate definition of the process?
2. Second, can you walk me through a balanced (cost, quality, throughput) set of metrics for the process (and show me today’s metrics in addition to last month’s metrics)?
3. Third, can you write me a formula for the process?

The answers tend to range. Many have a definition of their process, but it’s not deep or accurate enough to explain how their business really works. Most have some metrics, but they’re generally single sided or unbalanced. And as for that last question—a formula for their business process? Most often I work with clients who have no idea what a formula for their business might look like.

What would it look like to really understand these things about your business?

Let’s take a deeper look at the case of Clifford Cancelosi, an ex-Amazon leader and former colleague of mine. Until recently, Clifford was a leader at a national home-service appliance-installation and repair business.

When Clifford started in the home-repair business, customers were consistently waiting seven to ten days for a scheduled appointment with a technician. This made it hard to create a business roadmap or even to prioritize the urgency of specific customers.

Luckily, based on his Amazon experience, Clifford knew what to do. He calls it “do the math.” So he set to work creating a set of equations to determine his daily effective repair capacity. After some thought, he realized that, at a high level, the company’s effective daily repair capacity for each technician was a function of three variables:

• The mean time it takes a technician to complete a job
• The mean time it takes a technician to move from job to job
• The percentage of times a repair job was completed in one visit

The formula for the effective capacity of a technician is then this:

(eight hours * percentage first-time completes) / (mean time to complete job + mean touting time between jobs) = effective daily capacity

So, if the mean time to complete the job is 2 hours, the mean time to move between job locations is 0.5 hours, and the percentage of first-time completes is 75 percent, the effective capacity becomes

(8*.75) / (2+.5) = 2.4

That’s 2.4 effective jobs per eight-hour day.

Once Clifford had this equation, he could turn each variable into a metric—first-time completes, mean routing time, and so forth. He tracked each to keep tabs on that specific part of the business’s operations.

From there, he analyzed the possible errors that might affect each metric. In the case of first-time completes, these included the following:

• Technician efficacy
• Wrong part on truck
• No part on truck
• Scheduling inaccuracy

The more the company grew to understand the subequations within the formula and what drove variations in each metric, the better understanding they had of how to improve business performance. That deep understanding allowed the company to build formulas to augment its manual decision making.

Today, that home-appliance-repair business has significantly improved its effective capacity formula. It has strengthened its first-time complete metric by creating a hierarchy of metrics that measures the critical customer experience and the root causes of each metric to improve the metric, reduce variability, and reduce costs.

If you’re struggling to understand how to measure and improve your business processes, this is a great way to start.

1. Pick a key process or customer experience. (In this case, the number of jobs a technician can complete in a day.)
2. Define the hierarchy of metrics. (What are the factors that affect that process?)
3. Build a formula from the variables.

Once you have your basic formulas, it becomes much easier to understand which parts of the processes could benefit from more data collection by connected devices.

In Clifford’s appliance repair business, for example, sensors were used to capture

• The actual routes of company delivery trucks and the actual time between stops using company-provided tablets. Once the company had actual route and wait-time data, they overlaid it with a driver’s planned route and expected routing time to identify factors that might improve the efficiency of drivers—for example, eliminating unscheduled stops, increasing productivity on each job, and creating more efficient routes for drivers.
• The movement of key inventory using RFID sensors. RFID sensors allowed company leaders to see when key inventory was loaded on a truck and when it was removed. This not only helped eliminate shrinkage, but it allowed them to prepare for inventory needs.

The next step for the appliance-repair business would be to work with appliance manufacturers to include connected sensors in the appliances themselves. That would allow the company to understand the problem with the appliance—and any needed parts—before the technician arrives, leading to better “first-time fix” metrics and, eventually, the ability to detect, model, and predict appliance failures.

CREATE EQUATIONS FOR KEY TOPICS AND PROCESSES

We’re using software and algorithms to make decisions rather than people, which we think is more efficient and scales better.

—BRIAN OLSAVKY, AMAZON CFO¹⁷

The hardest part of “doing the math” is getting started. It is a process that requires a deep understanding of your company’s operating environment, a talent for math, and a willingness to be transparent—all of which the average person inherently finds reasons to avoid.

It’s far easier to spend hours debating subjective reasons for the variances in your operating environment than it is to double down and perfect your equations.

One of the most important things to remember when getting started with the math is to not let the perfect be the enemy of the good. Don’t let yourself get hung up on fully optimizing your processes right away. The concept of a fully optimized business process is intimidating for most people. Rather than setting yourself up to give up before you even begin, focus on “doing the math” to drive continuous improvement.

Keeping your eyes on the prize—a key business objective that can improve the customer experience, increase the efficiency of business operations, and scale these impacts financially—is critical to “doing the math.”