Entertainment Goes Left Brain

It used to be that new TV shows were developed by creative people spitballing, or throwing ideas against the wall, then using small test screenings to determine how audiences might respond. The process was long and nondeterministic, with most ideas being discarded along the way.

In the United States, NBC, CBS, ABC, and Fox were happy to have Netflix as a distributor of their content. “We get to reach whole new audiences, what is not to like?”, said many a network TV executive. But seemingly overnight, Netflix has become a top-five content producer in direct competition with those established TV networks.

You see, Netflix viewed data as currency, and not just as an asset for its IT organization to manage. Netflix builds shows through a proprietary mix of right-brain creative talent and left-brain machine learning insights. It constantly taps granular data from over 100 million subscribers consuming 250 million hours of video per day to anticipate what content viewers will respond best to. That massive data advantage enables Netflix to produce shows that outperform those from other networks by a factor of two to one.

Moneyball Wins Games

As detailed in Michael Lewis’s popular book Moneyball, the Oakland Athletics made the Major League Baseball playoffs in the fall of 2003 with an annual payroll of $44 million. For context, the New York Yankees spent $125 million that same year.

Oakland discovered that letting the data speak determined which players should play better than the gut feel of experienced coaches and managers. For baseball purists this was heresy, but the A’s came out of nowhere to go from laughingstock to pennant contender.

The Oakland A’s organization literally viewed its data as currency. The organization understood that the more games played in the playoffs, the more ticket, merchandise, concession, and TV revenue the team would receive. Once its data showed that on-base and slugging percentages were far better predictors of offensive success than the batting-average and runs-batted-in metrics other teams were using, the A’s were not shy about changing how they made lineup and in-game decisions.

Oakland’s disruptive innovation altered how every team in Major League Baseball is run. The impact has even been felt in other sports, such as basketball and football, where data science teams are growing in funding, size, and importance.

Just as mechanization substituted for physical labor in the agricultural and industrial ages, data is being substituted for mental labor in the digital age. This is causing every industry to fragment into the data rich and the data poor. Which are you going to be?

Think Broadly about Your Data Assets

In our experience, it is not nearly that simple. Even in a small or midsized company, the data sets you’re looking to inventory are scattered across disconnected pockets of your business. Beyond the typical data residing in your IT systems, look for data sets that are:

• Known only to some industrial automation guru in one of your manufacturing plants, where the data is “owned” by production operations, not IT
• Buried in a past acquisition that was sponsored by one of your business units and no one talks about anymore
• Part of an RM&D solution within your services business that is monitoring equipment operation at customer sites
• Hidden in the installed base data management tool in your repair depots as part of service entitlement management
• Aggregated in someone else’s cloud as part of a SaaS (software as a service) application, such as Salesforce for customer information or Workday for employee data
• On a storage-as-a-service platform, such as Amazon S3 or Microsoft Azure Data Lake, that your marketing team chose for a major customer trial program
• Embedded within the indecipherable log files on your servers that track user behavior across enterprise and Web applications
• Available through your channel, marketing, service, or manufacturing partners as part of your contractual arrangements with them
• Located within the instrumentation of your products themselves, but not yet cataloged, backhauled, and aggregated into a usable form

The list could go on, but you get the picture. You have access to more data assets than you think but have fewer under management than you hoped.

Aspiring Goliaths are increasing their focus on one particular area: data from their customers. For example, Waze enhances its own data sets through granular visibility to the activity of every user of its navigation app. This provides Waze with a cheap way to know the real-time speed of traffic in any market in which it has a critical mass of customers.

John Deere has integrated data from sensors in farmers’ fields and equipment to deliver cost-effective predictive maintenance and unlock the potential of precision farming to increase crop yields. The average farm is on its way from generating 190,000 data points per day in 2014 to a projected 4.1 million data points by 2020.

For companies such as Waze and Deere, access to granular customer data presents an enormous opportunity for cocreating new products and services. Ensure that your data asset inventory has sufficient focus on this rich source of potential data rights.

Understand Your Data Liabilities

On the right side of your data balance sheet, you probably have many more data liabilities than you’d expect. For example, you may be licensing data from third parties and incurring either in-cash or in-kind costs. Also, new data security, governance, privacy, and sovereignty regulations, such as GDPR in the European Union, make managing personally identifiable data more expensive and higher risk. In addition to Europe, governments in California, China, and India are becoming increasingly aggressive in this area.

Beyond data licensing and regulatory compliance, you almost certainly have data liabilities to your customers and suppliers in terms of what you can and cannot do with data that comes from your interactions with them. Just as with data assets, it is likely that your management of these data liabilities is far too fragmented today. That is becoming a high-risk approach.

Privacy breaches at brand-name companies, such as Equifax, Target, Saks Fifth Avenue, Panera Bread, Under Armour, and Facebook, have put data privacy front and center in the minds of customers. Companies like these are building high-value, hyperpersonalized experiences around granular, customer-identified data.

If you’re estimating the liability side of your data balance sheet, be clear about the true costs of acquiring and protecting that data. Catalog your obligations to be transparent about how data is used and secure the necessary data rights as you undertake your digital transformation. Aspiring Goliaths are pursuing new models of data anonymization and even providing customers with simple self-service portals that allow them to opt in or out at multiple levels of data sharing.

Rate the Quality of Each Data Asset

Now that you have your data balance sheet in place, it is time to rate each data asset in terms of quality. Company after company boasts about how big its big data is. “Our airplane produces multiple terabytes of sensor data per flight.” “We can see the condition of every bolt on our oil rigs.” “We have information on a hundred million customer purchases.”

While size does matter, it is just part of the story. The dirty little secret is that most of that data is unused—what we call “dark data.” It is sitting idle for a reason. Most of it is low quality. It lacks the attributes that make it useful in improving your core business and winning in adjacent markets.

Aspiring Goliaths are more sophisticated. They rate the assets on their data balance sheet against seven indicators of data quality:

1. Freshness. How recent is the data, and therefore, how predictive is it likely to be when applied to your current products, services, and operations?
2. Duration. Does the data cover an extended time period that reflects the normal seasonality and business cycle variations inherent in your business?
3. Context. Is the data tagged with metadata that makes the data valuable in understanding an end-to-end business process or customer interaction?
4. Consistency. Is each data item collected in a consistent way over time to allow trending and correlation analyses to be run with high statistical validity?
5. Attributability. Can the data in question be attributed to specific customers or machines, or has it been de-identified for data privacy reasons?
6. Reach. Does your data reach beyond the installed base of your current customers, products, and services to be reflective of your industry as a whole?
7. Exclusivity. Do you enjoy exclusive data rights, or is that data available to your current and potential competitors as well?

Pay Special Attention to the Edge

Domo’s Data Never Sleeps analysis shows that 2.5 quintillion bytes of data are being produced each day. That is 2,500,000,000,000,000,000 bytes. Forbes reports that 90% of the data in the world today was produced in just the past two years. Gartner predicts that by 2022, 75% of enterprise-generated data will be created and processed outside of traditional data centers or cloud infrastructure. That is up from just 10% today.

Most of this data is being produced at the edge—machines, cars, phones, and sensor networks. Little of it is backhauled to traditional IT systems today. It dwarfs much of the structured data that you’ve inventoried in your data balance sheet.

A few examples. Self-driving cars harvest massive amounts of sonar, radar, and video sensor data to learn about their environments and make real-time decisions about how to drive safely. Nike has made significant investments to extend its Nike+ platform of connected sneakers and wearables to collect granular data on over seven million runners. Under Armour spent nearly $560 million to acquire MapMyFitness, MyFitnessPal, and Endomondo to access data on 150 million digital fitness users, then launched Healthbox to extend its data collection into health and wellness. In industrial markets, platforms such as GE’s Predix and Hitachi’s Lumada collect and bring context to massive stores of machine and operational data.

As you finalize your data balance sheet, make sure that you have fully considered the future value of your equivalents of these edge data stores. In a digital future, you may find them more important than the ones you’ve traditionally run your business on.

Digital Customer Segments

Your efforts to maximize your return on data need to focus on four customer segments: internal make-it and sell-it teams, as well as external supply chain partners and customers. Just like Watson or AlphaGo, your goal is to augment the expertise of the people in that segment in order to make better decisions faster and more often. You will be shifting your company from fact-free to fact-based operations.

First up are your internal make-it teams, which might work in functions such as product management, design, business incubation, R&D, engineering, manufacturing, and service delivery. They invent, design, develop, produce, and deliver the products and services that your customers purchase. Those teams have the potential to act as both suppliers of new data—by designing additional digital instrumentation into their offers—and consumers of algorithmic insights. Show them which capabilities customers most value and they can stop designing ones that customers do not use. Help them understand the real-life way that customers use your products and they can manufacture them to fail less often and require less service. Identify the unexpected ways that customers want to use your products and services and your make-it teams can dream up entirely new solutions to grow your business.

Second, your internal sell-it teams include functions such as sales, marketing, lead generation, pricing, and distribution channels. They work to extract the maximum gross margin from the maximum number of customers every week, month, and quarter. Show them how to calculate prices that optimize across gross margin percentage and revenue growth rate for maximum gross profit dollars. Help them optimize their digital and traditional marketing spend for lowest cost per converted lead. Enable them to see future demand through the scoring of prospective customer activities—reading white papers, watching videos, subscribing to newsletters, attending webinars—that represent expressions of interest. Give every sales representative the next button that tells them the most likely customer in their assigned accounts to buy and the offer most likely to convert to a profitable sale. Propose get-well actions for end customers that have made a purchase of an offer but have not yet consumed it. Identify the most important impediments to customers buying and consuming more, so that your sell-it teams can design out the commercial friction.

Third, make sure your external supply chain partners can be reached via your procurement and supply chain planning teams. You might ask, “Why would I commit scarce algorithmic resources to help my suppliers?” Well, you are not doing it for free. Companies such as Walmart have become remarkably sophisticated at using data as currency in a literal way with their suppliers. Give them more granular visibility to demand so that they can avoid both excess inventories and out-of-stocks while balancing production in their factories efficiently. Provide them with detailed benchmarks about how their on-time, on-quality, on-quantity performance varies from their peers and what they can do to improve. Help them see through to end customer usage of their products so that they can design them for higher value, reliability, and performance.

Finally, external channel partners and end customers are increasingly dependent on your algorithmic insights. Help your channel partners see the most valuable ways to combine your products and services for maximum commercial velocity. Give your B2B customers visibility to the installed base of your products in their operations at a level of detail their own IT teams cannot match. Enable your best customers to evolve from time-based to condition-based maintenance for both higher uptime and lower overall maintenance costs. Provide your end customers with anonymized benchmarks for how other customers get more value from your offers more quickly and help your channel partners target complementary products and in those areas.

This is by no means an exhaustive list. It is merely a starting point in your efforts to prioritize the highest-value algorithmic insights within each of these four digital customer segments.

Master Machine Learning

Let’s turn our attention to the middle row of Figure 6.2—the algorithms that realize the value of your data for each of those four digital customer segments. It would not be hyperbole to say there has been an explosion of activity in this middle layer.

In year 2000, there were just 570 patent applications that mentioned “algorithm” in their title or description. By 2015, there were 17,000 such patent applications, a number estimated to reach 500,000 by 2020. If you are not already focused on acquiring or building an algorithmic advantage, then you are falling behind.

Some of these algorithms are horizontal in nature and can often be sourced from third parties. Examples include remote asset monitoring solutions for services, propensity to buy models for sales, cyber intrusion detection for IT, and spam filtering for all of us.

However, many algorithms are industry specific and likely to require internal development. Examples include fraud detection for financial services, price elasticity models for packaged consumer goods, object identification for self-driving cars, predictive maintenance models for sophisticated industrial equipment, and emergency room throughput optimization in hospitals. New algorithms in healthcare are even outperforming certified radiologists in diagnosing acute conditions, such as pneumonia and stroke.

The range of algorithms is broad, but most companies focus on five algorithm categories: asset optimization, operations optimization, risk optimization, pricing optimization, and sales optimization. Note that the internal and external customer segments on the top row of Figure 6.2 seek algorithmic advantage from many of these categories.

These algorithms can be built in two different ways—data science teams and continuous machine learning. Both are still important, but many aspiring Goliaths are in the midst of shifting their focus from the first to the second one.

Traditionally, companies seeking an algorithmic advantage have made a sizable investment in hiring a chief digital officer or chief data officer and staffing up a data science team. These highly trained (and compensated) statisticians, modelers, and programmers specialize in each of the five categories of algorithms in the middle row of Figure 6.2.

In fact, for the past five years data scientist has been one of the hottest jobs in Silicon Valley. Parents have stopped telling their kids to become doctors and lawyers so that they can focus their studies on math and statistics. As shown in data from McKinsey in Figure 6.3, demand for data scientists is dramatically exceeding supply.

For context, there were just 150,000 people employed in data science 10 years ago. Now the gap between supply and demand is almost that big. These individuals require a unique combination of a masters- or PhD-level math background, advanced data modeling and programming skills, and domain knowledge of the business problem that the algorithm seeks to solve. While data science is not going away anytime soon, basing your future algorithmic advantage on the hiring of a big-data science team is going to take too long and cost too much.

The second approach has greater potential for long-term success. Continuous machine learning is, in many ways, the great equalizer. Given the broad availability of both on-premise and cloud-based machine learning tools, it is becoming possible for every small, midsized, and large company to let its data speak.

At the most simple level, continuous machine learning is an automated way to find actionable patterns in large data sets. You might hear terms such as deep learning, neural networks, and cognitive models as specific approaches to how computers find nonintuitive patterns on their own.

While the basic math has been around for decades, an arms race is underway between Amazon, Microsoft, Google, Baidu, and Tencent to provide your company with a ready-to-deploy platform for continuous machine learning. These tech titans are taking advantage of step- function advances in computing horsepower through AI-optimized GPUs, the availability of massive integrated data sets to train new models, and ready access to AI development platforms, such as Google TensorFlow and Amazon AI.

This is great news for you and your company. The pace of innovation is rapid and accelerating. The cost of entry is coming down. If you are just starting out, self-directed platforms such as Google Cloud’s AutoML and Microsoft Azure Machine Learning allow your core developers to use your existing data to build algorithms without an army of data scientists.

As new methods such as reinforcement learning—machines training machines—continue to advance, you can expect this reduced reliance on data scientists to accelerate. The ultimate goal is to allow business people to ask the right questions of continuous machine learning platforms directly. Maybe the supply of doctors and lawyers in Silicon Valley will be safe after all.

For most aspiring Goliaths, investing in both data science and continuous machine learning is the best path to building your algorithmic advantage. Focus your data scientists on developing high-value algorithms, where their unique combination of math skills and domain knowledge is essential. In parallel, unleash continuous machine learning on those same areas, as well as the next tier of use cases for your four digital customer segments.

In the overlap areas, play king of the hill. That is, let the best approach win through ongoing A-B testing between data scientist–developed and continuous machine learning–derived algorithms. In the next tier use cases, let time be your friend as continuous machine learning solutions tend to get better with more time and bigger data.

Now that you’ve built your data balance sheet, learned to value data optionality, and used the digital value stack to maximize your return on data, let’s turn to a case example of a company that is certainly using its data as currency.

Company Readiness Self-Assessment

To complete your company readiness self-assessment, thoroughly read the grid in Figure 6.4, then identify the level of capability that your company has demonstrated within each row.

Career Readiness Self-Assessment

Repeat the exercise above with your career in mind. What roles are you playing in helping your company use its data as currency? Mark your self-assessments on the grid shown in Figure 6.5.

Rule 3 Readiness Summary

Now that you’ve completed your company and career self- assessments for Rule 3, fill in your readiness summary in Figure 6.6. Again, if you are doing these self-assessments online at www.goliathsrevenge.com, this readiness summary will be produced for you automatically.

You are halfway there—three of the six rules completed. Take a breath, and then it’s time to open up your innovation as TWC did using Rule 4: Accelerate through innovation networks.