- 16th April 2018
- Posted by: Manolis
What’s in the unexplored recesses of the “deep web”? An astonishing amount of raw data. And for the first time, technology is beginning to allow CIOs, business leaders, and data scientists ways to look at that data—and ways to unearth valuable business, customer, and operational insights.
Across enterprises, ever-expanding stores of data remain unstructured and unanalyzed. Few organizations have been able to explore nontraditional data sources such as image, audio, and video files; the torrent of machine and sensor information generated by the Internet of Things; and the enormous troves of raw data found in the unexplored recesses of the “deep web.” However, recent advances in computer vision, pattern recognition, and cognitive analytics are making it possible for companies to shine a light on these untapped sources and derive insights that lead to better experiences and decision making across the business.
Until recently, taking a passive, backward-looking approach to data and analytics was standard practice. With the ultimate goal of “generating a report,” organizations frequently applied analytics capabilities to limited samples of structured data siloed within a specific system or company function. Moreover, nagging quality issues with master data, lack of user sophistication, and the inability to bring together data from across enterprise systems often colluded to produce insights that were at best limited in scope and, at worst, misleading.
Today, CIOs harness distributed data architecture, in-memory processing, machine learning, visualization, natural language processing, and cognitive analytics to answer questions and identify valuable patterns and insights that would have seemed unimaginable only a few years ago. Indeed, analytics now dominates IT agendas and spend. In Deloitte’s 2016 Global CIO Survey of 1,200 IT executives, respondents identified analytics as a top investment priority. Likewise, they identified hiring IT talent with analytics skills as their top recruiting priority for the next two years.1
Leveraging these advanced tools and skill sets, over the next 18 to 24 months an increasing number of CIOs, business leaders, and data scientists will begin experimenting with “dark analytics”: focused explorations of the vast universe of unstructured and “dark” data with the goal of unearthing the kind of highly nuanced business, customer, and operational insights that structured data assets currently in their possession may not reveal.
In the context of business data, “dark” describes something that is hidden or undigested. Dark analytics focuses primarily on raw text-based data that has not been analyzed—with an emphasis on unstructured data, which may include things such as text messages, documents, email, video and audio files, and still images. In some cases, dark analytics explorations could also target the deep web, which comprises everything online that is not indexed by search engines, including a small subset of anonymous, inaccessible sites known as the “dark web.” It is impossible to accurately calculate the deep web’s size, but by some estimates it is 500 times larger than the surface web that most people search daily.2
In a business climate where data is competitive currency, these three largely unexplored resources could prove to be something akin to a lottery jackpot. What’s more, the data and insights contained therein are multiplying at a mind-boggling rate. An estimated 90 percent of all data in existence today was generated during the last five years.3 The digital universe—comprising the data we create and copy annually—is doubling in size every 12 months. Indeed, it is expected to reach 44 zettabytes (that’s 44 trillion gigabytes) in size by 2020 and will contain nearly as many digital bits as there are stars in the universe.4
What’s more, these projections may actually be conservative. Gartner Inc. anticipates that the Internet of Things’ (IoT) explosive growth will see 20.8 billion connected devices deployed by 2020.5 As the IoT expands, so will the volumes of data the technology generates. By some estimates, the data that IoT devices will create globally in 2019—the bulk of which will be “dark”—will be 269 times greater than the amount of data being transmitted to data centers from end-user devices and 49 times higher than total data-center traffic.6 Against this statistical backdrop, big data, as a business imperative, might be more accurately described as “enormous data.”
To date, companies have explored only a tiny fraction of the digital universe for analytic value. IDC estimates that by 2020, as much as 37 percent of the digital universe will contain information that might be valuable if analyzed.7
But exactly how valuable? IDC also projects that organizations that analyze all relevant data and deliver actionable information will achieve an extra $430 billion in productivity gains over their less analytically oriented peers by 2020.8
LET THERE BE LIGHT
When we think about analytics’ potential, the possibilities we often envision are limited to the structured data that exists within the systems around us. Dark analytics seeks to remove those limits by casting a much wider data net that can capture a corpus of currently untapped signals.
Dark analytics efforts typically focus on three dimensions:
Untapped data already in your possession: In many organizations, large collections of both structured and unstructured data sit idle. On the structured side, it’s typically because connections haven’t been easy to make between disparate data sets that may have meaning—especially information that lives outside of a given system, function, or business unit. For example, a large insurance company mapped its employees’ home addresses and parking pass assignments with their workplace satisfaction ratings and retention data. The effort revealed that one of the biggest factors fueling voluntary turnover was commute time—the combination of distance to the office, traffic patterns based on workers’ shift schedule, degree of difficulty in finding a parking spot, and length of walk from car to their workspace.
Regarding “traditional” unstructured data, think emails, notes, messages, documents, logs, and notifications (including from IoT devices). These are text-based and sit within organizational boundaries but remain largely untapped, either because they don’t live in a relational database or because until relatively recently, the tools and techniques needed to leverage them efficiently did not exist. Buried within these unstructured data assets could be valuable information on pricing, customer behavior, and competitors. Particularly in multinational companies, they may also contain potentially valuable yet untranslated data assets created for or generated in non-English-speaking markets.
What percentage of the data in existence today is unstructured? No one knows for sure. The generally accepted figure has long been 80 percent—known as “the 80 percent rule”—though recent estimates put the number closer to 90 percent.9
Nontraditional unstructured data: The second dark analytics dimension focuses on a different category of unstructured data that cannot be mined using traditional reporting and analytics techniques—audio and video files and still images, among others. Using computer vision, advanced pattern recognition, and video and sound analytics, companies can now mine data contained in nontraditional formats to better understand their customers, employees, operations, and markets. For example, a retailer may be able to gain a more nuanced understanding of customer mood or intent by analyzing video images of shoppers’ posture, facial expressions, or gestures. An oil and gas company could use acoustic sensors to monitor pipelines and algorithms to provide visibility into oil flow rates and composition. An amusement park could gain greater insight into customer demographics by analyzing security-camera footage to determine how many customers arrive by car, by public transportation, or by foot, and at what times during the day.
Low-cost high-fidelity surround cameras, far-field microphone areas, and high-definition cameras make it possible to monitor all business activities taking place within an enterprise. The ability to apply analytics to audio and video feeds in real time opens up profound new opportunities for signal detection and response. Such digital cues offer new ways of answering existing questions and exploring new opportunities. Moreover, in recent years data storage costs have declined by an estimated 15 to 20 percent, making the archiving of image and audio files a more realistic option for smaller organizations.10
Data in the deep web: As a dimension of dark analytics, the deep web offers what may contain the largest body of untapped information—data curated by academics, consortia, government agencies, communities, and other third-party domains. But the domain’s sheer size and distinct lack of structure can make it difficult to search. For now, only data mining and analytics efforts that are bounded and focused on a defined target—for instance, licensable data owned by a private association—will likely yield relevant, useful insights. Just as the intelligence community monitors the volume and context of deep web activity to identify potential threats, businesses may soon be able to curate competitive intelligence using a variety of emerging search tools designed to help users target scientific research, activist data, or even hobbyist threads found in the deep web. For example, Deep Web Technologies builds search tools for retrieving and analyzing data that would be inaccessible to standard search engines.11 Its software is currently deployed by federal scientific agencies as well as several academic and corporate organizations. Stanford University has built a prototype engine called Hidden Web Exposer that scrapes the deep web for information using a task-specific, human-assisted approach. Other publicly accessible search engines include Infoplease, PubMed, and the University of California’s Infomine.12
FLASHLIGHT, NOT INTERPLANETARY STAR
To be clear, the purpose of dark analytics is not to catalog vast volumes of unstructured data. Casting a broader data net without a specific purpose in mind will likely lead to failure. Indeed, dark analytics efforts that are surgically precise in both intent and scope often deliver the greatest value. Like every analytics journey, successful efforts begin with a series of specific questions. What problem are you solving? What would we do differently if we could solve that problem? Finally, what data sources and analytics capabilities will help us answer the first two questions?
Answering these questions makes it possible for dark analytics initiatives to illuminate specific insights that are relevant and valuable. Remember, most of the data universe is dark, and with its sheer size and variety, it should probably stay that way.
Lessons from the front lines
IU HEALTH’S RX FOR MINING DARK DATA
As part of a new model of care, Indiana University Health (IU Health) is exploring ways to use nontraditional and unstructured data to personalize health care for individual patients and improve overall health outcomes for the broader population.
Traditional relationships between medical care providers and patients are often transactional in nature, focusing on individual visits and specific outcomes rather than providing holistic care services on an ongoing basis. IU Health has determined that incorporating insights from additional data will help build patient loyalty and provide more useful, seamless, and cost-efficient care.
“IU Health needs a 360-degree understanding of the patients it serves in order to create the kind of care and services that will keep them in the system,” says Richard Chadderton, senior vice president, engagement and strategy, IU Health. “Our organization is exploring ways to mine and analyze data—in much the same way consumer-oriented companies are approaching customer data—to develop this deeper understanding.”13
For example, consider the voluminous free-form notes—both written and verbal—that physicians generate during patient consultations. Deploying voice recognition, deep learning, and text analysis capabilities to these in-hand but previously underutilized sources could potentially add more depth and detail to patient medical records. These same capabilities might also be used to analyze audio recordings of patient conversations with IU Health call centers to further enhance a patient’s records. Such insights could help IU Health develop a more thorough understanding of the patient’s needs, and better illuminate how those patients utilize the health system’s services.
Another opportunity involves using dark data to help predict need and manage care across populations. IU Health is examining how cognitive computing, external data, and patient data could help identify patterns of illness, health care access, and historical outcomes in local populations. The approaches could make it possible to incorporate socioeconomic factors that may affect patients’ engagement with health care providers.
“There may be a correlation between high density per living unit and disengagement from health,” says Mark Lantzy, senior vice president and chief information officer, IU Health. “It is promising that we can augment patient data with external data to determine how to better engage with people about their health. We are creating the underlying platform to uncover those correlations and are trying to create something more systemic.
The destination for our journey is an improved patient experience,” he continues. “Ultimately, we want it to drive better satisfaction and engagement. More than deliver great health care to individual patients, we want to improve population health throughout Indiana as well. To be able to impact that in some way, even incrementally, would be hugely beneficial.” 14
Retailers make it personal
Retailers almost universally recognize that digital has reshaped customer behavior and shopping. In fact, $0.56 of every dollar spent in a store is influenced by a digital interaction.15 Yet many retailers—particularly those with brick-and-mortar operations—still struggle to deliver the digital experiences customers expect. Some focus excessively on their competitors instead of their customers and rely on the same old key performance indicators and data.16
In recent years, however, growing numbers of retailers have begun exploring different approaches to developing digital experiences. Some are analyzing previously dark data culled from customers’ digital lives and using the resulting insights to develop merchandising, marketing, customer service, and even product development strategies that offer shoppers a targeted and individualized customer experience.
Stitch Fix, for example, is an online subscription shopping service that uses images from social media and other sources to track emerging fashion trends and evolving customer preferences. Its process begins with clients answering a detailed questionnaire about their tastes in clothing. Then, with client permission, the company’s team of 60 data scientists augments that information by scanning images on customers’ Pinterest boards and other social media sites, analyzing them, and using the resulting insights to a develop a deeper understanding of each customer’s sense of style. Company stylists and artificial intelligence algorithms use these profiles to select style-appropriate items of clothing to be shipped to individual customers at regular intervals.17
Meanwhile, grocery supermarket chain Kroger Co. is taking a different approach that leverages Internet of Things and advanced analytics techniques. As part of a pilot program, the company is embedding a network of sensors and analytics into store shelves that can interact with the Kroger app and a digital shopping list on a customer’s phone. As the customer strolls down each aisle, the system—which contains a digital history of the customer’s purchases and product preferences—can spotlight specially priced products the customer may want on 4-inch displays mounted in the aisles. This pilot, which began in late 2016 with initial testing in 14 stores, is expected to expand in 2017.18
Expect to see more pilots and full deployments such as these in the coming months as retailers begin executing customer engagement strategies that could, if successful, transform both the shopping experience and the role that nontraditional data plays in the retail industry.
GREG POWERS, VICE PRESIDENT OF TECHNOLOGY,
As a leader in the oil field services industry, Halliburton has a long history of relying heavily on data to understand current and past operating conditions in the field and to measure in-well performance.
Yet the sheer volume of information that we can and do collect goes way beyond human cognitive bandwidth. Advances in sensor science are delivering enormous troves of both dark data and what I think of as really dark data. For example, we scan rocks electromagnetically to determine their consistency. We use nuclear magnetic resonance to perform what amounts to an MRI on oil wells. Neutron and gamma-ray analysis measures the electrical permittivity and conductivity of rock. Downhole spectroscopy measures fluids. Acoustic sensors collect 1–2 terabytes of data daily. All of this dark data helps us better understand in-well performance. In fact, there’s so much potential value buried in this darkness that I flip the frame and refer to it as “bright data” that we have yet to tap.
We’ve done a good job of building a retrospective view of past performance. In the next phase of Halliburton’s ongoing analytics program, we want to develop the capacity to capture, mine, and use bright data insights to become more predictive. Given the nature of our operations, this will be no small task. Identical events driven by common circumstances are rare in the oil and gas industry. We have 30 years of retrospective data, but there are an infinite number of combinations of rock, gas, oil, and other variables that affect outcomes. Unfortunately, there is no overarching constituent physics equation that can describe the right action to take for any situation encountered. Yet, even if we can’t explain what we’ve seen historically, we can explorewhat has happened and let our refined appreciation of historic data serve as a road map to where we can go. In other words, we plan to correlate data to things that statistically seem to matter and, then, use this data to develop a confidence threshold to inform how we should approach these issues.
We believe that nontraditional data holds the key to creating advanced intelligent response capabilities to solve problems, potentially without human intervention, before they happen. However, the oil and gas industry is justifiably conservative when it comes to adopting new technology, and when it comes to automating our handling of critical infrastructure, the industry will likely be more conservative than usual. That’s why we may see a tiered approach emerge for leveraging new product lines, tools, and offerings. At the lowest level, we’ll take measurements and tell someone after the fact that something happened. At the next level, our goal will be to recognize that something has happened and, then, understand why it happened. The following step will use real-time monitoring to provide in-the-moment awareness of what is taking place and why. In the next tier, predictive tools will help us discern what’s likely to happen next. The most extreme offering will involve automating the response—removing human intervention from the equation entirely.
Drilling is complicated work. To make it more autonomous and efficient, and to free humans from mundane decision making, we need to work smarter. Our industry is facing a looming generational change. Experienced employees will soon retire and take with them decades of hard-won expertise and knowledge. We can’t just tell our new hires, “Hey, go read 300 terabytes of dark data to get up to speed.” We’re going to have to rely on new approaches for developing, managing, and sharing data-driven wisdom.
Deploying analytics technologies to help illuminate actionable insights not only within raw data already in your possession but also in derived data represents a potentially powerful business opportunity. Yet dialing up your data mining and analysis efforts and importing large stores of nontraditional data from external sources can lead to questions about data veracity, integrity, legality, and appropriateness of use. These are questions that few organizations can afford to ignore today.
On the flip side, deep analysis of more data from a variety of sources may also yield signals that could potentially boost your cyber and risk management efforts. Indeed, the dark analytics trend is not just about deploying increasingly powerful analytics tools against untapped data sources. From a cyber risk perspective, this trend is also about wielding these and other tools to inspect both the data in your possession and third-party data you purchase.
As you explore the dark analytics trend, consider the following risk-related issues and opportunities:
Sourcing data: To what degree can you trust the data’s integrity? If you can’t confirm its accuracy, completeness, and consistency, you could be exposing your company to regulatory, financial, and even brand risks. The same goes for its authenticity. Is the source of the data who it says it is? If not, the data could be recycled from other dubious sources or, worse, stolen.
Respecting privacy: The specter of privacy law casts a long shadow over audio and video data sourced outside the enterprise. In many cases, the privacy laws that apply to audio or video clips are determined by the nationality of the individuals appearing in them. Likewise, in some countries, even recording an Internet protocol (IP) address is considered a violation of privacy. In developing a dark analytics cyber risk strategy, you should remain mindful of the vagaries of global privacy law. Finally, data that may appear to be benign could, in fact, carry potential privacy risk if it has been derived from analytics. For example, analysis of customer data might suggest a correlation between customer meal preferences and certain medical conditions or even their religion. As you begin to curate and analyze data, how do you put proper controls in place and manage the associated privacy and legal risks? What liability could you face if you archive data containing such correlative findings?
Building predictive risk models: As you apply analytics to nontraditional data sources, there may be opportunities to create predictive risk models that are based on geography, hiring practices, loans, or various factors in the marketplace. These models could potentially help companies develop a more nuanced understanding of employee, customer, or business partner sentiment that may, in turn, make it possible to develop proactive risk mitigation strategies to address each.
Spotlighting third-party risk landscape: Global companies may depend upon hundreds or even thousands of vendors to provide data and other services. By analyzing data from nontraditional sources, companies may be able to create predictive risk models that provide more detailed risk profiles of their vendors. Some of the risks identified may likely be beyond your control—a point to keep in mind as you source third-party data.
Leveraging deep web signals: The data contained in those parts of the web that are currently accessible to search engines is too vast for organizations to harness. Expanding that already-infinite universe to include nontraditional data from the deep web may present data analysis opportunities, but from a cyber risk standpoint, it may also present considerable risks. The dark web represents only one small component part of the larger deep web, but it has, time and again, been at the root of cyber challenges and issues. As such, it will likely amplify the risk challenges and complexities that companies face should they choose to explore it. Proceed with eyes wide open.
That said, by applying risk modeling to bounded data sets sourced from the larger deep web, organizations may be able to further expand their knowledge in the realms of cybersecurity, competitive intelligence, customer engagement, and other areas of strategic priority.
Where do you start?
Three years from now, your organization may find itself overwhelmed by immeasurable volumes of unstructured data generated by Internet of Things devices. Working today to develop the discipline and tools you will need to manage and mine all of this dark data can help your organization generate data-driven insights today while preparing for even greater opportunities in the years ahead.
This process begins with a series of practical steps:
- Ask the right questions: Rather than attempting to discover and inventory all of the dark data hidden within and outside your organization, work with business teams to identify specific questions they want answered. Work to identify potential dark analytics sources and the untapped opportunities contained therein. Then focus your analytics efforts on those data streams and sources that are particularly relevant. For example, if marketing wants to boost sales of sports equipment in a certain region, analytics teams can focus their efforts on real-time sales transaction streams, inventory, and product pricing data at select stores within the target region. They could then supplement this data with historic unstructured data—in-store video analysis of customer foot traffic, social sentiment, influencer behavior, or even pictures of displays or product placement across sites—to generate more nuanced insights.
- Look outside your organization: You can augment your own data with publicly available demographic, location, and statistical information. Not only can this help your analytics teams generate more expansive, detailed reports—it can put insights in a more useful context. For example, a physician makes recommendations to an asthma patient based on her known health history and a current examination. By reviewing local weather data, he can also provide short-term solutions to help her through a flare-up during pollen season. In another example, employers might analyze data from geospatial tools, traffic patterns, and employee turnover to determine the extent to which employee job satisfaction levels are being adversely impacted by commute times.
- Augment data talent: Data scientists are an increasingly valuable resource, especially those who can artfully combine deep modeling and statistical techniques with industry or function-specific insights and creative problem framing. Going forward, those with demonstrable expertise in a few areas will likely be in demand. For example, both machine learning and deep learning require programmatic expertise—the ability to build established patterns to determine the appropriate combination of data corpus and method to uncover reasonable, defensible insights. Likewise, visual and graphic design skills may be increasingly critical given that visually communicating results and explaining rationales are essential for broad organizational adoption. Finally, traditional skills such as master data management and data architecture will be as valuable as ever—particularly as more companies begin laying the foundations they’ll need to meet the diverse, expansive, and exploding data needs of tomorrow.
- Explore advanced visualization tools: Not everyone in your organization will be able to digest a printout of advanced Bayesian statistics and apply them to business practices. Most people need to understand the “so what” and the “why” of complex analytical insights before they can turn insight into action. In many situations, information can be more easily digested when presented as an infographic, a dashboard, or another type of visual representation. Visual and design software packages can do more than generate eye-catching graphics such as bubble charts, word clouds, and heat maps—they can boost business intelligence by repackaging big data into smaller, more meaningful chunks, delivering value to users much faster. Additionally, the insights (and the tools) can be made accessible across the enterprise, beyond the IT department, and to business users at all levels, to create more agile, cross-functional teams.
- View it as a business-driven effort: It’s time to recognize analytics as an overall business strategy rather than as an IT function. To that end, work with C-suite colleagues to garner support for your dark analytics approach. Many CEOs are making data a cornerstone of overall business strategy, which mandates more sophisticated techniques and accountability for more deliberate handling of the underlying assets. By understanding your organization’s agenda and goals, you can determine the value that must be delivered, define the questions that should be asked, and decide how to harness available data to generate answers. Data analytics then becomes an insight-driven advantage in the marketplace. The best way to help ensure buy-in is to first pilot a project that will demonstrate the tangible ROI that can be realized by the organization with a businesswide analytics strategy.
- Think broadly: As you develop new capabilities and strategies, think about how you can extend them across the organization as well as to customers, vendors, and business partners. Your new data strategy becomes part of your reference architecture that others can use.
With ever-growing data troves still unexplored, aggregation, analysis, and storage are no longer end goals in the agile organization’s analytics strategy. Going forward, analytics efforts will focus on illuminating powerful strategic, customer, and operational insights hidden within untraditional and dark data sources. Be excited about the potential of unstructured and external data, but stay grounded in specific business questions with bounded scope and measurable, attributable value. Use these questions to focus your dark analytics efforts on areas that matter to your business—and to avoid getting lost in the increasingly vast unknown.