Text analytics goes mobile
Technology enables companies to make critical management decisions in a fraction of the time.
By Fiona McNeill (left) and Lisa Pappas (Right)
Text analytics is being embraced across industries as diverse as law enforcement, manufacturing, health care and insurance. It is a critical to fighting crime, spotting and correcting defects, reducing fraud and improving health outcomes.
Organizations are beginning to exploit the technology’s full potential by delivering real-time analytic results on a mobile platform. What is necessary to broaden the reach of mobile text analytics?
Real-time analytics applications, on the other hand, are multiplying dramatically. Cell phone providers, for instance, can instantly determine not only the creditworthiness of a potential customer but also that person’s potential lifetime value, and suggest the offer that will make the most of the customer experience. Shoppers receive customized offers whenever they visit a website where they’ve shopped in the past. And now, retailers want to make that offer directly to those shoppers when they are in or near the store.
As analytics has revolutionized the business environment, mobile usage has grown exponentially. A recent Aberdeen report notes that unlimited data plans are driving increases in the adoption of smartphones and tablets. This adoption is increasing the productivity of mobile employees and decreasing delays in alerting employees to key issues. Firms that deploy business intelligence (BI), including browser-based applications, demonstrate significantly greater return on investment.
As the Business Software Buzz blog described the Aberdeen report, companies can make critical management decisions in one-sixth the time required for firms that do not use mobile business intelligence – 26 hours compared to 165 hours . The Aberdeen study also found that top-performing mobile BI organizations achieved a “time-to-decision” that was three times faster than all other mobile BI users.
Still, most mobile and real-time analytic applications focus on structured data – the browsing and sales history of an individual or the credit score of a purchaser. The next breakthrough in efficiency, sales and safety will come from the 80 percent of the data that does not arrive neatly broken down into zeroes and ones. Unstructured data is where much of the analytics growth is occurring, and mobile applications represent the cutting edge.
There’s a big difference between explaining what happened in the past and understanding what will happen in the future. Suppose an appliance manufacturer introduces a newly redesigned product. At first, sales are strong, but then they drop off. The structured data doesn’t provide the context for the change or explain why it happened. That information is only available in the notes made from all the customer interactions. If the analytic capabilities are solely dependent on structured data, then choices for quickly determining why sales dropped can be limited, potentially confusing and certainly slower to arrive.
To better understand warranty claims, companies have poured great effort into categorizing repair calls so that unstructured data could be entered as structured data. While this has been a useful strategy, rushed technicians don’t always enter the right “diagnosis code.” Nor do the codes convey complex issues, as many times there was more than one issue associated with the problem; alas, that information was lost. Maybe the motor is broken, but some savvier technicians figured out that a part connected to the motor is the cause – and they’ve noted that in the repair ticket commentary, not using code terms but describing the same problem. Teasing out the patterns in structured data and then manually reviewing handwritten notes could take months. The manufacturer may eventually figure out what the originating problem is, but as time passes, the damage to customer confidence may be irreparable.
While this manufacturer scrutinizes the structured data from warranty claims and manually reviews written comments, the social networks are likely lighting up with complaints. Twitter, Facebook, influential bloggers and sites that solicit customer opinion gradually and steadily provide clues to the company’s customer confidence decline. Without text analytics, you would be hard-pressed to quickly score that content for relevance across sources, classify according to sentiment, and present as easy-to-read cloud tags to key managers who can make necessary changes. Years of brand building could be scorched in a matter of days.
For the manufacturers that are using text analytics, product launches include analysis of data gathered from social networks, analyzed in conjunction with incoming comments and mined to detect patterns. Warranty claims do more than simply corroborate the patterns – they provide the clues to the complexity and underlying issues. Is it a bad part, or is it something more complex? Is there a pattern to the repeat calls? Is this one issue, or is it one issue that leads to another? Thanks to text analytics, understanding of the problem will unfold in near-real time so solutions can be delivered quickly.
A major appliance manufacturer reduced warranty claims by 50 percent using text analytics. A major car manufacturer estimates it saved hundreds of millions of dollars with the same technology. The key for both companies was using text analytics to catch the problems early so they could correct issues before additional products left the manufacturing facility. With results like that, it is little wonder that John Elder, data mining expert and founder of Elder Research, finds “in some cases text is more valuable than all the structured data combined” .
Speeding Problem Resolution
Most repair personnel carry mobile devices with them. If they are repairing a washer and learn, in talking with the owner, that the problem started with a somewhat innocuous sounding squeak a few weeks before the breakdown, that could be electronically scribed immediately back to the company to help determine if there is a broader issue. Comparing that new information to the repair archives in real time may identify a different underlying issue and that could immediately be conveyed to the repair technician.
When technicians have not experienced an issue, they can access diagrams, report notes and instructions while at the customer site. When new customer issues surface at customer sites, messages are keyed in to be analyzed. By alerting the factory floor about emerging issues, the company can stop the problem at its source on the assembly line and work to ensure availability of replacement parts. Such speed could break the kind of vicious cycle that one appliance company faced when replacement parts were not readily available to fix the washer of an influential blogger.
Text analytics is not, however, a technology just for large companies. Take the example of a small dairy in the Midwest that offers home delivery. The dairy was receiving seemingly random complaints about sour milk. Analyzing the text of complaints allowed the company to quickly discover a manufacturing flaw affecting a certain variety of milk. It now plans to take that ability to a mobile platform so that delivery drivers can be alerted to problems before they’ve had a chance to drop off a substandard product. For a company that is offering a discretionary product (home dairy delivery), spotting, stopping and reversing problems is critical in order to build the brand.
Unstructured data also provides a wealth of information for organizations engaged in fighting fraud, including insurance companies, warranty providers and governments. Traditional data mining tools can create lists of names or addresses associated with fraudulent activities (illegal claims or unpaid taxes). Text analytics help investigators understand the relationships among addresses, individual names and claims activity.
Paired with structured data, the relationship between these different data points provides investigators with “top priority” lists of individuals or businesses that need to be investigated more thoroughly, along with visual maps that help investigators see previously hidden patterns of illicit behavior. When designed in “real time,” this can stop fraud before it happens or before a claim is paid out. Fraud investigators can take photos of locations or artifacts to supplement the case record to complete the file. And when extended to a mobile environment, claims adjusters can write checks on the spot with added confidence that the claim is legitimate.
Fraud comes in many different forms, in virtually every industry, and the analysis of unstructured data plays a significant role in rooting it out:
- A large U.S. manufacturing company saved $5.1 million in its first year using text analytics to discover warranty claim fraud. Prior to using the analytical solution, the company had only been able to audit a portion of its warranty claims. With mobile, results speed up significantly. Assessments on interview reports can be analyzed on the spot to identify links to prior fraud cases.
- A Pacific Rim country increased its detection rate for illegal cargo by 20 percent, without increasing staff, by analyzing structured and unstructured data together. When the country’s imported goods doubled without a corresponding increase in inspectors, the agency searched for a more efficient way to detect illegal cargo.
The previous risk management system selected inspection subjects based on the import or export business names or items imported. But the controls were easy to evade – illegal importers changed names and altered the kinds of items they imported or the foreign providers for their items. Text analytics were very effective in highlighting patterns indicating efforts to import adulterated food or avoid import duties. An automated system instantly flags those cargo inspection candidates. With the analytic results accessible on mobile devices, inspectors would have what they need to detect illegal cargo and file reports for further analysis.
Transforming Desires Into Sales
What if you could get inside the mind of a teenage girl? One fast-fashion retailer has done it already with a mobile-savvy customer base and a website that invites its customers to design clothes, upload their choices to social networking sites, and let members of their online community review and rate the designs. After the retailer began tracking and analyzing designs by mobile location, they responded to the unmet needs and succeeded in turning over stock an average of every four weeks (versus 12 weeks for traditional retailers) without employing heavy markdowns .
Isn’t that a structured data kind of analytics? Part of it is. Knowing that teens near a mall in Northern Virginia are particularly fond of lime-green tie-dyed tank tops does involve mining structured data. But understanding product issues that those teens commented on – such as that the shirts shrunk, colors faded or the seams frayed – involves text analytics. So does monitoring and analyzing the sentiment of the “yucks” or “OMGs” when designs are shared across networks like Facebook. Those empowered by text analysis and mobile access are equipped to pull faulty products off the shelves pronto and compel buyers to visit the store so they can be among the first to get the latest must-have top!
Complaints can serve as a kind of early-warning system in highly regulated industries like banking and pharmaceuticals. Drug recalls have often been preceded by years of complaints from individuals to regulatory agencies like the FDA. It isn’t always easy to spot critical trends in unstructured data. Text analytics can automate the analysis of complaints and serve as an even earlier-warning system.
After the mortgage meltdown, a regulatory agency with oversight responsibilities in the financial industry realized that warning signs were present in the unstructured complaint data that had flowed into the agency years before the real estate bubble actually burst. That agency has since invested in text analytics in hopes of stemming the next financial calamity.
Mining text from social networks could also provide “early-warning systems” for the manufacturers of medical devices – and those who regulate them. Failing hip replacements, for example, will show up increasingly in casual networking channels – along with more formal channels such as complaints filed with a regulatory body.
The insights from text analytics and the speed that mobile devices allow could have reduced a good share of the losses such organizations have faced.
Combining GPS and Text
Like the appliance technician, other industries equip field workers with laptops that transmit updates to the central operations. A U.S. cable TV and Internet service provider equips vehicles with GPS tracking in addition to laptops with Wi-Fi access. When completing a service call, technicians type details into the online ticket system. Text entered is analyzed with recorded geolocation data. This real-time combination has at times enabled the company, based on clusters of related complaints, to identify looming system faults related to localized climate conditions and to proactively perform maintenance.
Regional or system-wide issues are communicated via alert to update call center representatives, who can then assure customers calling in that the problem is being addressed.
Data on court proceedings, warrants, probation and parole history prisons, fusion centers, DMV and other relevant sources can be brought together and analyzed to give an unparalleled view of offenders and potential threats. Whether it is an officer approaching a car, a judge issuing a sentence or an analyst evaluating a potential threat, professionals at all levels of criminal justice and law enforcement can make more efficient, better-informed decisions when they can access the insights surfaced through text analytics.
According to “Vision 2015,” a report created by the U.S. Office of the Director of National Intelligence, “Information overload already presents a profound challenge to our business model. Given these challenges, the analytic community has no choice but to pursue major breakthroughs in capability.”
Imagine the effectiveness of intelligence agencies using analytics to detect patterns in the social networking activities of radical extremists. For example:
- Consider police forces that analyze not just the structured aspects of criminal activity (where and what crime occurred) but the unstructured aspects, such as how a home was broken into, and details regarding what was stolen and what was left behind.
- What about national crime-fighting agencies that collect the details of seemingly random acts of violence across jurisdictions to spot patterns that could lead to an arrest? Such details typically reside in the case file text and can be accessed and analyzed through the laptops of field police agencies.
The future is here. Statewide organizations have already proved that combing information sources improves proactive policing results. Internationally, it’s already helping in the war on terror as government agencies analyze the structure, character, interactions and methods associated with terror operational networks. That intelligence is made more widely available when well-established business intelligence tools push the analysis to field-level, secure mobile devices.
Revolutionizing Medical Care
Researchers and physicians are using analytics to improve patient outcomes, reduce hospital stays, better coordinate care and study the most useful and cost-effective ways to prevent injury and illness. Traditionally, researchers have worked with diagnosis-related code groups (DRGs) that assign a number to a diagnosis. Researchers have long known that DRGs have limits. No two cases of pneumonia are quite the same. It is sometimes difficult to understand severity using DRGs to factor in extenuating circumstances. Sometimes secondary DRGs are not adequately recorded (an infection following a surgery). However, manually checking written medical records is time-consuming.
A U.S. research hospital used text analytics to search physicians’ records to better understand the factors involved in prolonged hospital stays. It discovered that some prescription medicines and medications in combination were more likely to lead to long hospital stays than other drug alternatives. This is not the type of analysis that could easily be picked up by solely analyzing structured data, because the pattern identified was in side effects from these prescribed medications – text information, in this case, found in the medical file notations.
A hospital in Denmark runs all of its dictated medical records through a text analytics solution to classify DRGs and help identify complications, such as pneumonias, that occur after a surgery. The hospital’s chief surgeon considers this approach critical to being able to research the cause of secondary complications from hospital stays and ultimately reduce those causes. Extending this to a mobile environment including dashboards and alerts could further help physicians make informed changes to patient care even more quickly.
Text analytics can also help prevent workplace injuries. A European insurance company that administers workers’ compensation insurance for its national government combines structured and unstructured data to not only increase staff efficiency, but to help prevent workplace accidents. It discovered that certain types of injuries were more highly correlated with specific occupations and that within occupational groups, the activities leading to injury varied among different types of workers – to the point that proactive changes in procedures and additional training decreased time off from work due to injury. For this insurer, using the text data not only improved predictive models but also provided the necessary details to know what needed to be changed as part of their business process.
Taking Intelligence Mobile
With data volumes ever increasing and the court of public opinion gone social, information-centric organizations can be assured that they have a rich source of intelligence trapped within their text data. Building intelligence from all the available data – and taking it to a mobile environment when processes are time-sensitive – leads to more efficient operations, centralized knowledge and swift response to issues and opportunities.
Adaptability boosts bottom-line results. By analyzing both structured and unstructured data, you’ll better understand what needs to change and what can stay the same. By having analytic results available wherever you need, you’ll be more than adaptable. You’ll lead the pack.
Fiona McNeill, global product marketing manager for SAS, oversees product marketing for SAS Text Analytics. Lisa Pappas, global product marketing manager for SAS, focuses on business intelligence and data visualization.
- Business Software Buzz blog, “Make Critical Management Decisions Faster with Mobile Business Intelligence,” Jan. 13, 2011, http://www.business-software.com/crm/crmbuzz/make-critical-management-decisions-faster-with-mobile-business-intelligence/
- Jim Dion, “What Gets Measured Gets Done – Metrics and their Impact on Your Business – Part 2,” Running Insight, http://www.runninginsight.com/articles.php?id=31
- 35Following the 2016 presidential election, the first two months of the new year were quite tumultuous for the U.S. healthcare system. Interestingly, the Affordable Care Act (ACA) has survived so far, albeit the process to repeal the law has begun.
- 34The 2016 election is a watershed moment for the U.S. healthcare industry. Any presidential election and change of guards come with changes in policies. It happened in 2008 when President Obama was sworn into the office. That led to the establishment of the Affordable Care Act (ACA) or Obamacare. To…
- 31As healthcare organizations transition to value-based care, there is an increasing need for actionable information. Many organizations do not know where to start in building an information framework that assists with decision-making and drives actions. Provider organizations, particularly large, complicated health systems, have incredible amounts of data spread over several…