Text Analytics News

Awareness, Opportunity & ROI
		Tom H.C. Anderson Vendors didn't deliver what they had promised, and these clients were left asking, "so what?!"... they're starting to say "oh, so that's it!"
		Craig Rhinehart There's still a lot of education and learning that needs to happen... building solutions with real results will also drive an increased awareness in the market
		Diane Burley I think awareness is far greater today than ever before
		Kurtis Williams NLP, statistical analysis, key phrase extraction, categorization, clustering, sentiment, entity extraction... That's how the inductry sells the technology. But that's not how customers want to buy it

October 21, 2011

With the first West Coast Text Analytics Summit rapidly approaching, Text Analytics News caught up with four leaders in the text analytics software field to understand the current market and how text analytics is helping companies in various industries.

Capturing%20Investment%20Opportunities%20in%20Agriculture%202011

Q. Do you feel key stake holders in your company/your clients' companies understand text analytics and its capabilities? What if anything do you think can be done to improve awareness and use?

Tom (Anderson Analytics): Do most companies understand text analytics? No, there is currently a lot of confusion, however awareness is high and the low hanging fruit is becoming more and more obvious.

I’ve spoken with several fortune 500 companies who were early adopters of text analytics. Unfortunately many were burned early on by “enterprise scale” vendors who promised them the world. These clients spent a good deal of money, but the vendors didn’t deliver what they had promised, and these clients were left asking, “so what?!”.

After six years of answering the “So what?!” question within our specific use case expertise (marketing research), our firm is starting to turn some of these around, they’re starting to say “oh, so that’s it!”

I think that right now, the group that is beginning to really see the importance of text analytics are business intelligence and ECM vendors who previously had focused on aggregating data sources from surveys to mobile data. They realize now that to stay competitive they can no longer afford to disregard the 80-85% of information locked in unstructured data, and are rushing to incorporate effective, use case specific, text analytics offerings. Those who implement first will see tremendous benefit from having a better overall product in an ever more competitive space.

Craig (IBM): Companies across multiple industries are becoming more aware of Content Analytics solutions and the value they can bring to their organization. But there’s still a lot of education and learning that needs to happen. 

The challenges are there and not going away -- 80% of information is unstructured content and continues to grow at an exponential rate. Unstructured content contains valuable insight into business operations like potential process inefficiencies or unexploited opportunities. But traditional approaches such as keyword search or business intelligence tools are unable to surface that insight.  You need smarter solutions like content analytics to unlock this value.

IBM’s smarter planet initiative was founded in 2008 to provide new ways of monitoring, connecting, and analyzing the systems allowing business, civic and nongovernmental leaders to develop more efficient ways to manage these systems. It’s been a huge success in getting the word out about the value of business analytics.

Working in other industries, understanding key business problems, and building solutions with real results will also drive an increased awareness in the market.

Kurtis (Mindshare): There is an increasing awareness of text analytics capabilities within our customer base. However there are a lot of misconceptions and a general lack of knowledge of the use cases for each type of text analytics technology. If a customer has any knowledge at all, they usually just have a general sense that text analytics technology could help them process the deluge of customer comments they have. They don't understand what the resulting data can be used for or how to couple it with other structured metrics for synergistic results.

This is almost completely caused by the text analytics industry itself. Text analytics is both very broad and very deep with a myriad of different techniques: NLP, statistical analysis, key phrase extraction, categorization, clustering, sentiment, entity extraction, just to name a few. And that's how the industry sells the technology. But that's not how customers want to buy it. I like to think of it like a customer at a car lot who tells a salesman he wants to "buy a vehicle." But what kind of vehicle? Car? Truck? Unicycle? B2 Bomber? The problem is the customer doesn't even know those options exist, let alone how each solution might be optimally used.

For our industry, rather than sell text analytics techniques we've simplified the message of text analytics down to the three real use cases customers have: exploration, discovery, and monitoring. Our clients have tons of text and just need help efficiently navigating it; they want to explore text. Our clients want to learn surprising new things about their customers they didn't know before; they want to discover things in text. Our clients want to track and be alerted when known things go wrong in the flow of comments and social posts; they want to monitor text. We find that this messaging greatly simplifies how we teach our clients about text analytics so they can use it to build loyalty in their customers.

Diane (MarkLogic): I think awareness is far greater today than ever before. The "fuzziness" and "logic" behind analytics may be a mystery to many on the business side, but without a doubt they understand the need for relevant content - if for no other reason to separate the nuggets of actionable information from all the noise.

Q. Can you recount an example of how text analytics helped you or one of your clients improve customer satisfaction and or the companies’ quality processes?

Tom (Anderson Analytics): Of course, customer satisfaction were among our first use cases, and there have been many since 2005 as we’ve worked with firms from Starwood Hotels to Unilever, more recently leveraging OdinText to understand the fulfillment as well as website satisfaction for a prominent ecommerce site as well as more traditional brick and mortar retail shops.

Surprisingly, most companies have never really “listened” to their customers before, so everything is brand new to them. Don’t get me wrong, they were using some sort of customer satisfaction tracking like Net Promoter Score, but they haven’t had the resources to do much at all with the comment data.

This isn’t unusual at most companies believe it or not. So once we make the other 85% of the information available to them we can answer all sorts of questions that would have taken expensive ad-hoc research (focus groups and surveys) to address previously. They’re also impressed how they can monitor issues in real time, something that wasn’t possible just five years ago but is now becoming standard.

Perhaps some of the most interesting work we’ve done is to leverage unstructured data to link and model customer satisfaction and ROI on various capital and process improvement initiatives. Clients love this. Everyone wants you to “show them the money”!

Craig (IBM): A premier telecommunications company demonstrates its vision for information access and analysis by using IBM Content Analytics to address the following interesting use cases:

Customer Churn Detection: They process call center notes and customer emails to detect likely candidates for customer churn It’s a rules-based text analysis engine. By proactively addressing customer churn situations, they improved customer retention rates and customer satisfaction rates.

FAQ Generation: They process call center notes and customer emails to detect candidates for reusable knowledge. By helping to automate the creation of reusable knowledge, the company is able to help customers serve themselves, reduce customer contacts on expensive customer interaction channels, and improve customer satisfaction.

Root Cause Analysis: They use the text mining capabilities of IBM Content Analytics to perform exploratory root cause analysis of customer issues.

Changes made to the business as a result of newfound insight into customers include setting initial parameters of mobile phones based on Voice of Customer, starting a new Premium Club points program, and providing a better rate in model and service upgrades for loyal customers. Another major business change as a result was the decision to open kiosks in international airports, because of an increase in the number of calls regarding provisioning for overseas numbers. These kiosks have been a major success for this company.

Diane (MarkLogic): Virtually every one of MarkLogic's clients are using some form of text analytics - either statistical pairing at the very least, to full blown semantic enrichment, to developing schemas and NLP in between. These companies do so to provide the most contextually relevant information to their clients - which greatly improves user satisfaction. Additionally, companies are using text analytics to segment content - by entity, sentiment, topics, etc - in order to slice, dice and deliver content that may reside in silos - but which now can be married together.

In addition to the "traditional" forms of text analytics - we are beginning to see temporal data and geo spatial enrichments as well. So only information that is relevant to a given area is delivered - -or information that relates to a specific time of day is served, or conversely, so that expired information is eliminated.

Kurtis (Mindshare): For a major international quick serve restaurant group we were able to leverage large-scale natural language processing to discover that a certain geographic region had a spike in customers complaining about leaky coffee cups and stains on their clothing.

Digging deeper we learned that it was a specific paper products supplier who was selling cups with defective rims. This type of information was impossible to uncover using traditional customer feedback methods. Text analytics gave us aggregate and comment level intelligence that could be correlated with structured geographic data to improve customer satisfaction in a meaningful way.

Another exiting trend we're seeing is using industry tuned text analytics for specific use cases. For example we use a highly customized legal risk text categorization rule set to identify complains that might lead to safety problems and lawsuits. When a customer complains of foreign bodies in their food, an injury during a hotel stay, or a personal threat against an employee, immediate action can be taken by our system to notify a manager who can take action before problems spiral out of control.

Finally, we were recently using text analytics to correlate dissatisfaction in call centers with transcribed voice messages from customers. We were able to use NLP to identify common phraseology around "speed," "pace," and "fast." At first I thought the comments were indicating a speed of service problem - not getting helped in time. But as we correlated more natural language phrases the problem came into focus - the customers were complaining about the pace of the agents. They were reading too fast from their scripts causing the customers to feel unappreciated and undervalued. The problem was the customers were expressing that feeling in about a hundred different ways. By using the right NLP technology we were able to group those similar comments together automatically and correctly identify the issue. We then went further and connected the unstructured findings with geographic data and identified that the problem was only occurring at a single call center within a group of call centers, which meant there was a training issue that could be solved at that location; another great example of the value of integrating structured data analysis with unstructured data analysis.

Q. Have you attempted to tie any type of ROI analysis to the use of text analytics for your firm or your clients firms? If not, how is success typically measured?

Diane (MarkLogic): In customer facing search, success can measured by looking at the empirical data - of say, looking at the bounce rate from a search page before analytics and after. Again, in nearly every case, even rudimentary levels of text analytics will a yield a decrease in bounce rate.

The adoption of semantic enrichment is allowing for companies like ALM to create tailored information products and content-driven workflow tools for specific audiences; ie. New York Litigator, New Jersey Litigator. Rollouts of new products tightly focused on a field of law within a given geographic area are completely enabled because of semantic enrichment. There would not have been any feasible way to accomplish it without analytics - so the ROI is all of the forecasted revenue associated with these new derivative products.

Kurtis (Mindshare): At its core Mindshare is a customer experience measurement company. Our currency is satisfaction and loyalty, both of which can be correlated with sales revenue, though it varies per industry. Because we collect both structured satisfaction data in the form of survey questions and unstructured data in the form of customer comments we can create powerful ROI stories that demonstrate the value of analyzing customer comments. 

We are also able to measure operational improvement measures that have direct and immediate impact on both satisfaction and profitability. For example, a major car rental agency we work with is able to use text analytics to measure vehicle maintenance issues, fleet condition, check out procedures, and staff efficiency. Increased customer satisfaction leads to increased loyalty which directly impacts repeat business and increased sales. Operational improvements also drive satisfaction and can increase profitability as well through efficiency improvements.

Taken together, these two opportunities paint an amazing ROI picture for text analytics.

Craig (IBM): Each customer applies its own ROI based on the use case or business problem being solved. Here are a couple of examples of how our customers have been extremely satisfied with the results after implementing an IBM Content Analytics Solution:

IBM Japan Business Services Co., Ltd. runs business transformation outsourcing services for internal IBM group companies as well as external clients. The company offers on-demand contact center business that handles customer inquiries by phone, email and Web, and it provides back-office support for contract management and other functions.

The organization needed to improve the quality of customer support it offers by enhancing agent training. The existing FAQ-based training program was not sufficient: agents were not prepared to handle many real-world problems. Many of the first-line agents had to pass along inquiries to more experienced personnel, and as a result, the workload of those secondary agents grew rapidly.

The organization implemented IBM Content Analytics software to collect essential information from unstructured texts, ranging from text documents to customer voices, and then categorize and analyze information. Managers use the software to examine complex, challenging and long customer service interactions that primary agents could not resolve on their own.

Primary agents can handle more inquiries without having to transfer calls to secondary personnel. In the month prior to training changes, 202 inquiries were transferred to secondary agents. Six months after changes were made, only 16 were transferred. Before changes, primary agents could address fewer than 40 percent of inquiries without help. Six months later, they could handle more than 60 percent.

* 92 percent reduction in number of issues needed to be handed over to secondary agents
* 88 percent improvement in rate of calls handles by primary agent
* 88 percent decrease in number of calls on product bugs

Another example is specifically in healthcare.  One of the largest nonprofit health care organizations in the United States in partnership with a renowned research Medical University had a treasure trove of historical information trapped in unstructured clinical notes.

This healthcare organization and university researchers are now able to analyze unstructured information to answer key questions that were previously unavailable. Questions like: Does the patient smoke? How often and for how long? If smoke free, how long? What home medications is the patient taking? What is the patient sent home with? What was the diagnosis and what procedures performed on patient?  Preliminary results recorded have been pretty amazing. Using ICA, they have been able to

* Accurately and automatically extracting patient social history with Precision/Recall of 90%/93%
* Reduced the time to review medical records by 10-fold
* Accelerate the pace of research projects.

Tom (Anderson Analytics): Yes, as I mentioned earlier, and this really is super important. There are so many ways to look at ROI. Unfortunately often the first type of ROI question we’ll get is “how much does text analytics cost compared to human coding?”.

While depending on the scale, there are certainly savings by using an automated process, we have to explain that coding is just one small part of text analytics (at least as we see it). Furthermore comparing human coding to text analytics isn’t really logical. Human codes are far more limited in terms of data mining. All they give you is a high level frequency count.

So to answer your question, we’ve been able to measure ROI beyond cost vs. human coding on several levels, from evaluating the increase in business due to capital expenditure and new product introductions to various campaign effectiveness measures.

I do think ROI will become even more important in the future, and we should all be thinking about how to do a better job measuring it.

TAKE ME BACK TO THE HOME PAGE

Developments, Insights and Innovation
Daniel Graham	Tom H.C. Anderson	Catherine van Zuylen

September 28, 2011

Text Analytics News caught up with three professionals at companies that are pioneering text analytics and big data to ask them some questions about their personal experiences in the analytics industry and the future outlook of the text analytics market.

Q. When and how did you first get into text analytics, and what was it that attracted you to this technology in the first place?

Anderson: From 2001 through 2005 I worked on Starwood Hotels Customer Satisfaction Program, and during that time we had over a million VOC Surveys coming in per year. It was very different from other types of research I had conducted in the past, and while we were using very advanced analysis on quantitative data points I realized that the text comments could add so much more explanatory power to the analysis.

In graduate school my main area of interest was in gaining consumer insights through data mining, and though text mining (NLP) was still relatively new at least in terms of adoption, I started exploring the various software already on the market and found we could more easily answer and solve various business questions by leveraging the unstructured data available to us. So in 2005 I started Anderson Analytics and during the past six years we’ve leveraged these new techniques in a number of ways I hadn’t even thought of at the time.

The rise of social media during this time certainly helped propel this field and offer even greater opportunities. Expanding beyond survey research to working with data from Bulletin boards, LinkedIn, Facebook and Twitter has certainly kept it exciting.

Graham: Around 2000, I was at IBM working with Watson Labs on text analysis projects. At the time, we were working with a stock brokerage in Manhattan with their call center. Consumers would call in and tell the contact center person dozens of personal things, some of which could be used to improve service and also sell more to the consumer. I learned a great deal about ontologies, text processing, and text analytics. At about the same time, a few Professional Services people were able to help Ford Motor use a data warehouse to solve the Firestone Tire blow out problem affecting their SUVs and trucks. There was lot of text processing and data warehouse analytics needed for that.

Being in analytics for most of my career, the use of text analysis was fascinating. There was huge potential in many applications. I think the complexity of the text problems were also attractive – these were tough computer science problems which just makes it all the more fun when you solve them. I have always been fascinated with why the computer can’t read. This is a start.

Zuylen: I’ve been involved in the text analytics space for nearly a decade. I was attracted to the text analytics space because it provides the ability to truly organize the world’s textual information – transforming research papers, tweets, emails, and billions of other documents into actionable insights that have the power to transform human knowledge and drive real global understanding and innovation.

Q. How does your company use text analytics?

Graham: Teradata builds data warehouses for the Global 3000 and mid-market customers. We are helping our customers add text into the data warehouse, primarily in the area of customer relationship marketing. Numerous customers have already been doing text processing and feed that into the data warehouse for analysis. It’s simple enough to capture emails and tweets in the data warehouse.

Deriving value from these allows Marketing people to assess likes and dislikes about their products and how better to promote the products. This drove us to partnerships with Attensity, SAS, and Clarabridge and our acquisition of AsterData. AsterData opens up the social media analytics and social networking graphs so we can tie together consumer preferences with key influencers in the marketplace. Sifting through an internet full of text needs extreme scalability and parallel processing coupled with the text operators AsterData provides. Its exciting.

Anderson: Initially Anderson Analytics leveraged what was already on the market. Notably early on we were heavy users of Text Mining for Clementine. It was very much about understanding the best tool for the job. We found that no one vendor had the best solution for every situation.

There are more social media monitoring companies now than I can keep track of. Eventually, after trying so many different tools on so many projects you begin to develop a set of best practices. We published and won awards for some of our early white papers for leveraging text analytics in traditional market research as well as social media analysis. Finally we realized that vendors weren’t really offering something for the needs of our main use case (market research), so we started to build our own internal tools and are now in development of OdinText, which truly leverages what we have learned over the years.

Zuylen: At Attensity, we use text analytics to bring order out of chaos so companies can better connect with their customers. We take billions of social media, email, CRM records and other documents in 32 languages and both analyze those and route them to the appropriate person for action.

Q. What do you think the future of text analytics hold for us?

Zuylen: Not since the Tower of Babel have we been presented with the ability to rapidly share information on such a global scale – which I believe has the power to transform the world in such disparate efforts as cancer research, world diplomacy, customer understanding, and more. As we continue to innovate, it will become easier and easier for us to “read” more languages and structure an ever-growing amount of information so that we can more effectively learn from each other and accelerate our improvement of the global condition.

Graham: Text is one form of “big data” – it’s high volume and continuous in nature. Right now, the hot spot is in social media analytics, things like Facebook and Twitter where consumers voice opinions. There is a lot of work to do to harness this treasure trove of data and convert it into marketing decisions that grow revenues. Similarly, a lot of brand management and customer support will get immense value from a large corpus of text pulled from the internet into the data warehouse. The transition from anecdotal decision making to decisions based on trends derived from huge amounts of text can only be good for us all. So it’s easy to predict social media analytics will drive text analytics into the mainstream. This will open up dozens of other text analytic uses such as fraud detection, risk analysis, healthcare research, government trend analysis, security applications, etc. It will take another 10 years before the computer can read, so this is a fun time of innovation.

Anderson: Certainly as I said earlier, social media is an interesting and game changing technology. It’s now more important than ever to utilize text analytics as the relationship between company, employee and customer.
While we do not know yet where social media will go, and all the ways we can leverage this for competitive advantage, I think it’s equally important to not forget about many of the other sources of text data. As I support development and innovation in this field, I feel very strongly that in the near future true value will come less from incremental advances in NLP, and more from use case expertise provided by professionals from domain experts. Ours happens to be consumer insights, but there will be many more software options leveraging text analytics in various fields like pharmaceutical, finance, security etc., all providing value to the field by incorporating their unique knowledge into methods and software applications.

TAKE ME BACK TO THE HOME PAGE