CFO World

What's driving the rise of text analytics and its role in CX

We look at how brands are increasingly tapping into text analytics to improve customer experience delivery and understand their market better

When Norman Peledeau first began developing text analytics software 20 years ago, he could barely get a meeting with potential commercial clients. Now he finds himself at the helm of a company developing one of the essential tools of the social media age.

Peledeau had been tasked with analysing a large volume of text. But when he couldn’t find any simple tools to assist, he set about developing his own.

“I was doing that analysis by hand, and I thought there must be something better than this,” Peledeau tells CMO. “At the time, very few companies were interested in any kind of text analysis, but there were a lot of academic researchers in cultural science and communication doing that.”

Now his company, Montreal-based Provalis Research, has created text analytics software products used by more than 4000 organisations across 80 countries for tasks including media and survey analysis and market research.

The market Provalis plays in is now predicted to at a compound growth rate of 17 per cent until 2023, when it will be worth US$23 billion. And much larger players are incorporating text analytics into their offerings, including SAP, IBM, SAS Institute and OpenText Corporation, as well as a host of smaller established players and startups.

What’s driving take-up

According to Peledeau, one use case accounts for a large swathe of that growth: Social media monitoring.

“This is what has driven a lot of market research companies to look for text analytics tools,” Peledeau says. “When I started and was presenting our tool, people were looking at it and not seeing the point of wanting to analyse text data. And when they were doing surveys, they were not putting open-ended questions in because they know it was time consuming to analyse. Things have changed.”

Peledeau says analytics is now also commonly applied to transcripts of focus groups and contact centre interactions, with some brands now specifically encouraging clients to provide written commentary through blogs, comment pages and online communities.

The recent rapid uptake of text analytics has also been witnessed by SAS Institute.

“Organisations that we speak to now are asking us more and more about text,” says SAS director for presales, fraud and cloud for A/NZ, Dominic Frost. “I would say on just about every RFP response we have now there is now a section on text.”

Such is the demand for text analytics that for experience management specialist, Qualtrics, it is a core requirement alongside its traditional survey tools.

“We wound up with the responsibility for not just helping clients collect the data, but making sense of it,” Qualtrics product manager, Jamie Morningstar, says. “So how do you take the data that has been collected and make it really actionable and really powerful? And that is where a lot of our analytics products come in.

“Text analytics is all about making the text actionable… what text analytics delivers to those clients is the ability to quantify that text data so it can be really actionable and analysable, right alongside the quantitative data they are collecting.”

Morningstar says while the most common usage is to determine how an organisation is performing on the topics it cares about, increasingly analytics being used to find those topics they didn’t care about, but should. These might include emerging or latent issues that pop up too infrequently to be otherwise noticed.

She describes one example at a home Internet installer. “There was a small but powerful underlying theme when installers were smelling like cigarettes,” Morningstar says. “That was a really big driver of dissatisfaction for that very small number of clients reporting that. It’s not common, but it is something that is pretty easily affected by those installers if they know that it matters.”

Another example came about at a website hosting company, where ‘site’ was a commonly used term.

“But then another term that came up was ‘sight’,” Morningstar says. “Even though ‘sight’ and ‘site’ are not synonymous, these customers were mixing the words up. If our machine learning tools hadn’t surfaced this a whole body of related concepts would have never been seen.”

Up next: How text analytics is improving the way brands gauge customer sentiment, plus combining data sources

Page Break


Sentimental analysis

One of the common applications for text analytics is for analysing customer sentiment, through determining the adjectives and other descriptive terms that consumers use when discussing brands and products.

Much of this capability was pioneered by Massachusetts-based company, Lexalytics. CEO, Jeff Catlin, believes his firm was the first to offer a sentiment analytics capability when it launched in 2004, thanks to its focus on analysing natural language.

“The first market that was a strong pick up was marketing, for what has become known as social listening, and particularly in sentiment analysis,” Catlin says. “Historically, we were a company that provided a lot of the backend technology to the social listening providers – the Sprinklrs and the HootSuites and all those guys.”

While the technology is constantly improving, Catlin and others readily concede there are many tasks where it can’t do the job as well as a human.

“I’m not sure we are ever going to be all the way there – it is a really hard problem,” he says. “And you know how hard the problem is when humans can’t do it.

“The way you do actually measure the accuracy of sentiment is what’s called inter-rater agreement, which means you have humans read something and you ask them if it is good or bad. If you can get them to agree nine out of 10 times, you are doing great. But the fact is well trained humans can’t agree more than nine out of 10 times.”

Where the machines do shine, however, is in the volume of words they can analyse – something no commercially-viable team of people can match. And on the accuracy front, companies such as Lexalytics have developed domain-specific variations for industries such as airlines, hotels and pharmaceuticals.

“That gets accuracy up within three to five points of humans,” Catlin says, adding that the stage of development will be around emotional measurement.

“In contact centre conversations, ’disappointed’ is not nearly the same thing as ‘angry’. But if they can figure out that you are angry, they can route the calls differently, because the likelihood of a churn is very high. That is more than sentiment, because you can measure impact from it.”

Those capabilities have been enhanced further through the use of machine learning, to allow the software to self-tune its accuracy based on feedback.

In recent years, Catlin has seen clients increasingly use the technology as an early warning system to determine when sentiment might be turning. The rise of social media has meant that bad news can travel fast – often spreading far and wide before the organisation affected is even aware of it.

“We’ve got customers like Microsoft that are very worried about missing things particularly racial or sexist, for instance,” he says. “There are a thousand ways to be racially insensitive. The technology is getting better at synonym resolution, and the software learns more and more of them.

“You can feed it data and try and make it be racially insensitive. And that is where machine learning really shines.”

Ultimately, Catlin believes the technology will expand range of factors a marketer can monitor.

“Marketers are currently very much built around knowing things they know they want to know,” Catlin says. “They don’t do so much ‘tell me what I need to know’, or unguided discovery. That technology is getting quite a bit better.”

The power of combining data

While text analytics can reveal a treasure trove of information about what customers say, its power can be amplified further when combined with other data sources. According to SAS’s Frost, clients in the past 12 months have begun realising text only provides part of the client story.

“What we are seeing is organisations now using text as one of the assets and combining it with other information,” Frost says. “The CMO role is evolving and the CMO is now being asked to be the purveyor of all insights for customers. They are being forced more and more to have a holistic view of the customer and to provide these capabilities into the sales team.”

It is in interactions with the sales team where these capabilities of combining text and other forms of customer data might prove most useful, Frost added.

“When you rang that call centre, they already had a position on whether you are a churn risk or an upsell/cross sell opportunity, or somewhere in the middle,” he says. “They can combine that with what you are saying and what the agent is saying and look and see if the conversation is moving towards or away from a sale, or towards or away from a churn, and prompt the agents in real time.”

Check out more of CMO's coverage of improving your measurement and data-driven approach to CX:

Follow CMO on Twitter: @CMOAustralia, take part in the CMO conversation on LinkedIn: CMO ANZ, join us on Facebook: https://www.facebook.com/CMOAustralia, or check us out on Google+:google.com/+CmoAu