The Bots Are Coming, and Fast

The fact that different applications require different bots has posed a challenge. “Ordering a pizza is different than helping a customer determine their account balance,” says Amy Livingstone, vice president and general manager of contact center solutions at Nuance Communications.

Livingstone should know. The Nina voice bot that Nuance introduced in 2012 is used by both Domino’s and several leading financial institutions, including ING, USAA, and U.S. Bank.

Chatbot ecosystems geared specifically for the contact center are also taking shape. In addition to Nuance’s Nina, heavyweights making their mark in this segment include IBM’s Watson and Salesforce.com’s Einstein, but a bevy of smaller companies, including Assist, Astute Solutions, Chatfuel, Cogito, Digital Genius, Interactions, Kasisto, Msg.ai, [24]7, X.ai, and XOXCO, are also vying for attention.

And because bots need to be tied into back-end CRM systems, vendors like Aspect Software, Creative Virtual, LivePerson, Salesforce.com, Sparkcentral, Synthetix, [24]7, and Zendesk are among the countless vendors that have been enhancing their CRM solutions to work with various chatbots and digital assistants.

Bill Meisel, president of TMA Associates, noted in a recent report that the total digital assistant market landscape is made up of about 170 vendors, including the providers of full digital assistants, messaging apps like Facebook Messenger, specialized solutions, core technologies, and niche players. Of those, only a few dozen can support the building of full specialized text or voice digital assistants, he says.

And while many of the current applications emphasize text right now, Meisel suggests a greater role for speech very soon. “Adding a speech recognition front end is a relatively simple next step” once natural language processing (NLP) has been incorporated into the application, he says.

Meisel also expects the nature of the speech interface in most digital assistants and bots to change.

With voice interactions, front-end speech recognition converts the speech to text to allow natural language processing. Today, that process is carried out largely by cloud-based large-vocabulary speech recognition that is often independent of the specific application.

“In principle, since the context is limited, the speech recognition could be tuned to the application, and thus be more accurate, but that would involve specialized training of the speech recognition software,” he says.

Meisel also expects to see more integration between natural language processing and speech recognition, “so that the language that the NLP can understand will impact the probabilities in the speech recognition algorithms’ language model.”

Also look for NLP itself to change, particularly as companies experiment with neural networking, machine learning, and artificial intelligence. “It can be tolerant of speech recognition errors if so designed,” Meisel points out.

Generational Divide

Even though the market for bots is young, two generations of software have already taken shape. First-generation solutions, which were put in place in the past few years, perform simple functions and operate like IVR scripts. These include the basic financial service bots that let customers check their account and credit card balances, examine recent transactions, and determine when payments are due.

Autodesk, a provider of software for the architecture, engineering, construction, manufacturing, media, and entertainment industries, was at the forefront of this wave of adoption. As vice president of operations, Gregg Spratto constantly searches for ways to improve Autodesk’s contact center, which operates with 250 company employees and another 200 outsourced individuals. It soon decided to automate routine product activation queries, which number between 25,000 and 30,000 a month.

In the fall of 2015, the firm evaluated solutions from Digital Genius, IBM, and Kasisto. “Our system is complex, and IBM’s Watson enabled us to build what we needed quickly,” Spratto says.

After piloting the system for product activations, Autodesk extended the system to 40 other common requests. The changes have had a significant impact on the contact center, which handles about 1 million calls a year. Efficiency improved, but the biggest impact was on costs, which average about $50 per transaction for internal resolutions and $13 for those handled by the outsourcer. Interactions handled by the bot cost just $1.

Now a second generation of solutions is starting to take shape. These solutions promise to leverage analytics, AI, and eventually machine learning to dramatically transform the customer experience. They have the potential to provide businesses with more insight into customer desires and shape how they service and market to them. “What corporations would like to do is monitor customer interactions and have the system determine if it is appropriate to try to upsell the client,” Beccue says.

In this second wave, we’re starting to see financial services firms use bots to make their mobile sites searchable; connect customers to live support; provide help with product and account questions; or supply maps to branch locations. Flower companies can use bots as gift concierges, issuing special occasion reminders, answering customer questions, making gift suggestions, processing orders, and sending out shipping updates.

Making the Move

While bots have tremendous potential, they also present companies with many obstacles. First, any company looking at bots needs to decide which department (marketing, sales, IT, or the contact center) will drive development and which business processes could stand to be improved. “A mistake that many firms make is focusing their bot development on marketing functions, selling more products,” Aspect’s Goebel says. “They may be better suited to enhancing customer service.”