Exploring How to AI-Enable IVR for the Contact CenterExploring How to AI-Enable IVR for the Contact Center

While preparing for my upcoming Enterprise Connect 2019 session, “What It Really Takes to Put AI to Work in Your Contact Center,” I happened upon a small innovative company called Inference Solutions. I was intrigued because some well-known brands use Inference technology, yet the company is somewhat enigmatic -- unknown to me until it reached out and told me how it’s changing the way contact centers work.

Who Is Inference Solutions?

Inference offers a platform for creating intelligent virtual agents (IVAs), sometimes called chatbots or intelligent bots. The founders, who came out of Telstra Research Labs, started Inference in 2005. The company’s initial focus was on developing self-service speech applications centered around complex dialog models. These models were difficult to construct by hand, but their creation could be automated using “grammatical inference,” a technique that takes spoken conversations in any language and builds a model using speech-to-text and natural language processing (NLP). The word “Inference” in the company’s name was derived from this concept of grammatical inference.

Between 2005 and 2012, Inference provided consulting services that created these models for large telecommunications companies, organizations, and enterprises, including the Australian Post, Pizza Hut, and United Healthcare. However, even with Inference’s automation techniques, crafting these models remained difficult, expensive, and complicated.

In 2013, Inference re-evaluated its strategy and philosophy, focusing on how to make speech applications and other human interaction technology easier to use by anyone. With the emergence of software-as-a-service platforms, the company believed it could create a SaaS-based offering to abstract the technical complexity so that organizations of any size could easily use technology for automating and simplifying human interactions.

The result is a cloud-based service that allows organizations to create IVAs with the aim of offloading many mundane tasks from live agents, as a way either to augment or replace them. The company seems to be onto something given that name brands including Marriott, McDonald’s, Toshiba, Toyota, KPMG, Virgin Australia, and Kroger use its technology.

What Is an IVA?

IVAs and human contact center agents are similar in that both have skills and the ability to do things. Like live agents, virtual agents can be “hired” and “trained” (programmed) to have different skill sets based on business needs. Examples of IVA skills include the ability to understand human speech in more than 100 languages, respond using text-to-speech, detect DTMF tones, authenticate using biometric voice techniques, provide IVR, and do tone analysis. An Inference partner will typically configure several IVAs, each with a different combination of skills. For example, a basic IVA may have the skill to use touch-tones for simple interactive voice response applications while a more advanced IVA might incorporate machine learning, NLP, and speech synthesis.

Unlike with human agents, a contact center can easily scale the number of IVAs with a given skill set up or down as the business requires.

IVAs can do all sorts of tasks to care for customers. They can authenticate callers using voice, make appointments, reset passwords, look up orders, take payments, survey customers, and answer all kinds of questions -- all in an omnichannel, phone, Web, text, or SMS environment.

IVAs function like an outsourced contact center in that an organization purchases IVA services and pays to use them for as long as needed.

The key to making all these capabilities work seamlessly together and easy to use is Inference Studio, a technology abstraction platform. Like some other IVR tools, Studio provides a graphical user interface that allows customers to create tasks using the IVA’s skills with little or no programming. Because Studio has built-in interfaces to advanced third-party NLP capabilities from Google, IBM, Nuance, and others, customers can use Studio to readily build speech applications that have historically required extensive development and/or professional services work.

A simple intelligent virtual agent, created in Inference Studio, that automates the customer service tasks of collecting payments and reporting outages.

Studio has a library of prepackaged tasks that can help you accomplish your business activities using the skills in the IVA you’ve purchased. Task examples include (among many more):

IVAs can operate across different inbound and outbound channels, including text chat, SMS, Web chat, and voice. For skills that require voice interaction with callers, IVA creators can choose different voice genders in instances where it may be more effective to use a woman’s voice or a man’s voice.

For doing data dips, scheduling, processing, and bill payments, IVAs can interface with third-party apps via APIs. Inference has created APIs for a number of applications including:

In addition, users can create their own middleware to add custom data integrations.

Click below to continue to next page for an Inference proof point from partner BluIP

BluIP Proof Point

When researching a vendor company that’s new to me, I often try to speak with customers and/or resellers that are implementing the solution. One of these was Jeff Coomans, director of vertical solutions for BluIP, a BroadSoft (now part of Cisco) partner specializing in solutions for the hospitality, healthcare, and enterprise verticals. BluIP deployed an Inference IVA in the hospitality industry.

The challenge the hotel chain asked BluIP to solve centered around calls into the reservation system. Because every call answered by a live agent incurs cost, the hotel was looking for a way to eliminate the need for reservation sales agents to respond to non-sales queries yet remain available for sales-related opportunities, all while optimizing the experience for the guest and the agent.

BluIP chose to approach this challenge using an Inference IVA that supports conversational speech and NLP. This IVA would interface with guests, responding autonomously to as many non-sales queries as it could while sending sales-related queries directly to live agents. BluIP used the Inference Studio interface to design the IVA, which uses the integrated artificial intelligence/NLP engine from Google’s Dialogflow.

As with many projects, there are complicating factors, and this project had several of them. For example, the hotel chain outsources its contact center operations to multiple technology providers. Furthermore, it has different contact centers for each brand in the hotel’s portfolio. In order for this solution to become operational, all the players -- the carriers, the outsourced contact centers, BluIP, and the hotel’s contact center specialists -- had to work together.

To equip the IVA with the information needed to respond to non-sales queries, the hotel’s property data, which was stored in a data warehouse, needed to be made accessible to the IVA. This data warehouse was designed for service queries from the hotel’s internal systems, but it hadn’t been built or structured to handle external queries. So, the hotel had to make some adjustments on the data side so that the cloud-based IVA could fetch the right information once the intent of a caller’s query was discovered via the NLP.

An additional constraint was that if the entire process of interacting with the IVA was too complicated or took too long, the customer would simply transfer to an agent anyway. To keep the customer engaged, BluIP programmed the IVA to speak to the customer using natural language, and through a sequence of prompts and example phrases, help the customer rapidly identify the right hotel property and the specific information about the property in which the customer was interested.

Following a proof of concept, the hotel chain has funded the project and is rolling out the IVAs this year.

As the BluIP project leader, Coomans offered up a few lessons learned from this experience:

Coomans’ experience has led him to believe that virtual agents are the way of the future. He commented, “It’s hard to beat a well-designed virtual agent that can interact with people 24/7.” His belief is that virtual agents will replace or augment many of the existing contact center agents over the next five to 10 years.

What Does an IVA Cost?

Inference and BluIP were a bit coy about IVA pricing. Like live agents, IVA cost is skill set dependent. They told me the best way to think of pricing is to imagine how much a real agent with the necessary skills would cost: The IVA would then cost some fraction of the live agent’s cost. For example, if a human agent cost $28,000 annually and the IVA cost $4,800 per year, there would be some significant savings if a virtual agent was able to replace some live agents. Furthermore, by using IVAs, the live agents could become more productive because they would avoid mundane queries, giving them more time to focus on selling and upselling.

As with live agents, an IVA can only be on one call at a time, so organizations would need to use their workforce optimization software to plan accordingly so that enough IVAs were concurrently available to handle the call volume.

Inference sells virtual agents to its channel partners for a fixed price based on the skills configured within a particular IVA. The partner then adds markup as it sees fit. There is no charge for the tasks. BluIP charges the end-user customer using two pricing models: One is usage-based, and the other is transaction-based. Usage-based implies that the monthly IVA cost is computed based on how much time the IVA agent is interacting with customers. In transaction-based pricing, a single customer interaction from the start of a call or text session until that session ends is considered a single transaction.

Inference sells through approximately 30 carrier channel partners. Because of its strong relationship with BroadSoft, Cisco’s acquisition of BroadSoft allows the company to also sell through Cisco’s traditional channels.

See More at Enterprise Connect

We will have other real examples of AI in the contact center at Enterprise Connect, taking place March 18 to 21 in Orlando, Fla., in the session titled, “What It Really Takes to Put AI to Work in Your Contact Center.” In this session, we’ll discuss chatbots, next best action, and behavioral pairing. I look forward to seeing you there. (Register now, if you haven’t already, using the code NJPOSTS to save $200 off the cost of attendance.)

Disclaimer: The author has not been compensated by Inference or BluIP for writing this article. He believes this is an example of innovative AI-related contact center technology that would be of general interest to the communications market.