Everyone has had at least one bad experience when dialing into a call center. The robotic audio recording, the limited menu options, the repetitive elevator music in the background, and the general feeling of time wasted are all too familiar. As customers try to get answers, many times they find themselves falling into the infamous spiral of misery, searching desperately to speak to a live agent.

While virtual assistants, mobile applications and digital web interfaces have made self-service options in customer service more accessible to users, phone dial-in experiences overall continue to fall short in meeting expectations. In fact, 98% of customers would rather talk to a live person than deal with a recording when they call customer service.

Over the last decade, the adoption of voice recognition technology has transformed how consumers interact with brands. There are interactive voice assistants that can create personalized music playlists at home and AI assistants that can quickly take your orders at a drive-thru. With the rise of generative AI, speech recognition, deep learning and natural language technology, businesses are now taking huge strides that promise a new era for call center operations.

At IBM, we have been working with clients across various industries to help them successfully modernize their call center operations by improving deflection rates and customer self-service with powerful AI voice assistants. With the help of IBM Consulting® and the IBM watsonx™ Assistant conversational AI solution, the State of New Jersey now handles an average of 6,200 calls and 800 call center hours a month through its AI voice assistant. The AI voice assistant helps these citizens get answers and support for aid over the phone.

IBM continues to work with clients globally to deliver tailored solutions that can take their customer service to the next level. Today, we are excited to introduce a new set of generative AI features that are poised to deliver enhanced call center experiences over the phone with AI voice assistants that are intuitive, expressive and sound eerily similar to a human agent. These features provide seamless support to live agents with insights and automated, optimized workflows.

Introducing IBM watsonx large speech models now available to users

With the rise of ChatGPT in the last year, we have seen generative AI take off for chat-based experiences in a big way, using large language models (LLMs) that span an insurmountable number of topics. But what about voice? Businesses are now adopting a new form of technology designed to enhance speech recognition for AI voice assistants: Large speech models or LSMs. An LSM is an advanced artificial intelligence model designed to understand, process and generate human speech with high accuracy.

Based on transformer technology, LSMs take vast amounts of training data and model parameters to deliver accuracy in speech recognition. IBM research and development teams have been hard at work designing new, state-of-the-art LSMs, purpose-built for enterprise use to support various customer support use cases such as self-service phone assistants and real-time call transcription.

Starting today, watsonx Assistant users with the Telephony/Voice add-on subscription can use these LSMs in English right off their instance to enhance speech recognition in phone channels. This means businesses can say goodbye to automated recording experiences and introduce powerful AI assistants that can understand human language much like a live agent would. In addition to English, we have new languages available in beta today: Japanese and French, and plan to expand to Spanish and Portuguese later this year.

So, what is the difference between these new LSMs and traditional speech models?

These large speech models are IBM’s most competitive and accurate yet. Based on internal benchmarking, this new LSM is already outperforming OpenAI’s Whisper model on short-form English use cases. The out-of-the-box performance of this English LSM word error rate (WER) was 42% lower than that of the Whisper model across 5 use cases for customer service and we expect this model to improve its performance over time.

Introducing new streaming capabilities with CCaaS provider Genesys

New streaming capabilities are also available to users of watsonx Assistant and Genesys. Clients can now set up an integration with Genesys throughvia our new streaming integration. The streaming service allows users to quickly set up an AI assistant in the Genesys architect flow and allows for a continuous exchange of data between the two services. This integration gives clients the ability to take advantage of the full breadth of what both solutions can offer with capabilities such as analytics, voice biometrics for authentication and call recording.

Some of the benefits include:

• Enabling the exchange of metadata on a turn-by-turn basis
• Promoting cloud native integration patterns
• Greater simplicity than SIP trunking
• Single web socket endpoint connected to Genesys

How does it work?

In just a few simple steps, a user can drive seamless connection to Genesys.

1. A call is made to a toll-free number
2. The call is routed to CCaaS over the public switched telephone network (PSTN)
3. CCaaS opens streaming connection with watsonx Assistant
4. Assistant signals transfer to CCaaS over Streaming API
5. CCaaS solution redirects the call to a live agent queue

This entire process can be done by using the Genesys Audio Connector. Customers can use watsonx Assistant to develop and design a conversational AI bot in the backend while keeping Genesys as their front end; allowing Genesys Cloud to call watsonx Assistant™  actions in architect call and message flows. The ability to set up an AI assistant within Genesys and exchange data between the two services will give clients the tools to deliver even more accurate responses.

Unveiling conversational search for customer self-service

Last year, IBM announced the beta release of conversational search, a new generative AI capability that brings IBM large language models and conversational AI to deliver policy-grounded answers to customers with little-to-no training needed. Today, we are making this new capability generally available to Plus and Enterprise users of watsonx Assistant and watsonx Orchestrate™ out of the box, and it is now enhanced with the latest IBM Granite™ models purpose-built for business use.

The IBM conversational search function offers a transformative approach to virtual assistance, enabling teams to build and deploy conversational experiences powered by generative AI with reduced manual effort while delivering customer self-service support on various topics within minutes. Combined with the semantic search capabilities of watsonx Discovery, conversational search orchestrates a dual-stage process, seamlessly retrieving relevant content from business documentation before synthesizing contextual responses by using IBM Granite LLMs. Such integration not only offers responses grounded on business content, but instills trust through traceable sources, thereby ensuring responsible AI usage and richer user experiences.

This functionality empowers users to accelerate the build process while increasing the scope of topics that chat interfaces can cover, facilitating diverse and sometimes complex interactions with clients.

Take your call center AI assistants to the next level with these new generative AI features that are now available. For more information about these new features, reach out to your IBM representative and schedule a custom demo today.