Your Tech Talks in 20 Podcast Subscription has been confirmed!
Please be sure to whitelist @email.genesys.com to ensure you receive Genesys emails.
September 13, 2023 – Duration 00:24:25
More than three-quarters of senior executives agree that AI will be a critical part of their customer experience (CX) operations in the coming years, finds research from Economist Impact. And 80% say they’re already seeing gains in customer and employee satisfaction ratings and loyalty by using AI improve their experiences. In this episode, Rahul Garg, VP Product, AI and Conversational AI at Genesys, explains how organizations are using conversational, generative and predictive AI to optimize the customer experience and enhance the employee experience. He also predicts the next big thing in AI for orchestrating engaging and empathetic experiences.
Want to listen more?
Subscribe to our free newsletter and get updates in your inbox on new episodes released each week!
Subscribe to our free newsletter and get new episode updates in your inbox
VP, Product, AI and Conversational AI at Genesys
Rahul joined Genesys in 2021 as VP of Product, AI and Digital Self-Service. He’s focused on building the best conversational AI products for Genesys Cloud. Prior to Genesys, he worked at Pypestream, where he served as Chief Product Officer and was on the IBM Watson team before that.
Here are conversation highlights from this episode, edited and condensed. Go to the timestamps in the recording for the full comments.
Rahul Garg (01:28):
I’m the VP of product for AI and conversational AI. My team is tasked with all our conversational tools, like bots, agent assist and knowledge products. Over the last six to 12 months, we’ve taken on a bigger role with generative AI. My team is building the next generation of bots and agent-assist tools that take advantage of generative AI.
Rahul Garg (02:25):
We look at AI in two ways. One is predictive AI, which is meant to do is drive predictions throughout the customer journey. Think about trying to predict if a user is going to engage in specific actions. We try to predict the right agent to send the user to, and then even post-call, we create predictions and analytics for the call itself — was it negative, was it positive? Was the agent helpful or not?
The one benefit of predictive AI is you don’t have to do anything. You just turn it on. The models are already created, they don’t require data to move forward. We’ve done a lot of that legwork for you.
And then there’s conversational AI, which is meant to mimic agents for menial tasks. For example, having a knowledge article that can help a user self-serve or a bot that can provide the pieces of information an agent needs.
When we talk to a lot of our customers, they spend a lot of time saying, “Hey, 30% of my calls are level-one calls,” like a password reset or a return for an e-retailer. You don’t need an agent to do that. There’s a way to do it with a bot. It guides you through that process.
Conversational models are very configurable; they’re built to speak the language you speak, understand the language your users speak. They require more work, but there’s a bigger return on investment because you can deflect calls and help your agents in real time.
Rahul Garg (04:40):
No, it’s definitely not just bots. Bots are a starting point, but it’s really about augmenting humans. Bots are only as good as what you create. What we see is people will create bots and add knowledge to them to make them better. They’ll add integrations in the backend to make the bots do more than just giving answers. For instance, AI could recommend pinging a customer proactively because she’s in a recall group. And then a bot could have that entire conversation to process the recall without having a human involved.
Rahul Garg (05:42):
It’s driving customer satisfaction. If you look at conversations you have with your friends, you don’t really spend a lot of time talking to them about things you expect them to understand and know some of the context. With AI, we can do that in real time. If I know you as a customer, know you bought 10 things from me, I can have a deeper conversation with you.
Many people hate the fact that Amazon knows so much about them because it’s listening in at their house. But I love it. It makes my life so much easier. I forget what I was going there to search for, and then it tells me, “Hey, this is what you’re talking about.”
If you could help drive those enriched type of personalized experiences without a lot of lift, it’s definitely going to drive customer satisfaction — and we’re seeing major gains on the agent side too.
We’re offering agents things like call summarization right now. If you look at agent calls, 40% of the time is spent at the end of the call to wrap it up. They summarize the call, they’ll write all the notes, they’ll update their CRM records. Why can’t we automate that to make their lives easier? With auto-summarization, agents are doing 40% more work on calls with customers versus spending all that time on back-office work.
Rahul Garg (07:31):
I’d say, today, it takes most of our customers about three to six weeks to build a bot. We give them tools to make it easy. They have a lot of data already. If you are a contact center, you have treasure trove of conversations with your consumers. We have tools that can uncover the top 10 things your consumers are talking about. Next is to train our Intent Miner tool with that data. Once those models are trained, you can set the process flow for each intent and then build them into the system.
But the lift isn’t building AI. It’s building the integrations on the backend.
Generative AI is flipping that paradigm even further. We’re working through some concepts of building bots from previous flows and calls you’ve had. You tell it, “I want to automate this queue with a bot” and we can take previous data from that queue, understand the intents — because 50 calls were for password reset, 25 calls for updating contact information — and then you have a bot ready to go.
The way software is developing, and AI is developing, it’s making it easier for the common person to use it. We don’t tell any of our clients that they need data scientists to go build this stuff. We’ve made it so that most business analysts can say, “This is the business problem I’m trying to solve. This is how I can go build it.”
Rahul Garg (09:43):
A little bit, but I think you’ll see more designers than IT people. They’re going to be more looking at how to design the perfect conversation or perfect experience for their customers.
There’s still going to be need for IT. You still need to integrate with your backend systems, your CRMs. IT is still going to have to do those API integrations, but your IT team won’t need to build IVR flows anymore. Your conversation designer is the one now saying, “Ginger is going to call in for these five things. These are the different flows, and this is a conversation I want you to have with her.”
Rahul Garg (10:50):
Yeah, I would bet on it — because we’re using agent conversations to design conversation flows automatically. So, if they know the conversations, they can then check whether the AI built a good conversation.
I think agents are going to elevate themselves; no longer do level-one tasks. Level-two tasks are going to be more focused on harder problems to solve, as well as delivering that empathetic experience when a customer is irate or on a different emotional spectrum. There are places where you’ll always need agents to provide empathy that AI will get close to, but it won’t ever provide human empathy. So, I see agents being key to the success here.
Rahul Garg (11:59):
Some companies think they can use chat GPT on its own. Even if nine times out of 10 the answer will be right, can your business risk being wrong that one time out of 10? Probably not — especially in regulated industries. If you’re a government business, you can’t have a bot give the wrong answer.
You still want somebody in the middle to say, “This is the correct answer.” A lot of the generative AI starting points for us are going to be with a human in the middle. With agent assist, for example, the agent sees the summary before it goes to the backend system. The agent sees a knowledge article that was retrieved before it goes to the consumer.
Then there’s generative AI helping bot designers build bots, creating the training data for them — but then showing it to them to review and approve. Because some models will veer off script. Sometimes they hallucinate, and we do a lot of work to prevent that, but you still want to make sure that the technology is up to par. We put in pre- and post-processing into our API calls to make sure that that hallucinations and toxicity don’t happen. But the best way to prevent it is to have a human in the middle.
Our purpose is meant to use AI to augment your agents, augment your supervisors, augment your bot designers, so they can get their jobs done quicker, easier, faster at high quality.
Rahul Garg (14:13):
Giving the agent the answer right away or giving the customer the ability to fulfill what they’re looking for without being on hold or asking an agent is true Nirvana. Average wait times should be zero at a certain point.
When you call an airline, it shouldn’t take two hours to reschedule your flight because there’s lightning in the area. A lot of airlines are working on offering better self-service within their apps. But why can’t that be conversational? I had an issue a couple of months ago where my flight was booked for the wrong day by me not looking at the calendar properly. It took me 40 minutes to rebook. It should have been three clicks on the website saying, I want to modify my ticket to this day.
Rahul Garg (15:30):
There are two types of businesses. One is, “Can I just get rid of all my agents and use AI?” No, I don’t think that’s going to happen anytime soon. But the question then is, “Where do I get started or what do I do first?”
I tell them: Turn on predictive models. Let your data help you in real time. Turn on summarization, give your agents a superpower to finish a call one minute faster. Then the next step is to build knowledge. Even if it’s in five different places, get your knowledge repositories in order. Share that data across all your conversational areas, because once you bring in knowledge, you can use it with a bot or an agent immediately.
You could see value day one, and then your true nirvana within a year.
Now, I know a lot of folks that will say they have to slate in this type of work, so it will take more time. But if can you turn it on as a preference in the admin console, you don’t have to change anything. You’ll start seeing value and then can sell it to your boss saying, “Look, I didn’t do anything, and I got this. What if I just spent an hour doing this?” And then they see more and more value.
If you don’t start now, you’re probably going to get left behind by your competitors.
Rahul Garg (17:49):
Everyone will tell you that AI will solve everything; it will drive your ROI right away. It will in some cases. It won’t in some cases.
So, look at what you’re doing and how much work it is. Ask your current and prospective vendors what they include, whether there’s a big services cost and how long to ROI if you do the work versus them doing it.
What to think about in terms of where to get started is, “What data do I have? What data do I want to use? What data do I need to go get approval from legal to use? What data do I need to get AI up and running?”
Rahul Garg (19:48):
I think you’ll see more emphasis on using AI to augment human agents and consumers.
We talked a little about summarization, for example. We’re looking at other things we can do to help those agents, like building agent-facing virtual assistants to help them with tasks and get them through calls quicker. So, I would say agent assist is probably my biggest area.