Transforming the customer and agent experience with conversational AI

I want to take a moment to quickly talk about what we do at Quantiphi. We’re a team of 3,500+ professionals. We started in 2013, and have seen 3x growth YoY. We've also gained some key recognitions like ISG and IDC in the past few years.

We’re an AI-first digital engineering services company. We provide applied AI solutions to our customers, set up the infrastructure, and provide data solutions to them as well.

We also have a strong partnership with Google Cloud. We’ve been working with them for the past six years, and have also managed to be the Google Cloud Partner of the Year, every year consecutively.

About five years back, we started this practice called conversational AI, primarily because Google launched Google Contact Center AI at that point, which caters to customer experience solutions using AI.

We were one of the first three launch partners that were shortlisted, primarily because of our experience with Dialogflow, even before it was Dialogflow and was called Api.ai.

Because of our experience delivering such solutions, we managed to be the launch partner for that, and have been running this practice for the past five years ever since.

I'm here to talk more about how you can reimagine your customer and agent experience and reduce operational costs using conversational AI. Let's take a look at this.

Top challenges associated with contact centers

We've been going through some challenging times in the past few years. We just got some stability post-pandemic, and now there are economic factors that are affecting our businesses: continuous global disruption, the threat of economic recession keeps looming, and our customers are expecting better experiences every day.

During these times, it’s very important for businesses to be able to sustain. And what we mean by sustain is ensuring that they’re reducing the operational cost while not compromising on the customer experience. And to be able to do so is a very delicate mix that we need to find.

Also, if you compromise on your customer experience, that's your brand. The impression of your company, be it good or bad, depends on how you actually serve your customers. That's what we want to talk about more today.

But before we actually get into that, let's just take a step back and go through some of the challenges that we face as customers.

Aren't we all eagerly waiting to press nine on a call to speak to an agent? It's crazy how we first have to go through that rudimentary part, listen to that music, and then they keep saying, “Your call is important to us.” But is it really important? You're serving all the other customers at this point, and it's taking a long time to get to speak to someone.

Getting to speak to someone means you have to go through a long process. And if your query is between Monday to Friday, 8 pm-5 pm, you’re good, but if not, you're out of luck.

Customer experiences are primarily very important, but agent experience also counts for your businesses. Your agents are overwhelmed with the surge in call volumes day in and day out, and your average handle time is increasing on a daily basis.

Looking for information and responding to the same questions over and over again exhausts the agents, and you as an organization or a business also have to go through a lot of training and onboarding costs to get these agents. But at the same time, there's a huge turnover rate that you have to keep going through.

Employees are trying to understand how we can automate the processes. Where do we start? What are the right touch points for us to get started?

To be able to do so, we also have to make data-driven decisions. How can we leverage what exists already to get actionable insights to keep going?

And from an organizational standpoint, we all want to cut the cost of our contact center, which is ever-increasing due to huge turnover rates, agents leaving, hiring, onboarding, and all of that.

So, these are some of the challenges that we're all facing.

How Quantiphi enhances contact center operations

I quickly want to introduce you to total experience, as we call it, which is a sum total of not just focusing on customer experience, but also thinking about all the people that constitute your business today. These are the employees, the agents, the affiliates, the vendors, etc.

All of this combined leads to total experience and also ensures that you’re providing a multi-channel experience, not just limiting it to one channel, but to multiple omnichannel experiences.

It's a loop; your agents and your employees are happy, so your customers are also happy. Your customer experience cannot do without the other experiences, which is why it’s a total experience.

How do we do this?

A customer calls and speaks to a virtual agent first, and then the call gets transferred. Then the live agent comes in. But even the live agent is getting assistance with AI-assisted responses. But I'm just going to break down the solution in a more graphical way.

The experience for your customer remains the same. With the input, which could be in the form of chat, voice, or email, they still reach out to your organization in the same fashion. What we do in the background is integrate with your existing contact center platform. This could be Genesis, Avaya, or whichever ones you use.

In case you don't have one or you want to move to another one, we also provide an end-to-end contact center as a service (CCaaS) platform as well, which is in the form of a Google Cloud Contact Centre AI platform.

That experience remains the same in the sense that they still reach out on the same number of the same channel. We integrate with that, and in the background, we use Google's Dialogflow, which is the conversational code that actually provides the best NLU capabilities to be able to interpret and provide a response and get back to the customer.

But how do we train these virtual agents?

You need a corpus of data to go back and analyze why people are calling. Take all your historical call recordings and your metadata from whichever CRM tool you use today.

We perform topic modeling. We have a bunch of machine learning models that we use to gain insights as to why people are calling, what their sentiment is, and how agents are responding, stuff like that.

Based on that, we’ll then create a roadmap for you and start training the virtual agent on those high-impact use cases that can actually drive containment and also lead to resolution.

This is also integrated with any back-end systems that you use. It could be Salesforce or ServiceNow. It could be Epic for healthcare or LMS for an education company.

We integrate with your system so that there are dynamic responses as well. And that's how the virtual agent is then built. The virtual agent is then capable of handling queries that are coming in, and can also resolve those queries firsthand. There's the first call resolution that happens.

But there will always be times when there is some level of human in the loop that’s needed. Let's say a virtual agent hasn’t solved a particular query or they can't handle a certain intent. That’s when we need to transfer to a live agent.

There's a seamless transfer that can happen to the live agent as well. And while that transfer is happening, we ensure that we equip the live agents with the transcription between the virtual agent and the customer. They start off from wherever the virtual agent left off, and the customer won’t have to repeat themselves again.

With the transcript, there’s also Agent Assist in the background that listens to the conversation and provides real-time recommendations. The recommendations keep popping up, and then towards the end, there's a conversation summary that can be created.

That conversation summarization is one of the key components of Agent Assist, and it can also automate the after-call work that your agents need to do with respect to going back and adding those case notes on your CRM tools.

This solution can reduce your average handle time drastically. Hypothetically, about five minutes can come down to between two and two and a half minutes, so you end up saving a lot of time on those calls. You can cater to more customers and agent productivity increases.

And all of this can have a functional layer of insights running in the background, which can just keep generating insights. It's not just insights with respect to how your contact center was performing before, but even how it’s performing now that you've implemented this technology. Are we seeing the right containment? Are we seeing the average handle time going down?

And then there's also a feedback loop mechanism. Once the virtual agent has been trained, deployed, and it's live, you’ll also get to know why people are calling and what queries aren’t getting handled, and that's when the feedback loop comes in. We keep training it over and over again, based on whatever insights we generate.

This is how we do it with Quantiphi and Google Cloud’s technology combined. We provide engineering services, virtual agents, Agent Assist, and insights. We also provide a cloud-native platform that you can use in case you want a holistic solution and an end-to-end platform that can do all of this.

And any custom NLU that’s needed for your organization to actually train it on your industry jargon, that’s also something that can happen.

So, those are some of the services we provide.

Total experience transformation success stories

Since we've been doing this for so long, we've already worked with about 125+ customers and done about 200+ implementations. We've gathered a lot of knowledge as to how the market works and what people are looking for.

Based, on this, we’ve productized some of these offerings. We’ve built out some industry-focused, pre-built virtual agent templates that can quickly help you get started on those high-impact, prominent use cases for your industry. And then you can continue to build upon those to customize it for your business.

Gatekeeper connectors are some of the productized modules that we have within Quantiphi that we use to connect it with your integration points, which basically are the back-end systems that you use today. We also have some connectors that are readily available to be able to do so.

And then we've built out the insights and analytics framework which has all the standard KPIs that you want to see to measure the performance.

While we're doing all of this, we’re making sure that we also provide advisory services throughout, ensuring that when a customer wants to get started, and when they want to make an assessment of where they stand in terms of customer experience maturity, we help them understand the stage from a maturity assessment standpoint.

We also help them understand how you can get started on this transformation roadmap with ROI analysis. We also perform that.

And once the technology has been deployed and it’s being used by your agents, one very important aspect is ensuring that the employees are also getting a hang of the technology because you suddenly introduced something, and things could just go haywire.

How do they adapt to that? How do we go through an organizational change management process which leads to the right steps and processes that we need to take so that the adoption of the technology is being maximized to its best potential?

That's a bit about how we do it, but for your organization, if you want to get started from a total experience transformation standpoint, what we recommend you do first is select those KPIs that actually matter to your business. What’s the business value that you want to drive? You start off with that first. It could be from a customer experience standpoint, reducing operational costs, or agent productivity.

You select those KPIs first that you want to solve. You even put a measure to it. Currently, let's say my IVR containment rate is about 10%. How do I get it to 40% in the next four months? So, that's how you set the business problem first.

And then we help you get started with the key aspects that you need to consider to be able to start doing this. Which channel do you want to consider? Is it chat? Is it voice? Is it email? What industry solutions are already there in the market that can be leveraged, and what is it that you want to customize? What integration points do you need with your current systems?

For example, if I want to do a password reset or customer authentication, I have to integrate with a CRM tool. Does it have the right API endpoints?

The other one is privacy and compliance issues that we all have to think about while we're building these solutions. What’s the scale at which you want to do this? You look at your call volume and average handle time, and then you can select the department that you want to tackle.

You start first with one department or one channel, you maximize the potential of that, and then you keep growing. But starting is what’s important. And that's what we wanted to touch upon here.

Now, I also want to put this into perspective a little bit with some real-life examples of what we've done with some of our other customers. The State of Illinois is one such customer that actually made sure they made good use of the crisis. As they say, never let a crisis go to waste.

So, when the pandemic hit, there was an unprecedented surge of unemployment claims, and everybody wanted to claim their benefits. There was a huge surge in call volume, but the agents were practicing social distancing. Supply and demand became an issue, of course.

At that point, we built out a virtual agent within two weeks, and first deployed it on their website in the form of chat that was first taking in all these unemployment-related questions. Then it actually started performing claiming unemployment benefits as well. We then took it live in the next two weeks on voice.

It was an omnichannel solution, and it catered to about 3.2 million inquiries handled in the first two weeks itself. The State of Illinois estimated annual savings of 100 million dollars just through this virtual agent, and 40,000 after-hours calls were answered per night. Whenever the agents weren’t available, there were still calls that were getting handled.

That's a little bit about the State of Illinois and how we did it with them.

Another success story would be with an insurance company where they had rudimentary, IVR retirement-related queries, etc. But the step that we did with them was we actually took a very strategic and systematic approach to how we wanted to solve this problem.

We first started off with insights. We first figured out why people were calling, and how their current contact center’s performing today. Then we built out the virtual agent by identifying those touch points. And that's how the virtual agent was built to address all these retirement-related issues.

We saw a 40% increase in call containment, a reduction in average handle time by three minutes, and estimated savings of about 50%.

We integrated with their Genesis system as well, and currently, we're also doing Agent Assist. After we've handled the automation aspect from a customer standpoint, we're now moving to agent productivity.

All of this is great, but our favorite, and our hot topic of this year, is about generative AI, and how generative AI is paving the way we do things for a better future, especially in the conversational AI domain and how we interact.

Baioniq: The regenerative platform for responsible and ethical AI

Late last year, when ChatGPT was launched, we were introduced and exposed to generative AI models called large language models (LLMs). Although they already existed before, they came to light because of ChatGPT. Now it's changing the way we do things and companies need to start adopting this.

But for organizations to adopt this at an enterprise level, there's a good, bad, and an ugly side to it. Let's talk about that a bit.

The good is, of course, that they’re generative. They're problem solvers, human-like, and low-effort as well. But there's a bad side to it, especially from an enterprise standpoint - the responses aren’t factually correct.

For example, it can talk about DNA computing, but DNA computing doesn’t actually exist. Because generative AI kind of hallucinates and it can generate any kind of response, it picked up DNA and computing and just created its own response which is factually incorrect. But still, it provided a response.

That’s dangerous. But what’s ugly and even more dangerous than this is when you ask questions that can actually cause harm to humans or lead to malpractices.

For example, you can ask a question about how to make a Molotov cocktail, which is dangerous, but it still gives a response.

You can't control how these LLMs are behaving or what kind of responses they're generating, and that actually brings us to just summarise the drawbacks that we face with LLMs at an enterprise level.

From a customer standpoint, it’s fine. All of us are using ChatGPT, we get great responses, everything's great. We decide what we take.

But from an enterprise standpoint, you don't want to take the risk of customers asking anything and there are any kinds of responses that are getting generated.

It's not trained on your proprietary data. It’s built on all of this data that you don't control, especially the ones that are publicly available. And there’s a lack of control over your output. It can hallucinate and generate any kind of response. You never know what it gives.

The task-specific training needs are another element that’s very important for enterprises. It needs to have domain knowledge, it needs to have a sense of why people are reaching out, and it should be able to have the business logic that’s integrated with your current business.

And the other is, of course, data privacy and security measures that all enterprises have to take into account.

With these challenges, although we want to leverage generative AI and large language models in our enterprises, this becomes a hindrance. And what I'm going to talk about next is how we can overcome these challenges and still leverage generative AI, but the ones that are enterprise-ready.

This is Baioniq, our enterprise regenerative AI platform:

You can use any of your proprietary data sources and integrate Baioniq with it.

You can use any of your generative AI large language models that exist in the essence of it, but on top of it to ensure that it’s getting the right domain knowledge,

You can use your propriety data sources, you can integrate it with something called Compositor, and we have certain modules within Baioniq also. That's the integration layer that can integrate with any of your propriety data sources and pull in the information from there.

The other aspect is having the foundation model repository, this is where any of your large language models that exist today can come into play. Then there is some domain expertise that you need to add. That’s the section where we have calibrate that does that.

And then comes the library or the warehouse of prompts. Any kind of prompt engineering that you need to do can also be automated through that aspect.

How do we have an interface that puts all of this into perspective? And how do you actually want to use it? What's the consumption layer for your customers? It could be in the form of enterprise search, it could be chatting through text, it could be voice, or it could also just be using RPA.

That's where Blink comes into play, where we have modularity connectors that actually decide the consumption layer.

Summary

Baioniq and generative AI in general helps you increase your knowledge and workers’ productivity.

The essence of using generative AI is not to replace humans. Humans that use AI are replacing humans that don't use AI. So, it's just changing the way we do things. It's making it smarter and more economical as well. What you could do in months of building all of these solutions, now you can build it in days or even in a few minutes for that matter.

It's very important for organizations to leverage this technology to stand out in the market and make sure that all of these things that we discussed today, especially from an operating cost standpoint, we're able to use that and leverage such technology.