Designing Conversational Experiences for Call Centers

Watch The Event Again

Event Q&A

What are the solutions that organizations are using today for their IVR needs and call centers?

Sarandeep: I think a lot of organizations still are using the IVR from a long time ago. We have to truly go and press the button. But we
also, see that that is becoming cumbersome for customers, it does one size fit all, it is essentially, just if you want to direct customers to particular queues route them properly, that's what it actually does. You can't expect customer service, or self-service out of it, because people want the choice to really go and figure out where the agent is and then just go and get the agent. Right. But what we seeing is that other technologies are picking up this visual IVR, this positional IVR, that is getting a lot of traction, and companies are moving in that direction.

Karthik: I think in terms of IVRs, I think as Sarandeep pointed out, the technology itself is changing quite a bit. Right? So there have been IVRs in the past as well capable of speech, right? It's just not to the level that it is today. Without getting too technical. You know, speech-based IVRs have been around for quite a while, but they've used legacy technologies that are simply not trainable enough. And as a result weren't applicable to the diverse set of accents out there, which constantly made it a very frustrating experience. But as well as you know, if you're in a country like Canada, and now more so than in many other countries around the world, right, you have a myriad of accents you need to deal with native speaking speeds, and then you need the ability to keep training the system to keep up means exactly right. So the trainability aspect was a big one in the legacy IVRs which was very, very difficult to almost impossible in some cases. And I think the biggest window of opportunity Here is the majority of speech technologies, right, and as Sarandeep mentioned, you have the legacy IVRs, who are integrating these technologies is a bit more challenging. Then you have a new breed of 5g rs, the likes of Twilio Amazon Connect, that really are API based and make integration of these technologies very simple.

Emanuel: I'm not going to begin to understand or pretend I understand technology at that level, all I know is that we're utilizing new services for our current system. Right now we're utilizing the Genesis platform, which is pretty common right now, as well as new ones for the front end, which is, like 75% of Fortune 500. Companies are utilizing that for sort of the front-end things, but exactly what it was described there, the purpose for Genesis was an infrastructure that's very, you know, it has a lot of comprehensive sort of services and suites all in one and allows us to be, you know, fully integrable, with other systems like chatbots, and even digital now to kind of pass along information for us to really understand customer intent and things like that, and really allowing us to make scalable types of decision moving forward.

How many organizations implement natural language understanding based upon a script for their IVR call flows?

Maria: mean, it depends on the resources that every company has, and the budget, I would say him, I mean, to get intelligence data market data, you would need to like look at a report from Gartner or something like that they don't make an analysis of the market share and market data. And but as far as I know, and what I have seen, is mostly IVRs, that is structured, first, like having a natural language question, which captures the first intent. And then usually IVR, move to a directed dialogue sort of style. Just you know, at once we have established that, like, first intent, we usually use direct dialogue, just to facilitate it better. And I think maybe the others will agree with me that it is extremely difficult to create a whole call flow just with training, it's extremely difficult. Even with machine learning and neural networks, it will be extremely challenging, because there are, for example, many intents that are similar, their speech recognition problems because of poor signal because of accents. I'm sure. I'm sure people agree with me in general. I mean, it's very demanding. So I think most of them do first in natural language, and then direct dialogue.

Karthik: I agree, like behind the scenes, there is even in a natural language-based IVR, there is a certain amount of scripting going on, right? Because at the end of the day, what's important to realize is whether you're using an IVR, or a chatbot, or whatnot, it's a business process in action. And executing a business process is a sequence of steps, right? And you can't violate, like, if you give certain information where it's not intended to be, you know, you should you can trigger a certain process to act a certain way. So you still obey the business process, so to say, but having said that, though, the big opportunity with natural language IVR is non-nonlinear information access, right? So unless a scripted IVR that can just take one input at a time, right? If I say something like, hey, I want to book travel to so and so plays between so and so dates, right there, I gave you a like probably five pieces of information that in the regular IVR would have taken five steps, right. So you have the ability now to jump directly into Step six, versus, you know, going through the same sequence of asking those data points.

What do you think is one of the most popular use cases for IVR is known in call centers or is there something maybe that's, that's new that maybe wasn't around 10 years ago?

Karthik: I think in terms of use cases, the use cases haven't significantly changed. Like I mentioned, you know, the opportunity now to really get a lot more done and a lot less back and forth. Tax. Right. So typically, when you look at the fundamental purpose of what an IVR was doing, your first hope was that you could, you know, get customers and answer and potentially not engage a human. Your second hope, is that, okay? Now if you do have to engage a human that at least lands in the right place, so that, you know, you're being more efficient and not bumping the call around, I think that went a step too far. And you know, people started using dog designs and trapping people in ideas, I think we've all had that experience. You know, but looking going back to the use cases, I think, you know, some of the most popular ones, of course, are information gathering, trying to gather as much information as possible upfront, before you hand it to a human, the very fact that someone is calling you they need assistance, and the process is complex, it's, you know, there's going to be a certain amount of human touch and judgment needed, which is why you couldn't just automate it on your website or your mobile app. Right. So when someone calls in, I like looking at it in two steps. One is low-touch interactions. And then high touch interactions, low touch interactions can potentially be fully automated end to end, provided you have the digital maturity to do so for example, appointment scheduling, or checking a request. Or they can be automated without ever engaging the human. Right, and the second, the high touch ones, you know, a lot of those business processes are going to be bespoke, or you know, systems aren't integrated, and hence, you need a human, or, you know, you just need a human to stop fraud, for example, you know, you can't have a bot getting of a price match requests, right, you need a human to have that judgment, or, you know, cancellation of your service, you want to give a shot at retaining your customers, you don't want to just automate those, even if they could be automated.

Sarandeep: I would like to add that use cases also differ, but you know, you know, your, the industry you are in, you know, that are some basic use cases that you want to automate. But I think the best way to figure out the use cases is to look at your data, look at your call logs, understand what your customers are calling for you do speech analytics if you need to, really look at the data in detail, and then figure out okay, these are my, these are my cases that can like, you know, Karthik said, these are my low touchpoints, or high touch point pieces, these are the things that I can really automate using IVR. Right, and try to attack them first, rather than doing all of that together, have that strategy in place that you know, you know, that this is what I'm going to start to automating IVR, or, you know, whatever, whichever way. So use cases are a lot dependent on what your customers are calling for you and what the problems are. And that also means that you have the potential to, you know, stop those calls from coming. If If you can, you know, have your digital channels solve that problem, also. So I would say that you take a data-driven approach.

Emanuel: I will say from a bank's perspective, they're pretty like standard stuff you see across the board, like, you know, checking account balances, like people will recall 510 a day just do that, right? The client transactions, obviously, password resets, funds, transferred, activating cards, right, making debit card payments, things like that. So I would say for us, our IVR has the intention to allow customers to self serve towards all these use cases. But sometimes you've got when you dive deeper into many of these calls are centered around again, broken digital journeys, meaning something wasn't clear on digital and resulted in a call and maybe was discoverability issues and things like that. But to be honest, you like I think the call is alluded to, sometimes it actually doesn't matter if you put on digital and not people are just going to call regardless, right? We found that when it comes to our clients, money transfers are getting rejected at a point of sales, for instance, that, you know, triggers this sense of urgency, this will instinctively result in a call because they simply want to touch or sorry, talk to a human, for the human touch to reassure them that everything is okay. Right. So there's some stuff is like, how do you get around that right? So in for us, we're trying to do ways to be more proactive before the issue even takes place, like turning on alerts, for instance, right? So when you're approaching the threshold, you get a warning, and you don't spend over it. And that doesn't result in a call later on.

What are some of the challenges for building NUL based virtual agent for IVR?

Sarandeep: I think, the dialect, the accents, all of that remain a big problem. Now, again, I am in India, which is a multi-language country, there are a lot of companies that are doing work in specific regional languages. The data is, as you try to eat a lot of data to train these training models. So I think, one the training data, for building the models, one that is important, you know, that is a challenge for, for specific languages that are not English. Because you may get a lot of data for English, but for other languages, it might be a challenge, then understanding dialects in accents is actually a problem. What also is that when somebody calls you, there might be a lot of background noise. So, you know, how do you really get those things? And then, you know the training of the data model itself is a very repetitive task, right? So the data quality the bot does not understand or doesn't understand the first time and then you train it with the new accents, you know, variation utterances and all of that. So that is a cumbersome task, which does mean that you know, it's not a one-time effort, it's a journey that you have to undertake, and to continue evolving, you can't expect, you know that you've created a conversational IVR. And they are a constant one day, and secondly, you'll have all the calls being answered by them and will be 100% accurate. That's not gonna happen. What do you know, when you're trying to tell the customers? Yeah, that's a reality, you have to really think of it as a baby that, you know, you have to really evolve and grow. And that's where businesses do tend to struggle because they say that we've done such an upfront investment, and why should we have to wait to get that ROI? And, you know, those are some of the things from a business and technical perspective that I see as being challenging. 

Maria: I totally agree with Sarandeep, and on top of that, regarding, specifically the NLU part of it, not the speech recognition? I would say that, yeah, the huge problem is that the vast set of valid utterances that in that first try, and people and our clients and people that are not familiar with the industry and clients and everyone else, they are surprised when they even for a very simple service. For example, in the passport office, let's say, people, are shocked by the number of different intents that they have. And each intent has 1000s of utterances, 1000s of variations of the same thing. And everybody speaks with different words and expresses themselves differently. So it's really, really challenging. And I would say the second challenge is finding the right balance between having too many intense and risking miss recognitions and confusions between many many similar intense to too few intense, where then you are excluding a big chunk of your customers or users.

Karthik: I'll just put on you know, a different tact and give another perspective, which for me, the opportunity with NLU based IVR is a conversation like yours is really non-nonlinear information access like I was mentioning earlier, right? For example, the customer can give you a lot of pieces of information in one shot, what I found often organizations struggle with is the lack of digital maturity, right? They might not have the right set of APIs to do something with that, right? And the business process is too rigid. So now you're forced to structure a conversation. That's very unnatural, right? It's not the way someone would necessarily speak because you've designed the process, as per the infrastructure you have and not the other way. Right. And so it really has that digital maturity to tap on the opportunity that the natural language processing, conversation IVRs provide, for me is a big, big challenge that's gonna, you know, it's going to take some time to mature and address.

What are some opportunities for crafting personality in an IVR? You know, the human element to it?

Maria: I think, the more the higher the percentage of recorded utterances in the IVR means that it's going to be more natural. And, of course, I mean, there are texts to speech systems that are amazing. And they sound extremely natural. But they are mostly reserved for films or RPGs. There are companies like semantic that have amazing text to speech. But I think that if you were to deploy this kind of quality, tuned, a synthetic voice that sounds amazing and emotional. And it would ruin the call because it wasn't the bandwidth may or may not be able to take it back in on a serious note, we were using text to speech we can control with a markup language, that standard on every platform. Standardized, we can push on pace, the intonation. But I think most clients will choose a recorded voice, and we actually have a recording studio at our company just to make sure we can even record on the fly if needed. And I think, for now, we're sticking to this. Yeah. until it becomes amazing.

Sarandeep: It is important that the personality is reflective and, is part of your brand. Right? It's after all you are it's your brand that speaking so it's important that you have all the stakeholders, the corporate communication team aligned your branding team aligned when you finding this personality. And make sure that this is part of your overall corporate strategy as well. And of course, then the voices and the accents and everything can be planned accordingly. But for me, I think the most important part is to really figure out what that personality is going to be. And you really need to think about and brainstorm on that quite a bit. 

Emanuel: I think you need to start really with looking at this stage in which customers are actually entering the channel. As well as the exit paths, and goals for that journey. And to be honest, from a behavioral standpoint, we learned from our research that in many cases, customers are answering our system quite anxious or distracted multitasking and not really paying attention to the menus or options, and quite frankly, callers do not enjoy interacting with call centers. Like we've got it and heard that through some of our breakouts. In many cases, the journeys begin beginning with anxiety and expectations of frustration, being annoyed, long wait times all that stuff, right. So, not really a great way to start, right. But aside from that, like voice and tone, they're really our own, they really are the only few modalities that we can be leveraged to bring life and brands in those audio channels. So you know, we don't have visuals or interaction design, like digital would have to help guide, for instance, right? And so when you're considering stuff, like another thing to kind of consider is like real recordings versus text to speech for those locales, what do you do?  I think there are definitely pros and cons. And that screen kind of indicates that for you, either or recorded talent is more costly longer to deploy. But there's more trust associated with the right text to speech, super rapid quickly to deploy integrated multiple languages, but oftentimes not considered trustworthy rates, if not done correctly. So either way, I would recommend a more consistent approach to your voice personality to help build familiarity and Alliance with your brand. And while your voice should remain consistent, in my opinion, it's really tone, I think that may change according to the context of your messaging. So the same way you would use a different tone with telling a friend a certain situation when delivering bad news versus good news, for instance, right? So be mindful of the exit paths, and then kind of work backward. A joke of the day, for instance, might not be the most appropriate method for a declined transaction or lost and stolen card. Speaking of which, if you guys are familiar with Zappos, the online retail stores. So they utilize jokes of the day throughout their whole messages to keep things light and fun. So when crafting the message, be clear, singular, CTA, remove corporate jargon and use repetitious CTAs throughout your whole sequence. 

What are your thoughts about using NLP for data we may need from a customer? They're hard to gather via NL example someone's full name or email address.

Karthik: If the question is particularly about natural language, understanding, or conversational IVRs, the challenges with some of this data, but I've addressed it at some point, which is, it's a lot there are many technological components in action. But the most fundamental one that is pretty much made or break is the speech-to-text module. The speech or text basically is responsible for transcribing what people say into text. Right? So especially if someone provides an email, someone could say, let's say person name, one two@gmail.com. And then someone can say person name twelve@gmail.com. How do you transcribe those equally? to us? It's very intuitive, right? That someone meant one, two, and not Twelve. So it's those kinds of challenges. And a lot of those, you know, firstly, of course, it's the understanding, but secondly, it's even the transcription of those. And today, the technology isn't there. Right. So if you were to tell me that, is that possible today? The answer is no. Eventually, we will get that the right technology gets better iteratively.

Sarandeep: You could also look at different multi-channel experiences here, right? Everybody has a smartphone. So if you know if there's an interaction that you think probably difficult on voice, maybe send an SMS or a request where you can have a visual IVR, where they can fill in the information, all of those things are available and possibilities that depend on how digitally effective or transform the company is if they can do these things. But the technology is available, it's just a matter of time that these things become and are used more often.

What are some of the design considerations for building a successful handoff?

Karthik: Okay, when we talk about handoff, just to clarify so that everyone is on the same page. Handoff is really about involving a human at the right point, right. So the idea is to leverage automation as much as possible. Now, sometimes you may be able to automate the whole transaction, a great example is scheduling an appointment or telling you the status of a certain order, or a certain you are activating going back to Manny's use cases activating a credit card, these are easy to automate over voice. At some point transactions will get complex security fraud reasons, you have to involve a human. So strategies for a successful handoff, of course, is first and foremost, gather as much information as possible upfront. And I think in the breakout session, I was chatting with Catherine if I remember what she was saying. They did both an IVR that gathered all the information, and then the agent would repeat and ask the same pieces of information. That's an absolute fail, right? The idea is to take time off the human so that you know you're very efficient, you provide all the information upfront to the IVR. And then, boom, you're right up. So right without asking any other unnecessary question. The second one is knowing your customer, I think it's vital that you don't ask things that you're already supposed to know about them, you know, especially if once you authenticate them, it's unnecessary to ask them certain pieces of information beyond what's needed to identify them. Once you know what their issue is, you know who they are, hand them over. And don't ask them unnecessary questions. That another important design consideration I would say for handoffs is to degrade gracefully. And that's a very important design consideration. Because I think Sara and Maria were mentioning earlier, it's going to be impossible for a conversational IVR to know everything. So there are going to be instances where you don't know the answer, or you wouldn't be able to you've never heard that before. It's key to degrade very gracefully and immediately bring in a human and not trapped. So it's okay to ask for clarification, let's say a couple of times at most. They could repeat that I couldn't hear you. Clearly, that's fine. Maybe a couple of times, but don't go into a loop that you then trap the customer. Take them to a human ASAP, and rob them to the right. Human Rights skill sets based on what you've been able to infer so far.

So compared to designing for other voice-enabled modalities for virtual assistants or smart speakers, cars smartphones were the specificities of IVR?

Maria: I actually think that both in terms of design, and in terms of infrastructure are extremely similar. There are not many differences. We have flows both for IVRs. And for native apps or third-party apps in voice assistance. We have flows we have intense, we have slots, we have universals, we have everything. We have API, data fetching, we have everything. So it's very similar. And in fact, all the voice apps that are being built now all the design principles are based on, you know, IVR designs of the past, that's how it started the whole thing. So they're not similar. They're very similar. If there's one thing that I would say is different in IVR. It's the DTMF fallback in the design. And so if you're speaking to your Alexa, you don't have the DTMF fallback. Or if you're speaking to your car, you can't, you don't have anything to press, you can't press and, and the other thing that is a bit special when you're working with IVRs versus voice apps on assistance and platforms is that we actually get limited access to recordings. And having those recordings. Of course, they're all anonymized and loving recordings, we can train the system better to recognize better and more phrases. And the hurdles that all the voice developers from other platforms have to go through is that they don't get these recordings, they only have sometimes the transcripts. So they only have what, for example, Alexa, or Google, or Siri, what they recognize or thought the person said, and sometimes that's not very helpful or not helpful. I think that's the special thing about working with IVR.

So the question is how to select the right platform for an NLU based IVR?

Karthik: I think one of the first ones is with any technology it's useful things that matter, not cool things. And usually, it's very important to build a strong business case. And just speaking about it from experience. You need to be aware that the minutes or seconds that you save off, each customer interaction is actually worth the cost you'll invest right in, in having a conversational IVR. So needless to mention, that's a very important consideration. The second one is the trainability of the system. I think a lot has been emphasized so far about the need to constantly recognize up-and-coming intense topics customers may talk about, and then be able to train the system and train even the speech technology, right to recognize new accents. Recognize, you know, be able to be noise resilient, if I can put it that way. You know, I'm calling from a noisy background. And also remember, there's a certain amount of sampling that happens in your phones as well, when you're calling in, there's automatically a downgrade action of the voice that's happening. And it's important to be resilient to those and you know, not get thrown off. Having said that, though, what I would definitely say is the way an NLU based IVR works is one of the first things that happens is the speech or what you say gets transcribed to text. And everything from that point on is triggered off the text that you recognize. So the speech to text to speech to text technology is really the critical chain, a critical link in the chain, right? If that's not good, nothing else matters because it will misfire and probably trigger the wrong intent. The wrong intent will trigger the wrong response, leading to more customer frustration and the loop cycles down very, very quickly write, I would say a key consideration is the accuracy of the transcriptions, if that's not working, well, you know, don't even attend because it's not going to be very successful.

How do you go about choosing what metrics are?

Emanuel: So I think when considering metrics, or main goals for the IVR, and IVR, features, for that matter, one is definitely all about call deflection, or call containment, we hear we asked ourselves, what can we do to ensure callers literally stay in our systems to self serve? Right, some decisions around have centered around, you know, presenting, you know, clear menus and potentially utilizing proactive measures, meaning, you know, can we be proactive and avoid the call altogether in the future, and many of the decline transaction calls, for instance, in many cases are due to insufficient funds, right. So to get ahead of that, we try to find ways like working with our digital partners and turning on again, digital alerts, for instance, to get them to turn on those features. So in the future, don't know when they're dipping to that threshold. Another one is called migration to digital nudging digital example. What can we do on digital that could be deflected or be proactive on so think of hold message strategies, for instance, where we actually utilize that messaging, and present simple language and the benefit to self-serving through that language. And in work with, you know, we work with our digital partners to drive digital to digital properties, specific to call reasons using distinct URLs, for instance, so within the whole message will say, you know, go here, it's quicker and faster, that kind of thing in hopes that, you know, a portion of those callers on hold, will remain on the line while they self serve, or uncover how to do it on digital. And oftentimes simple instructions on digital-related to how to update your account info, for instance, is easy enough that some simple instructions can help you do this online and hopefully end that call while you're waiting. Another one is reducing agent handling time, right? So ask yourself, what can you do exactly? What can you do to shed minutes on that call itself? So, again, we utilize hold messages to also prepare the caller. So if you calling for instance, about a billing issue, and we'll ask the individual to have that bill ready? It's amazing. It's amazing how many people don't. And then which eats away at Time, the time-based analysis here? Why you literally have the agents waiting on the phone while the customers find their bill loaded? or open their email, all of a sudden, my wallet? dropping everything, It's a wait time. So even ensuring even another one is like authentication, making sure that takes place right. Again, saving time or spending time with an agent, providing your card details and numbers, and like that's a minute right there. So considering really doubling down on medication, for instance, or doubling down on click the call functions within mobile to help alleviate that authentication issue, go directly to the agent authenticated. Another one, I'll be quick here. Another one here is a proper call agent routing. So matching customers to agents, you know, we've done this through ensuring call paths are clear. And sharing the numbers listed on our digital properties actually point to the correct destination, that sometimes doesn't happen for some reason. Now, and now we have AI and segmentation in place. And another thing is that again, authentication is also key when colleges don't authenticate or press zero, that actually takes more time for them. And then you get them basically bouncing around from agent telling their story from story to story. And again, eating away time anyway, money. In the middle here, below I have simple. Yeah, just a real simple, quick form anybody can use to try to determine the cost to serve cost savings, a real simple equation, you have. And how you do this to generate a cost or cost to serve equation is. And this is a lot of stuff we do is about actually quantifying our design efforts. So this is the formula example that I use average call times, let's say five minutes, right times cost per minute, let's say $1 per minute when you take into account all the services, the function, the fees, the all that kind of stuff, Times called volume will see that 100,000 collars cost to serve right there is 500,000 on that 100,000 caller list, right? And so after that, you want to divide by your cost savings goal, let's just use 20% for this instance. So you do 500,000 costs to serve divided by 20% is equal to $100,000 in goals and savings and then the fun part is obviously trying to determine design strategy. Obviously to reach those cost savings. And that might be stuff like, you know, improving discoverability on digital or being incorporating predictive proactive measures, right. And then finally, sales revenue lift is another one that we utilize, obviously. So thinking about ways to automate sales offers being presented to callers during those key moments, like stuff like overdraft protection, when we see customers constantly dipping into overdraft and being charged, or credit increases, stuff like that. So really trying to get proactive to help our customers manage your finances and avoid unnecessary charges.

Once the implementation actually happens, what are the success criteria that people should be looking at?

Sarandeep: I think Emanuel covered some of this in terms of your ROI, your cost-saving, etc. And, to me, the main KPI is for us to avoid the calls at all, we don't want the calls to come in. But if they do have to come in you are landing them at the right place. So for me, I think first call resolution. So it's the calling it a calling? It's important that they get that it gets resolved in the first instance, you don't have to really, you know, they should not be needed to call, they should not need to call you back again. Again, Emanuel covered that, what's the average handle time and you need to reduce that by another one, which is important to me is the CSET score, because the customers will tell you what the experience was, ultimately, track that in an ongoing, yes, the adult nature. So I think the CSET score is an important one, for me, if I have to look at the KPIs or the success criteria.

What, what are your thoughts on the kind of the future of IVR looking like, and especially in regards to the voice assistant?

Maria: I mean, obviously, we are moving towards an era of the smart home the Internet of Things, we will be talking to our fridges, I think, in the near future, talking to our window blinds. Technologies are here already, but it's just not widespread. It's not yet widespread. And many people are not even familiar with that yet. So we're not at that point yet where everybody has adopted it, as the smartphone and everybody has one or two and knows how to use them and given, so but I think we have a bit of time to get there still, as this is from both a technology perspective and the adoption perspective. So I think IVR is going to stick around for especially now in the post COVID era, where it's all about touchless technologies, clouds, this dejected stylization, and IVR has been a part of touchstone touchless technologies for decades. So I think it's, it's this post COVID era is, is enforcing the use of IVR rather, but I think it's possible that I don't know, in 20 years, or 10 years, we will still have IVR calls. But what we think now as IVR calls solely through the phone, it will be an IVR call through your smart speaker. So the same technology, but through your smart speaker or instead of your landline, for example, or from your I don't know, fridge or something else. Yeah, I think it will be there, but just the different shape and form.

Sarandeep: I think we will see a lot of technology evolution in space and this space is going to mature within the next four years. There are things like voice biometrics that we will have, we will that, will get mainstream adoption, right because your voice is your password and why not use that and stuff like that, but again, there are some legal and regulatory and privacy concerns at the moment for technology which need to really think about as we go along. But I think we are evolving fast pace. I think the industry is moving at a very, very fast pace. Every day New Tech, new techniques are coming, passing. A lot of other things that are that we're seeing, which we did not in the last 10 years, maybe. Right, I think we are right on moving at a very fast pace, and we will see a lot of the market might completely change over the next 10 years, I would say.

Karthik: I think, Maria hit upon some of this, the voice assistants will really look at the voice as a modality. If people want to speak, they speak, if they want to touch they touch, if they want to click, they click, they go to the device that will let them do that. The concept of the IVR is very tightly tied to telephony infrastructure, as we know it today. The real input there is voice and nothing else. So the future is very much, you know, using voice as a trigger, the second one I would look at is, which I'm seeing in certain companies go this route is your customer service reps, or, you know, call centers, the concept of call centers is really changing into contact centers. And it's about being asynchronous. So I can just tell with my surgery, what my issue is, or I can go put something into the mobile app, and the issue gets taken care of, I get a callback if you need more information from me. As I'm not going to be on the phone with you. It's not synchronous communication anymore, it's asynchronous, you will get an update when your issues are taken care of. And with that, it's far greater, far more efficient for a lot of enterprises to deliver service to customers. without necessarily having to be on hold, it's easier for the customer as well, because the frustration of incessant weightings and repetition is gone. So, you know, it's twofold. One is the voice as a modality and not the phone as a device. And the second one is the move to asynchronous support.

Emanuel: First of all, I think those are gonna be absolutely like, kind of a blend between all these technologies and utilizing voice, whether that's in your car or on your fridge, or whatever the case may be. But looking at use cases we just heard that are associated with issues being spoken, and then getting resolved and updated only when they're resolved. I think those are the kind of things we're trying to strive for. So it becomes less taxing on our systems on our employees and allows us to really get back to relationship building, I think what we're trying to do is the free up capacity for relationship building. And so all those like high level, kind of like future vision approaches are things we're looking at to try to get to that point. So I agree, everything we heard here, it's gonna be, changing I think, in the next couple years.