Conversational Messaging Podcast by Gupshup

Episode 007: Conversational AI: Humanizing Interactions Between Brands and Consumers

June 24, 2021 Beerud Sheth and Srinivas Vijayaraghavan Season 1 Episode 7
Conversational Messaging Podcast by Gupshup
Episode 007: Conversational AI: Humanizing Interactions Between Brands and Consumers
Show Notes Transcript

In today’s episode of the Conversational Messaging Podcast, Product Head of Conversational AI, Niyati Agarwal joins Beerud and Srini to discuss how Conversational AI technology changes the course of brand interaction with consumers. Tune in to know how Conversational AI impacts your brand reputation, reduces your manpower’s effort, and optimises your business costs. 


The “Magic” of Conversational AI Technology

Conversational AI technology is ideal whether you want to have an unstructured (free style) or structured (straightforward) communication with consumers. Through natural language understanding (NLP), AI understands the intent of consumers’ query and provides the right responses. With advanced techniques, AI can solve even the most complex queries no matter how complicated consumers phrase them, which can ultimately have a positive impact on the brands’ reputation. 

“You take 10 people, and they'll have 10 different ways of asking the same question. That's where you need artificial intelligence. You need this field called natural language processing to understand the user query.” - Beerud Sheth


The Evolution of Conversational AI Technology

Niyati points out that AI uses “reactive” and “proactive” techniques. Aside from the expected resolution, AI also offers other recommendations based on the consumers’ history. These techniques prompt the incorporation of AI to other business processes like marketing, human resources, ITSM, and such. Conversational AI technology is a good space to explore, as its concepts and channels continue to evolve in a way that greatly reduces the efforts of not only the consumers but also the brands. 

“We are trying to not replace humans. We are trying to have a more human-like conversation that has the power of automation and intelligence a machine could have.” - Niyati Agarwal


The Ease of Conversational AI Technology

The demand for Conversational AI technology rapidly increases, so small brands should not be hesitant to explore and tap companies who specialise on this type of technology. Companies like Gupshup put everything together by providing pre-defined industry templates and other tools you’ll need in automating your consumer interactions. Beerud sees a great development in Conversational AI technology’s human style mimicry, and he expects many brands to have their own “Alexa” in the next few years. 

“We are automating them, and we are providing tools, so that this becomes a mundane task for you. It becomes something which you do not need to go to your IT team to implement.”  - Niyati Agarwal


To know more about The Power of Conversational AI Technology, listen to this episode. 



Bio:

Niyati Agarwal 

Product Management Head for AI


Resources:

Gupshup.io

Gupshup Conversational AI

 GupShup_007

Mon, 6/21 9:47AM • 25:20

SUMMARY KEYWORDS

ai, conversational, talk, query, conversation, nlp, message, understand, device, technology, marketing, human, happening, realise, capabilities, people, business, plug, evolved, user

 

00:01

Listen to insights on how conversational messaging is changing the way businesses and their customers engage. Join Gupshup CEO Beerud and VP for Marketing Srini, and an array of guests for conversations about conversations. This is the Gupshup Conversational Messaging Podcast.

 

00:28

Hello, and welcome to another episode of the Conversational Messaging Podcast brought to you by Gupshup. As always Srini and Beerud here, we bring a new topic every week. And this week, the topic is conversational AI. And some of you our listeners may have heard about this, it's a pretty hot topic, it's there's a lot of buzz around it. And just to set it up, I think it's a kind of a holy grail for technology in general, where, you know, humans have sort of had to learn machine languages and programming languages to interact with computers and phones. But at the same time, these machines are also being taught to really interpret what human beings are trying to say or what they're trying to talk or what they're trying to type. I think a key enabler of that end capability is conversational AI. So Beerurd, let me ask you first, I know, you've written a lot about this on how this whole machine to human interaction, how it has been changing over the last 20 to 30 years. So how do you see that transformation taking place? And what's the role of conversational AI, in that? And in that process, maybe you can also sort of demystify this whole heavy term for our listeners.

 

01:42 Beerud

Sure. Well, I think as we've, as we've discussed in an earlier podcast episode, I think on the big screen, human computer interfaces have been evolving over the last few decades, it was desktop in the mid 80s, went to the web sort of endeavor to them in the mid 90s. It went from sort of this client side metaphor to a server side framework where you use a browser to visit websites on the server side. And then on the small screen, again, we are in this similar transition where we started with mobile apps, which are client side experiences, and now moving to chatbots, and conversational experiences. So and you use the messaging app, which is analogous to the browser to go interact with different chat bots, which are sort of analogous to, to a website. Right, so so as that's happening, basically, the net effect of this is for a consumer, it's as easy to chat with the business as it is to chat with a friend. Right? So if you have any questions, you know, you want to deal and offer you want to buy something, you have a question you want to report a problem, fix anything you recognise, you know, consumer should simply be able to engage in a chat conversation and be able to resolve it. Now, when you look at conversational experiences, right? There are very broadly speaking two kinds, right, you can have a structured conversation experience, and many of the listeners will be familiar with, you know, you've seen messages with quick replies that they give you an option, right? For example, you know, which size do you want small, medium, large, and then you click on, you know, small, let's say, and then it asks you what colour Do you want blue, green, or red, right? And so on, there could be multiple options. So you could have a carousel, you could have some structured elements that you can interact with right? Now here, it's a it's, this is a structured conversation, which can be programmed with a, with a very straightforward, menu driven kind of flow, right? It's like a flowchart of a conversation, and you're going from one step to another. And it can, it can be linked together in that sense. And you can do this without needing any AI, per se, right. But the other kinds of conversations are the unstructured conversations unstructured as sort of, you know, complete freestyle, natural language conversation where the user can phrase their query, in sort of literally any, any form, right, and they do it on their own, without any prompting. So when a user phrases a query, and sort of natural language, as we know, right, humans, you take 10 people, they'll have 10 different ways of asking the same question in different ways. And that's where you need artificial intelligence, you need in particular, this field called natural language understanding or natural language processing NLP to really understand the user query, right? So the, the chatbot, the software needs to understand the user query, what are they trying to ask, like, what's called the intent you need to extract the intent of the query? Are they trying to open an account, cancel an account, refund an item, you know, and so on, right? So you'd have to figure out the intent, and then also figure out the entities around it. Right. So for example, they're trying to book a flight, and then from which city to which city and so on, right. And in and actually in a sort of naturally phrased language, you know, in a query like that, it can be sort of a complex challenge, right. And AI, it needs a lot of advanced AI techniques to figure that out. So anyway, that's sort of conversational AI, is a set of technologies that help software chatbots, understand human queries. And then of course, you know, provide a response with the right information requested. But but that response is not possible unless you really understand the query, for which you need conversational AI.

 

05:38

Great, thanks for that explanation Beerud. And while usually on most episodes, it's you and me talking, I want to welcome our very special guests, to all our listeners, I want to welcome Nyeti to the show, and Nyeti is at Gupshup, she heads Product Management for all AI suite of products. And she's worked for many years for companies where AI has been has been at the core of product development and innovation. So first of all Nyeti, welcome to the show. 


06:03 Niyati

Thanks Srini, really pleasure to be here. 


06:07 

Yeah, that's great. And I mean, you heard Beerud giving us that, that context of how this whole space has evolved? How do you see conversational AI today in terms of the here and the now based on your experience of working across all these companies? 

 

06:21 Niyati

So when we talk about conversational AI, it's in one liner, it's basically me or a human talking to a machine, and everything that we talk about in that context where, let's say, I need a more personalised approach, or I need the machine to do something for me, that's all encapsulated together. So it's on the same lines where you know, I can manage my funds on my own, I can manage my finances on my own, but I still need a personal advisor, a personal finance person who can just talk or listen to my queries and understand me in length. And it's the same case where conversational AI, on one hand, it's, of course, solving my problems, but it's giving me that personalised approach to understand everything that I mean, and everything that I've done in the past and also understand what I'm here for to solve my issue. And I see it in the same context. So when I talk about, let's say, a conversational AI approach, of course, it means NLP it means machine learning, it means semantic analysis and everything on those lines. But it also means language understanding, it means a multilingual approach where I say something in some of the language on the board can understand. And it also means a more real time contextual understanding as well, which is not just you know what I'm saying here, it also means what's the time of the day? What is what day it is what conversation history have had? What is the tone in which I'm talking? What is my, you know, getting to depths of intense entities, sentence structures, and I think all of that in that context. And maybe it could also mean, where I'm talking on one channel, and I can have the same conversation leaned upon in another channel, or in another context, everything encapsulated together, that's what conversational AI is today. And that's what we are building, we are trying to not replace humans. But I think we are trying to have a more human like conversation, but yet have the power of automation and intelligence that the machine could have. So yeah, it's the perfect mix, according to me.

 

08:30

Now that's well put, so I have a follow up question for you in terms of so this capability sounds like it is fairly universal. So I mean, can just spend some time talking to us about how this space has evolved over time, like I have, from whatever I have seen that you have, like lots of companies that are specialising in conversational AI applied to a particular problem, like a marketing problem, or customer support problem. And there are certain companies that are sort of just building platforms that others can plug into and use. So how have you seen these different segments come up in the last few years, as the whole market has grown?

 

09:09 Niyati

You know, honestly, I start with, or let's say, I started with the approach where we used to go to people and tell them, okay, there is a chatbot, where you can talk and there is a robot like structure on the other end, which can reply and give replies to your answers. It started from very basic approach or FAQ or customer support. And there was a hard time where people did not understand AI, they thought it was magic. If it is able to answer my query, it can do anything. I can talk about, you know, when do I get a girlfriend and you know, what's the temperature like and the bots will be able to understand anything and everything that happened with of course, how Google Home Alexa, all those devices evolved. We moved into a more calculative approach. And I think the audience's have also been trained now and they've evolved their thinking has evolved and now, it's an audience where you talk about chatbot. And they understand that it's not just AI, it's about the used case as well. So you they're on an assistant and you want to solve your issue, that's the first game changer, how you're able to solve your issue that comes next. So of course, the AI is important know, what can be built without NLP approaches or AI technologies. But the first and foremost thing that the chatbot does is to solve your queries. So it started from, of course, the most basic thing that everybody wanted to talk about customer support, that became the low hanging fruit that, you know, everybody could achieve. Because it's those FAQs, there are pre trained questions, people found out a pattern, that there are a set of questions that everybody asks and everybody needs answers to, and that can reduce the effort on their customer support agents are, you know, kind of optimise their costs. So it started from there of competition, good nascent technology, and then, you know, as the growth and as the availability grew, as the need grew, people realise that there are other technologies or other verticals that can also be kind of utilised. For example, marketing became a one man show, I remember in my last company, we create a marketing suit, because when you talk about marketing, it's not just sending out one message. And people see that message, the view that message and they do what they want to do, it has to be personalised, you want to send out the right kind of message. So you know, it's more on the lines of, let's say, a reactive and proactive technique, while on one hand, your chatbot, you send out a message and gives a reply. On the other hand, it also has to be very proactive, meaning I can send you the right kind of message that you want to see or you want to hear about. So that became the essence of the marketing chatbot, and the whole marketing automation through into, you know, small things like abandoned carts, or let's say small things like I have not been active on this board for about 10 days. And that becomes a pattern for a chatbot to send me a message or a virtual agent to understand that pattern and engage me in a meaningful conversation. And while that happens, you know, the upstreams also realise that when marketing can do this, of course, sales can do this, the commerce part can do this. They can kind of prequalify their leads, they can collect relevant information from them, and then kind of qualify, what and who would be more likely to get or buy your product. So that became an essential part as well. And after that, of course, if you're selling, then there is upselling, if you sold a catch up to a person, you would try to sell Maggie as well. So that became like upselling product recommendations, and all of that came through. Now this was all happening D to C, direct to consumer. People also realise that, you know, while this is happening, this is good. Why don't I use this conversational AI concept with internal processes, or let's say HR processes, ITSM processes because the agenda was still same, there was a human talking to a machine. Now that human could be your consumer, or it could be an internal employee, both of them did the same things, they wanted to solve an issue. And then over and above, what I also see these days is internal processes talking to each other. There is where you know, we talk about one board talking to the other board, because if you've created a board with a persona, human, of course, that human could talk to other humans. So it's evolving. It's a, it's a new concept. And I think when these concepts are growing on one end, the channels also growing. So people are moving into voice enabled walls, they're moving into speech to text and text to speech. And you know, all of that. Yeah, it's a, it's an interesting field as a lot is happening.

 

14:03

That's really exciting. And Beerud, I think just picking off on what Nyeti said conversational AI sort of conjures up images of something that exists deep within a network or deep within a software system working its magic. But I think a lot of people don't realise that there is AI conditionally AI working on devices that we use every day, like, for example, a mobile phone also has these elements playing out. So you know, can you talk talk to us a little bit about how it's actually moving into the client side as well, apart from the service side, like on mobile devices?

 

14:39 Beerud

Yeah, I think you know, this is something Gupshup has some very unique experience in because any of the users have been using either Xiaomi or one plus devices, then they might have seen Gupshups on device AI technology at work. So what we have done, you're right, most AI is developed on on the server side, because, you know, it takes a lot of cloud computing resources to, to build models and to run models and so on. And it's very hard to shrink them enough and scale them on to the device, you know, and actually run it on the device with with the limited power and the limited resources available on the device, but we've been able to do that, right. And what we did was develop some AI models and software that could look at incoming messages, in this case, SMS messages arriving in the inbox, and he will automatically classify those messages into categories and subcategories. So for example, this is a banking message within banking, it's a deposit credit or a deposit, you know, credit or debit message and then within a credit, right, it could be into a savings account or into a loan account or something, right. So it figures that out, it does the categorization. First, the classification of message and secondly, it extracts the key entities, right? For example, what was the amount and which account they will come out of and then displays it? It does visualisation, right, so we've developed these AI models for classification and visualisation that run entirely on the smartphone device, right inside the messaging app, and sort of leveraging the same sort of natural language processing capabilities, and conversational AI capabilities to really understand the message, the intent of it, right. And these are just style. These are human scripted SMS messages that are automatically classified and visualised, visualised accordingly. So and I think, like I mentioned earlier, not only is that an AI challenge, but it's also a technical challenge to to shrink it to make it work on on the device, not in the cloud, and isn't to run it on the device is primarily because of, you know, user privacy and security, right? You these messages have to be classified and visualised right on the device, you cannot send it to the server side, because they will defeat the whole, you know, issue, they will cause a huge user privacy issue, and so on. So there are many benefits by and these are literally the first applications but you can go further, you know, it could automate certain actions, right, like maybe any bills that come in so long as it's within a certain reasonable amount, maybe paid automatically or reply to certain messages automatically, right. So imagine, at some point, you could have on device chatbots, right, things that work on the, on the device that could enable some very powerful and rich capabilities. So, so yeah, I think, you know, doing sort of AI on the edge, right, on the device is, is another sort of frontier, to or AI to have for conversational AI to have huge impact.

 

18:03 

That's a great point. And I think as these technologies proliferate, we're going to be putting them in the hands of more and more people. And that brings me to another related point on you know, any technology for it to sort of gain a critical mass has to be democratised. So that a lot many people irrespective of how mature they are, or how knowledgeable they are, or companies, irrespective of how mature or knowledgeable they are, in the in the field, can can use it and sort of plug into it. So NLP, for instance, I think, you know, is it possible today for nearly for example, for a company that is that doesn't have you know, AI engineers, or let's say, experts, like you working for them? Is it possible for them to sort of plug into already available libraries or capabilities? Let's say they're building out a port, and they want to add an NLP capability? What what options do they have? Like, what what can they do?

 

18:58 Niyati

That's, I think, the exact vision with which we're also working here, because, for example, you know, a restaurant owner is supposed to cook food and you know, take care of his business, they should not be concerned about what's happening on the AI end, or what can be done with chat bots, or what can be done with virtual agents. So we are here to solve that particular problem. So we enable everybody who, you know, these business users and anybody who comes to us, brands to leverage the AI tools, yet having that simplicity that anybody from, let's say, a developer to somebody who's coming from a marketing background and does not understand how to and you know how to use an API can enable our tools to make sure that they are also able to leverage the same kind of automation for their clients. So we have predefined templates. For example, these templates are industry specific, their use case specific, they also target to different personas of chatbots. And the reason we have these is so that it can be a plug and play application, somebody needs to just click on one button and click on a few buttons, and they have an automation suit ready for them. Similarly, you know, like, we're also talking about device AI, or we're talking about simplicity, in which use cases have that patterns enabled. Again, you know, if you, let's say 80% or 90% of your business use cases, whether it be a bank or whether it be a ecommerce site, or whether it be any business in this world have repetitive processes. So we are kind of automating them and we are providing tools, so that this becomes a mundane task for you, it becomes something which you do not need to go to your IT team to implement that. So the heavy loading is something that we will carry out at our end, and it will become DUI based experience, or it will become a five minutes job for you to come and tell us what you need to do. And we'll manage the rest. Because it's not that simple. You know, anybody can create a bot for sure. But to have that customer experience, or to provide the right set of automation to your customers can be difficult, you know, as a simple example, you know, we talk about IVR is, you know, you talk to an IVR. And there are 1-2-3-4-5 options, the first time it is, it's very interesting for you to press on first button, and then on the second button and the third button. But the next time you visit an IVR, you want that process to be automated, you know, exactly 1234 is the code I want to use. And I want to talk to a customer support agent. So that whole process where things are different for different kinds of audiences is not something that we need our businesses to worry about that something that will, you know, we'll have it simplified for you. And that's the essence of the platform and the various tools that we're building on top of it. So we have our local platform, we have no core platform, we have industry templates, we have a scripting tool, where everything that you want to do every kind of automation that you want to achieve can be done over a conversational or over limited set of intelligence that you'd like to have on the AI, NLP NLU sentiment analysis. And you know, all those AI tools that we talked about so heavily.

 

22:30

Right. Now, that's great. I think it sort of goes a long way in helping companies, like I said, across the maturity spectrum, to adopt AI and to sort of plug in NLP to do what they're doing. So in closing thoughts, Beerud, coming back to you like, how do you see the proliferation of conversation there, especially in the area of messaging, as we, as we see a lot of businesses sort of wanting to converse with their customers on messaging channels. So where do you see the next couple of years with this technology? How's it gonna drive messaging?

 

23:08 Beerud

I think, you know, it's an exciting time, because the state of the art in AI has improved dramatically over the last couple of years, especially around natural language processing. Okay. There are these, there's a lot of advanced academic research to create new language models that you can use as a starting point and fine tune it for a specific domain or for a specific purpose. Right. So the AI technology, conversational AI technology is already pretty good. The last remaining bit is to fine tune it to specific domains, and to sort of tie it into chatbots and roll it out. And I think that I don't, I don't, I do not expect it to take a long time, right, it will happen. So certainly within the next year or two, you're going to see very rapid proliferation of conversational AI into these chat bots that can mimic human style, you know, natural language conversations and I think that will be a really powerful opportunity. So I see it proliferating across a lot of businesses and brands. And you know, just like Amazon Alexa and so on a very popular pretty you're gonna have every business every brand, you know, building their own Alexa, if you will. We may you know, as consumers, we will all be talking to having conversations with, with brands, and quite possibly, we may all have more sort of AI friends than human friends. So much conversation with all of them having too much fun, I guess.

 

24:43

Yeah, I look forward to living in that world. It'll be interesting for sure. So thanks always Thank you, Beerud. And special thanks to Nyeti, thanks for sharing your views and to our listeners. We will join you next week with another episode and possibly another special guest. Thank you for your time today and thanks to both of you.