DRUID Talks Season 2 Ep#4: Architecting the Future Conversational AI in the Modern Enterprise

Join our host, Kieran Gilmurray, and his guest, Wayne Butterfield, Global Lead Contact Center Transformation at ISG (Information Services Group), to discover how conversational AI empowers existing technologies, accelerates digital transformation, and prioritizes employee experience in automation, and understand the pivotal role of Open AI - ChatGPT in enhancing customer interactions and driving innovation. Let's get started!

Kieran Gilmurray:

Welcome to DRUID Talks. Wayne Butterfield is a pioneer in the intelligent automation industry. His work has been the subject of research and books. Wayne is a prominent speaker and publisher of intelligent automation and AI content and is currently the Head of AI and Automation at ISG. Wayne, welcome to DRUID Talks.

Wayne Butterfield:

Well, thank you, Kieran. A very familiar face yourself as well.
Thank you. You're almost making me blush there.

Kieran Gilmurray:

If I could make you blush, that would make my day. After that, I'll worry about the questions. Let’s jump in for those who don't know. You'll be very surprised. I think most people will know you in the industry, but I don't think at times they realize that way back in the day, you were at the very sharp end of what we now call conversational AI and chatbots, but they weren't called that back then. Would you mind sharing your experience with conversational AI text and everything else with our good audience today?

What is your experience with Conversational AI (CAI)
to date?

Wayne Butterfield:

Yeah. So I just feel like it is almost a decade ago now, and you're right; I would say we’re living in an industry that reinvents similar terms. So conversational AI probably wasn't what it was called then, but we did have virtual assistants, even before that, if I think about where we started. So, back in 2020, just not long after RPA. So that was in 2010, and probably in about 2012, we had one of the largest online chat operations in the world at that point. A great team there pioneered this new digital channel much earlier than anybody else. And that meant that we had about 600 -700 hundred thousand online digital conversations, which, again, back then, was a lot. So, huge volumes of these online digital conversations, and that, for me, was the precursor to conversational AI as we know it today. It is very difficult to convert somebody from phone speaking to interacting with a human to interacting with a digital avatar. So, that stepping stone of online chat took people from the phone to texting people online and then being able to maneuver those conversations from a person to a bot.

And we had Lucy, who was our AI avatar at the time. You can tell that we were early on in the adoption there because the number one question asked of Lucy, which again was AI Avatar, was, “Are you married, or do you have a boyfriend?” So, 2012/2013 was when I first got my hands dirty with conversational AI. We used to call it NLP back, but like pure NLP back then, that opened me up to look into what a customer is asking and the responses we can give. From there, I developed my love for digitizing and automating the customer service group.

Kieran Gilmurray:

Interestingly, because everybody just assumes these technologies have appeared today, their foundation and life cycle have continued over the last 15 and 20 years. And that's why your experience is so fascinating, to understand what worked and what didn't. And it's funny hearing the question, are you married about it? You can't imagine anybody asking that this day, or at least I hope not. I hope we've moved on.

What about conversational AI in its current guise? Do you feel it empowers existing technologies and companies to be more digital and allows them to do that faster? Or what do you think its role is these days?

Wayne Butterfield:

I think it's probably necessary for many reasons mainly because, you know, we have evolved as humans, you know, this is connected to my hand now we're always on as people. I think because we're always on, we always need to be served, and that is incredibly expensive for brands to do if their only workforce is a human right. You need to have your silicone workforce; whether it's robotic process automation or conversational AI, these are silicone workers. And so any brand that isn't always on risks losing the opportunity of even unknowingly making their consumers annoyed that they've gone to their website. Their hours of operation do not align with when I have time, and therefore, I am annoyed without ever telling you maybe I've lost a sale. I think it's a necessity now, and it doesn't necessarily mean that every conversational AI conversation and deployment worldwide is and has been a good one. But the capability and foundation this type of interaction provides you is necessary as we think about the future. And you know technology continues and will always evolve. And so whether or not you agree that the current deployments of most conversation layer technologies are transformational, then you know what we've seen with current improvements in how this type of technology can converse. Just tell me that it's almost going to be impossible for most conversations in the near and medium future to be still completed by people. It's just it's just not going to happen. And so you know, for me, this is again a foundational technology, a foundational capability, and having that capability of not having that capability, I think, is going to be to the detriment of any organization, big or small, in the future.

Kieran Gilmurray:

Quite a massive statement, and again, I'm referring to moving back in the day to now. In contrast, at the very beginning, people objected to technology and then wanted a person, and we've almost gone 360° from what you're describing. But for that to happen, there must be benefits for the organization and the customer.

So what benefits are you seeing that are compelling businesses and compelling people to want the silicone, not the physical person or maybe not exclusively the physical person?

Wayne Butterfield:

Yeah. And I I'm a big believer in context strategy and context, and these other heap abilities and and and things you need to think about with, you know, with any deployment of conversational AI don't think you can only have a conversational AI strategy. It can't be standalone. I think we need to be mindful that we know the right time, query, and channel in the right context, and you know you need to think about it in those kinds of areas to make this a true success and not lead to people having frustrations. So you have to think about this as a new starter. You've always got room, and always, you know. Even in the future, you will always have agents and human agents who are at least for the next 5 to 10 to fifteen years. I guess you want to have to deal with certain queries, but you only want them to be dealing with those types of queries and anything that is simpler, transactional, and needs more of an immediate response. And I guess lastly, those that the customer is comfortable dealing with a silicone worker, you know, you're just going to have to have that kind of in place. I don't think the right thing ever to do is force a silicone worker on a customer. I think you need to be mindful that certain customers in certain demographics at certain times and scenarios are just not going to be relevant for conversational AI to take the place of people.

But you know that is there's a whole, there's going to be a whole new piece of, I guess, work for enterprises in the future to drill down into the guts of their organization to understand which queries you know with which context should be with people and those that should be with silicone work with conversational AI. And it's an interesting one that I think we'll see over the next decade now that we've started to get some of these technical capabilities in place. And you know there will be winners and losers; I suggest deploying all of this technology as we've seen today. But I think yeah, it's, without doubt, this is over the next ten years I think is going to be significantly more exciting and better than what we'd seen over the previous decade in this kind of area as we awaken to some of those, to some of those benefits. Yeah. And those benefits for me, if I think about it from a consumer perspective, the always on, you know that for me as somebody who is pretty much always on and always connected, I find it incredibly frustrating to go to a website into an app and have the sorry, we'll be back at 9:00 AM tomorrow. I'm sorry, I'm working at 9:00 AM tomorrow, but I am available at 11:15 tonight before bed. And it's, you know, in my view, a reasonably transactional type query that I may have. I just haven't been able to do it on your app or your website. Not particularly clear.

And yet you know you are now forcing me to have to wait and speak to a person because clearly, you haven't got silicone workers available in those though the kind of the later hours. This annoys me because I don't think I need to speak to someone to resolve this query. But I need more help than what you've given me so far. From a business perspective, there is always an opportunity for cost savings, but there is also an opportunity for increasing revenue. I mean, if mine was a sales query that I had, I may now be frustrated because I haven't been able to resolve that query, and I may have gone to your competition. So, you know, revenue is probably one we don't speak about enough with conversational AI. Of course, we always think about cost savings because we are deflecting contacts away from people into this more digitally automated channel. But that kind of always on that connectivity that you can have with your with your clients, especially some of the younger generations. As we think about the future, consider the longer term, the next 5 to 10 years, and the new generations with spending power; they will expect it. And for those that don't have it, there will be problems.

Kieran Gilmurray:

It's almost like an arms race to some degree. And I love what you're saying there. By the way, Wayne, I'm looking at five years plus because I worry that companies sometimes look at the next 12 months of their next short-term period and don't think a bit more strategically. And like you, I know want, expect, and demand service when I want because I know the technology's there, and if it's not there, it's so simple to move to someone's partner. It's phenomenal. It is phenomenally easy, just shifting gears a little bit.

In your recent article, "Top Five Intelligent Automation Predictions for 2023", you said a major RPA player will buy a conversational AI platform. Why do you think that? What does the business need? Or what was the reasoning behind your proclamation?

Wayne Butterfield:

So you think about it, and we have slightly different takes on this. Think about anything: clickety-click, right-hand work, or headwork. An RPA player is no longer purely an RPA player. We've seen this over the last 2-3 years, various acquisitions and strategic partnerships, and so even the RPA players now are more than just the clickety-click or the handwork players of the market, right? They have realized that they need to create an automation platform because automation of work is far more complex than typing, clicking, moving them out, copying, and pasting, right? We get that. And if you think about where the market is going and what clients are demanding, you know it's automation AI and automation everywhere. You can't get away from it. And so when I think about the digital worker, I often think about it across three or four, maybe 5-5 key areas, like the replication of the eyes, the ability to speak, understand, interpret, etcetera. Converse is a key component of most work, whether you're conversing with a colleague or conversing with a customer.

Without a conversational AI player, you have a digital worker who cannot communicate verbally, so you're missing this kind of key capability from your platform. And that, for me, is kind of why I'm thinking a conversational AI player will be acquired. Now, it's a UiPath that did go some route with purchasing a Reinfir a few years ago. Maybe it was about 18 months ago now, but that is generally retrospective and more conversational intelligence versus, I would say conversational AI chat-pointed type technology. So, I still think there's a gap there. I'd be really surprised if it doesn't get filled. I mean, I can't see how you can be an automation platform and not have that ability to converse and maybe give your boss a voice, give your if your PDF, you know, via IDP voice and ability to communicate. There's something that feels like it's still very much missing from every RPA player, maybe apart from Microsoft, who would say that their power platform already has that capability.

Kieran Gilmurray:

I wouldn't argue against Microsoft. They say a lot, but no, it's so true. The most natural thing we humans do is not type or click it. It is that conversational piece; therefore, the conversation pieces are missing for trying to replicate a digital worker to be somewhat supportive or equivalent to a human worker. Now you mentioned a quote clickety-click, and I can't help but playback yours, which is if you can think it and you can say it, then ChatGPT can create it.

Where do you think ChatGPT or generative AI fits in the context of conversational AI for businesses? What should companies do with it?

Wayne Butterfield:

I had my bill this morning for my Open AI monthly subscription. And you know, I thought I hadn't used it in the last few weeks at least. So, I think generative AI as a complete segment is incredibly interesting. I have spent thanks to another great thought leader, Danilo McGarry. I've spent the last couple of weeks deep-diving into avatars, deep-fake voice synthesizing, and speaking with those companies in that space, building my capabilities. So, you know, generative AI, I think, is we have a lot of hype riding away the business value. I still think we need to get under. I still see that with ChatGPT, there's cool technology. One of the other quotes that I use is "It's cool," but not corporate, and what I mean by that is technology. I think it always has been that you can talk to your mates about it, but when it makes it into the enterprise,  it's more of a worry than something you're getting value from right now. It's more, I guess, not scaremongering in the enterprise. There is a real understanding of the value of how you can change work because many of the examples you see everywhere again call, but is it going to transform the way that I work? Is it going to save me time on a regular occurrence, and when I remain regular, will it save me 5 minutes every hour? Not only will it save me 5 minutes every day, but I can save 5 minutes every day by doing something completely different that will be much more secure and easier to implement than something like ChatGPT in the enterprise. But this is how I think about all of these technologies and how we all have to think about them up until you know.

November, so less than a year ago. Most people did not think they would have this big a jump in technical capability in the language space, and others just think about where we will get to now. This focus will be on it in the next 12, 24 and 36 months. These technologies are as bad as they're ever going to be. They're only really going to get better. Now, we may find that transformers can only get us to a certain point, and everyone now is appearing to, you know, pick nitpick at some of these capabilities. I just think we need to be mindful of GPT, Open AI, you know, many other organizations are all working on fixing some of these problems, things like Copilots and other capabilities that are going to be generated on the back of these capabilities of where business value is going to come from. So, what do I think about about ChatGPT right now? I think it's cool. I use it when I think it's appropriate.

I think it's some great user cases for it. You could take this transcript from this conversation and say I need you to write me the social media brief to discuss this. Great. Do you know what? It saved me 10 to 15 minutes. And that could transform my day if I did this every hour, every week. Brilliant. I just saved myself half a day. So, I think we're still in the early days of getting through regular value from these capabilities.

But once we've cracked that nut, and believe me, I spend much of my time thinking about it. Once we crack the nut, I don't want to say it will be transformational because I feel like every technology could be transformational. But I feel like there is a step forward in changing how we work. Whether or not that delivers a significant amount of time-saving, I think, is still to be known. Still, for things like composing and summarizing, I think we're just going to use technology very quickly to be able to do all of those things. Just like going back to conversational AI, we will use conversational AI to do very simple transactional activities. Again, the crossroads between an enterprise using ChatGPT and its current conversational deployment is not corporate. My big worry is about an enterprise-like certainty, the entire premise of a large language model. If it is going to give you something based on prediction, it will not necessarily be exactly as you want. Some people would call those hallucinations. Therefore, I think we just need to work a little more on things like guardrails to ensure that what we say, what we respond, or what they respond to is still classed as corporate. But you know, I think this entire space is so exciting, and I think we have years of new stuff coming out that is just really going to be interesting to be a part of and interesting to see how the enterprise changes the way that it utilizes this type of technology moving forward.

Kieran Gilmurray:

Interesting,  I love your perspective on that. I saw the stat the other day. I was listening to the HBR webcast, and they said last year, I think, in Silicon Valley, there were 80 generative AI start-ups. Now there's 8000, which shows you hype or interest, or all of a sudden, and I'm very vocal about supporting ChatGPT or the OR the derivatives while putting guardrails around it. I used to have Google open every day, and now I've got Grammarly ChatGPT, Bing, Bard, and meta GPT, and I run through each of them to do many different things. Active to passive sentences, ideas, proposed summarizing text, you name it, doing some data analysis, and whatever else. But I get what you're saying, and you see this on the hype curve as well, and the development, get the technology out, fix some of the things, move it forward and round and round you go in an iterative phase. There are a lot of parallels, as you and I will have recognized with our PA back in the day. So it'd be interesting to see how it develops and grows.

If you were to characterize conversational AI in one sentence wing, how would you characterize it?

Wayne Butterfield:

That's pointed me on the spot. If I were to characterize conversational AI in one sentence, what would that be? That's the most difficult question you've asked me today, mate.

Kieran Gilmurray:

Refer to ChatGPT.

Wayne Butterfield:

That's exactly what I was just going to say. I've just copied this down. Give me ten responses to this question. Now, this would be my use of ChatGPT today. So Wayne, GPT, as I'm speaking to you, would be that this is hard, the automation of human conversations with brands. Is that what I say? No, that's a poor one.

Kieran Gilmurray:

I'm glad I got your stomach here because you could answer 99.999% of anything about intelligence.

Wayne Butterfield:

Automation and I have too many faces. It's such a wide scope here. But the conversation lay to me is the automation of human conversation, right? That's what it is to me. The automation of human conversation is in a nutshell, but that's almost too simplistic, and it's so much or can be so much more than that. Because when I think about the conversation layer, I think about it as three things. I think about it as chatbots and voice bots automating human conversation. I think about it in the sense of understanding conversation in things like aging assistance technologies in the IVR. Then, I also think about it as being able to interpret conversations for use in analytics using conversational intelligence. When I think of conversational AI, I think about it through those three lenses.

We spoke about chatbots and voice bots in the main today. And that's what got me stumped because conversational AI is those three things, and that's a long sentence, which is much longer than just one sentence. So that's why I had me stumped. But then I went for the simple automation of human conversation.

Kieran Gilmurray:

I love it. We'll run it through ChatGPT because it's interesting. I wonder and worry: do people think of conversational AI as a chatbot? Which, to me, was a simple question and answer a long time ago, truly back in the day. And now you see, you know, chatbots plus AI to get towards conversational AI, plus NLP plus a whole host more to get to conversational AI. Then, use and integrate conversational AI with conversational business applications, ChatGPT, or other variants to suddenly move the conversation dramatically. It's an interesting space. It is.

Wayne, thank you so much for today. We should have ChatGPT in the background next, then ask it how we do, and then could you score us and give us ten other questions that we could have asked or answered? Unfortunately, we don't have that amount of time. Look, it is an interesting space. I'm glad you brought us through the history of everything there, and that's exciting, but probably why you or I are in this industry more than anything is brilliant. Tech keeps happening and happening and happening, although I do see the cycle shortening and shortening and shortening, and I wonder whether that scares the life at me now or excites me. There is a lot of high-brown conversation on ChatGPT, but there's a lot of promise and value. I see stuff like, you know, Microsoft Open AI when integrated securely and companies implement that correctly, it's good. As you mentioned, we have been very conscious of the governance, risk, bias, and everything else that we cannot move away from. But that is what makes, as you mentioned earlier on, you know, something that augments something that saves you money. And again, I love that you said it as well, something that now drives income as well.

So it's not one thing, as you mentioned. It's not one sentence as we closed. It's many, many things. And I appreciate you calling those all out for the audience today. Thank you so much.

Wayne Butterfield:

Indeed, no. Thank you for having me, Kieran, as always. It was an absolute pleasure speaking with you and having our minds meet as we discussed, you know, this and another topic. So superb. Thank you very much for having me.

Kieran Gilmurray:

Thank you.

The next episode of DRUID Talks is scheduled for January. Subscribe to be notified at https://druidai.com/talks.