Latest / The Adults In The AI Room / Chris Featherstone Talks About Building Omnichannel Virtual Assistants - Episode 1

Transcript

Chris Featherstone Talks About Building Omnichannel Virtual Assistants - Episode 1 · The Adults In The AI Room

• 0:00
  SPEAKER_00: We're going to be talking about omnichannel virtual assistants.
• 0:03
  SPEAKER_00: So my name is Seth Early, and I am founder and CEO of Early Information Science.
• 0:08
  SPEAKER_00: We are a professional services firm that works in the area of information management, making information more findable, more usable, more valuable.
• 0:16
  SPEAKER_00: And I am joined by Chris Featherstone.
• 0:19
  SPEAKER_00: Chris, do you want to give a thumbnail of your background?
• 0:22
  SPEAKER_02: I'm Chris Featherstone, and I lead business development for our AI and speech language services at AWS, specifically over speech recognition.
• 0:31
  SPEAKER_02: And I've had a rich opportunity to work with a number of customers across all industry types centered around how to maximize speech recognition and utilizing AI services in order to pull the really rich data out of that information in terms of augmenting human spoken conversations, both in short form, like intent-based, and also in long form.
• 0:57
  SPEAKER_02: So non-structured conversations.
• 0:59
  SPEAKER_02: So thanks, Seth, thanks for the time and appreciate the opportunity to uh to speak with you on this topic.
• 1:05
  SPEAKER_00: Absolutely.
• 1:06
  SPEAKER_00: And what's interesting about uh what AWS has done is they've done a lot of the heavy lifting in terms of the heavy-duty mathematics and machine learning uh to provide platforms that can really support people at multiple levels.
• 1:19
  SPEAKER_00: You can be a uh data scientist, a data software engineer, go in and tweak algorithms, or you can be a business user and actually uh use it to configure um solutions for business problems.
• 1:30
  SPEAKER_00: So great to have you, Chris.
• 1:32
  SPEAKER_00: Here's what we want to cover.
• 1:33
  SPEAKER_00: We're going to talk about the information retrieval continuum and thinking about uh virtual assistance and chatbots on that continuum, because I always think that you know, chatbot is just a channel and it's a mechanism to access information.
• 1:46
  SPEAKER_00: We still need to have that information and a good uh structure.
• 1:50
  SPEAKER_00: Uh, we'll talk about omnichannel when we talk about omni-channel and virtual assistance and voice and chat.
• 1:55
  SPEAKER_00: What do we mean by that?
• 1:57
  SPEAKER_00: And um, and then talk about the differences between voice and text channels.
• 2:01
  SPEAKER_00: We lose uh a lot of um uh nonverbal, uh a lot of the nonverbal uh cues and clues uh when we go right to to text, and there's ways of of dealing with that.
• 2:12
  SPEAKER_00: Uh we want to talk about voice to text, text to voice, and some of the nuances there.
• 2:17
  SPEAKER_00: Think about, you know, just discuss at a high level things like intent classification and analytics and sense, but this is not a deep dive.
• 2:24
  SPEAKER_00: This is really just getting into the high-level topics and uh discussing them, and then thinking about how you can uh provide contextual triggers for human agents, right?
• 2:34
  SPEAKER_00: By by monitoring a conversation and then uh providing some type of a uh of a script or a piece of uh content that would help the human agent in that real-time interaction.
• 2:45
  SPEAKER_00: So that's another way of thinking about how voice uh recognition, speech recognition can be applied to to assist humans.
• 2:54
  SPEAKER_00: Uh thinking about you know, when you're getting started, what do you need to look at in terms of use cases and scenarios?
• 2:59
  SPEAKER_00: Where should you start?
• 3:00
  SPEAKER_00: And really, what are the obstacles to going to production, to really scaling these things?
• 3:05
  SPEAKER_00: And so again, we're we're touching on these topics at a fairly high level.
• 3:09
  SPEAKER_00: Um, but let's look at the first issue, the information retrieval continuum.
• 3:14
  SPEAKER_00: I like to say again that chatbots are simply a mechanism to retrieve information.
• 3:19
  SPEAKER_00: And just like any other uh mechanism, uh, it requires certain data sources, it requires some mechanism for interaction, uh increasing levels of uh sophistication of information architecture, different nuances around the experience, and then the enabling technology.
• 3:36
  SPEAKER_00: So when you think about this, you can kind of, I'm just gonna build this slide for now, but if you look at search to begin with, you you know, search can be applied to any text source, right?
• 3:47
  SPEAKER_00: You you you can you can you don't need metadata, everything is better with metadata, but uh you don't have to have it.
• 3:53
  SPEAKER_00: Uh and it's a search box and it's a list of documents.
• 3:56
  SPEAKER_00: When you start moving up the continuum of more sophisticated uh information retrieval, and you can actually use the same architecture, uh knowledge architecture to support these other classes of applications, you know, knowledge portals are really getting more uh uh rigorous in terms of defining the structures of the data, defining use cases, and maybe integrating multiple sources.
• 4:20
  SPEAKER_00: Uh there's a case study I I have in the book in that I talked about uh with applied materials, where they had about 14 different data sources that needed to be integrated, and that was done through a knowledge portal.
• 4:31
  SPEAKER_00: But again, it's just it's an interaction or retrieval mechanism.
• 4:34
  SPEAKER_00: Then you start looking at when there's no hard and fast definition between virtual agent and intelligent assistant, but we're looking at an increasing level of sophistication.
• 4:43
  SPEAKER_00: And when you start looking at these, you're now recognizing language and and uh and text and recognizing intent to provide an answer, not just to provide a list of documents or you know, people don't want 100 documents, and one of those documents being a 300-page uh uh document, uh they want an answer, right?
• 5:03
  SPEAKER_00: The 300-page document doesn't help, the hundred, you know, search items uh returned doesn't necessarily help.
• 5:09
  SPEAKER_00: Uh, we've done this kind of work for insurance companies where you have very large documents around policies, but then you just need to surface a specific piece of information for a question.
• 5:20
  SPEAKER_00: And you can do that uh via a chat mechanisms.
• 5:23
  SPEAKER_00: And then when you start looking at more sophisticated applications and looking at really the intelligent virtual assistants, you're you're you're able to have more of a conversational interaction, you're able to have uh uh greater uh levels of depth in terms of context and information sources and be able to switch context as necessary.
• 5:45
  SPEAKER_00: And so again, when you start thinking about these things, they really are about information retrieval.
• 5:50
  SPEAKER_00: At the end of the day, that's what it is.
• 5:52
  SPEAKER_00: We built um the All-State Business Insurance Expert many years ago.
• 5:56
  SPEAKER_00: It was a search application, and the client wanted to put an avatar on top of it.
• 6:01
  SPEAKER_00: And I didn't think that was a good idea at the time.
• 6:02
  SPEAKER_00: In retrospect, it was brilliant because people would say, Good morning, Abby, how are you today as All-State Business Insurance Expert?
• 6:08
  SPEAKER_00: And it used some mechanisms that we're going to talk about.
• 6:11
  SPEAKER_00: Um, but again, it's a retrieval mechanism.
• 6:14
  SPEAKER_00: Chris, what are your thoughts here when you start looking at this?
• 6:17
  SPEAKER_00: And you what what kind of uh uh nuances do you do you think are important will be to look at this continuum?
• 6:23
  SPEAKER_02: Well, I think I think at first the discussion warrants a little bit of context as well to our history, because I think the the key was, you know, you and I actually visited a conference together and it was fun because, you know, we just struck up a conversation in the back centered around the topic that was being discussed, which was, you know, I think it was centered around call centers, but how do they put you know provide innovation across it?
• 6:47
  SPEAKER_02: And the person that was speaking, I remember it was somewhat frustrating yet comical because they were utilizing old methods of thinking to try to apply it to next generation approaches.
• 7:00
  SPEAKER_02: And you and I know as well as I do that it's just not going to work, right?
• 7:04
  SPEAKER_02: Because we can't utilize, you can use old patterns, but you can't utilize, you know, old infrastructure.
• 7:10
  SPEAKER_02: I think the the key thing was um, you know, after I introduced myself, you were like, oh, you deal with AI and ML on a daily basis.
• 7:18
  SPEAKER_02: And the I I think the first question out of your mouth was so what do you how how important do you believe that ontologies and taxonomies and classifications are for the data and structure?
• 7:29
  SPEAKER_02: And I was like, I think it took me back a minute because everyone that we talked to, including customers and partners, automatically jumps to the solution space as opposed to the foundation space.
• 7:43
  SPEAKER_02: And the key point that I was like, you know, unbelievable says, thank you for actually asking the right questions, which it just even got us to the point of even, you know, these these discussions and these seminars was centered around, okay, a methodical thinker who actually can break this down and really float into the foundation space first and what needs to be there in terms of what can be built upon.
• 8:09
  SPEAKER_02: And I like it.
• 8:10
  SPEAKER_02: This this slide is an eye chart, but if you take a minute to just simply just look through it and look in terms of a nice, easy breakdown of where, and I think this is what you should be seeing it from an external perspective is where do I fit within this matrix?
• 8:28
  SPEAKER_02: And I could fit you know in each location, maybe some as further or maybe not as um mature, but where do I fit in this?
• 8:36
  SPEAKER_02: And what can I do to really kind of you know break these foundations down in terms of where to start?
• 8:42
  SPEAKER_02: And I think that context of you asking that first question was the most intriguing thing for me to say, yes, this is a huge problem.
• 8:50
  SPEAKER_02: People don't know where to start, and yet we've been at this for how many years now, decades, and yet we're still providing information um agents and chatbots and all this kind of stuff out in the market that just are not hitting the mark yet because they're bringing back way too much information or not the right information or missing the intent altogether and/or not understanding the conversation, right?
• 9:14
  SPEAKER_02: So it's super important.
• 9:15
  SPEAKER_02: But anyway, I'm I mean I diverge a bit in terms of bringing that in, but I love that historical context because it sets the stage for the foundations that you you know that we're putting together.
• 9:25
  SPEAKER_00: Absolutely.
• 9:26
  SPEAKER_00: And yes, that was a really interesting conversation because uh most of the time when I asked those questions, people just don't even know where to how to answer it or don't really think about it, or they they think that the AI is going to solve those problems, or you know, uh they're going to train their their their ML uh you know using some source of information, but they're they're very fuzzy about it.
• 9:50
  SPEAKER_00: And the point here is that it really does need to be very intentional and very structured in terms of you, you know, you can you you need the knowledge, you need the answers, you need the details.
• 10:01
  SPEAKER_00: And uh a machine learning uh algorithm AI by itself, it can do it in certain instances if you have large amounts of data.
• 10:08
  SPEAKER_00: So if you said it's like uh machine translation, if I have a lot of uh examples, I can translate from one language to another.
• 10:16
  SPEAKER_00: Uh I can translate questions to answers, right?
• 10:20
  SPEAKER_00: If I have enough data, but that's not necessarily reliable.
• 10:23
  SPEAKER_00: And if you're in a regulated industry or you have specific, or if you don't have all that data, I mean you'd need a large amount of data.
• 10:30
  SPEAKER_00: So the point here is that there's you can apply uh AI and machine learning and and various algorithms to improving um the ability and the efficiency of building a foundation, but you still need to be intentional and have human judgment around around that foundation.
• 10:46
  SPEAKER_00: Um and and again, I think uh what's important here is to is to imagine a future where we will be just talking to these virtual assistants day after day.
• 10:58
  SPEAKER_00: We are going to be interacting with them uh just as naturally as we interact with with um our colleagues.
• 11:05
  SPEAKER_00: And in my book, in the first chapter, it talks about um this gentleman, Alan Perkins, and how he goes about his day.
• 11:12
  SPEAKER_00: And his day starts with interacting with virtual assistants, his own virtual assistant that knows his preferences, will renegotiate for him, will do the stuff for him, and with the other assistants and the other virtual uh agents at other organizations.
• 11:26
  SPEAKER_00: So he's checking his banking portfolio, his portfolio, financial portfolio, he's checking parts availabilities, he's arranging travel, he's doing all sorts of things just via voice.
• 11:37
  SPEAKER_00: One day we will get there.
• 11:38
  SPEAKER_00: That is the future.
• 11:40
  SPEAKER_00: We know these things are really bad today, but they will get there.
• 11:44
  SPEAKER_00: And the organizations that are doing the right work today to build a foundation are going to be the ones that will be able to take advantage of that.
• 11:52
  SPEAKER_00: And for others who are not, it's going to be existential.
• 11:55
  SPEAKER_00: I mean, they're gonna find that they're that they're gonna be blindsided by competitors and and being able to provide these capabilities.
• 12:02
  SPEAKER_00: So it's really important to start thinking about what that foundation is.
• 12:05
  SPEAKER_00: Now we look at Omnichannel, it's really looking across devices, it's looking across channels.
• 12:11
  SPEAKER_00: You know, you're trying to maintain context, uh, even if you're switching devices or you're starting a process in one place and and and finishing it in another, starting on your your phone, finishing on the website, talking to an agent, and then shifting back to uh uh uh an app.
• 12:30
  SPEAKER_00: So there's lots of different things, but it's important for the virtual assistants to be able to provide that consistent messaging, uh, to be independent of channels, to be able to escalate to that human agent.
• 12:45
  SPEAKER_00: You know, again, this is really foundational work that has to be done.
• 12:49
  SPEAKER_00: And when you think about messaging, messaging uh should be a part of your content operations and your content publishing so that you can publish once and consume anywhere.
• 13:00
  SPEAKER_00: And there are some organizations that have been working on this for years, and most of the time I see a lot of fragmentation in these AI initiatives where people are building separate groups to do AI content, and that's just not the way to think about it.
• 13:15
  SPEAKER_00: And it leads to knowledge fragmentation.
• 13:17
  SPEAKER_00: You should really be looking across your information ecosystem and looking at all the content they're publishing and start to think about ways of standardizing and normalizing that and componentizing.
• 13:28
  SPEAKER_00: Again, we talked about picking up from one device.
• 13:30
  SPEAKER_00: I have a uh starting, maybe starting on a smartphone, chatting with customer support, not losing that transaction as you go through this.
• 13:39
  SPEAKER_00: And uh what was really interesting, uh this one of these last points here, uh, seamless integration, uh voice and real-time help.
• 13:47
  SPEAKER_00: I was trying to buy a product from a um I won't say what company it was, but it was a uh a place where you buy stuff for your home and uh and and your yard and so on.
• 13:59
  SPEAKER_00: And I was trying to buy um something for my water feature to clear clear off the uh uh the allergy.
• 14:07
  SPEAKER_00: And I got to the website, I couldn't find it, I started a chat.
• 14:12
  SPEAKER_00: The chat was was uh was not um it was not with a live agent, it was a it was a a text uh uh chat bot or it couldn't help me.
• 14:21
  SPEAKER_00: And then it started to ask it, it shifted me to a human, but the human did not have any of that context.
• 14:28
  SPEAKER_00: They did not have the chat that I had started.
• 14:31
  SPEAKER_00: I couldn't even they couldn't even see what I was doing or see what was in my cart or understand.
• 14:36
  SPEAKER_00: I had to copy and paste the product information.
• 14:40
  SPEAKER_00: In fact, I had to do that when I was uh first talking to the bot.
• 14:44
  SPEAKER_00: So I had to copy and paste this stuff multiple times and re uh and give that agent my problem over again.
• 14:52
  SPEAKER_00: So he had no context, he was not able to look at what I was doing.
• 14:55
  SPEAKER_00: That was a really, really bad experience, and that was not what we want.
• 14:59
  SPEAKER_00: We want to be able to maintain that context and that information.
• 15:03
  SPEAKER_00: Uh any uh any thoughts on on your this piece?
• 15:06
  SPEAKER_02: You're you're outlining the issues right now with the difference between reactive and proactive.
• 15:13
  SPEAKER_02: And when we look at it from the perspective of just providing information, it's 100% reactive.
• 15:18
  SPEAKER_02: I can have uh you know an environment that's super, super good and has a lot of really rich data.
• 15:26
  SPEAKER_02: And like you said, it's only as good as the information that it has and that it's indexed and that it knows.
• 15:32
  SPEAKER_02: So I can ask it as many times in 50,000 different ways, meaning being vague or whatever, and it's gonna bring back really good, hopefully, results because it's also got reinforcement learning.
• 15:43
  SPEAKER_02: Now, here's the problem.
• 15:44
  SPEAKER_02: Like you said, what if it doesn't have that information?
• 15:46
  SPEAKER_02: The minute it doesn't have that information and it fails, what does it do?
• 15:50
  SPEAKER_02: It then escalates.
• 15:51
  SPEAKER_02: And the only thing we have now to do is to escalate to human because humans are proactive and they can actually free think for what you need.
• 15:59
  SPEAKER_02: Even when they don't have the context, it's a hell of a lot better if they do have the context.
• 16:03
  SPEAKER_02: But now what we're doing is we're moving from this ability of saying, listen, all of the information that I have is just super proactive, or sorry, is it reactive?
• 16:13
  SPEAKER_02: Then I move to a human and I needed to be reactive.
• 16:17
  SPEAKER_02: The difference of having a virtual assistant or you know, in a virtual bot or something and a digital assistant is that reactive to proactive nature.
• 16:27
  SPEAKER_02: So when we think about this in terms of designing the conversations, and we'll get into this a little bit more.
• 16:33
  SPEAKER_02: Do I want to design this for a reactive conversation where I'm just providing information?
• 16:37
  SPEAKER_02: Well, now I'm just duplicating what's on the website, and or maybe some knowledge-based articles, or do I want to design this for a proactive experience where it's now utilizing and thinking and giving me suggestions and helping me through the process and you know, doing that?
• 16:55
  SPEAKER_02: That's a true digital system.
• 16:56
  SPEAKER_02: That is a hell of a lot harder, though, to design that type of piece because to your point, Seth, you've got data, you've got context, right?
• 17:05
  SPEAKER_02: And you have to pull all that together.
• 17:06
  SPEAKER_02: And then you need additional data sources potentially as well to pull all that together, as well as um, you know, seamless integration, like you said in the in the last the uh the almost last point there, you need behavioral information as well.
• 17:20
  SPEAKER_02: And that behavioral information isn't necessarily collected through, let's say, a textual bot.
• 17:25
  SPEAKER_02: So, how do you collect the behaviors of somebody that's working through these types of things to gain that collection?
• 17:32
  SPEAKER_02: Because, for instance, you know, culturally as well as uh demographically, we have different needs through different channels to communicate and actually solve and provide problems and actually get to the point that we need to.
• 17:45
  SPEAKER_02: So my behaviors are going to be different as a digital immigrant to a digital native that's coming in saying, Oh, I know I only do things through social.
• 17:54
  SPEAKER_02: Well, that's just a different channel.
• 17:55
  SPEAKER_02: Well, can they do everything through social?
• 17:57
  SPEAKER_02: And can I gather all that non-verbal data to actually help understand behavioral context?
• 18:03
  SPEAKER_02: So we're moving from this continue, not only information-wise, but going from a reactive scenario and providing all that rich data.
• 18:11
  SPEAKER_02: But now let's add on top of it the behavioral data and behavioral science to get to a more reactive and proactive type of an assistant.
• 18:18
  SPEAKER_02: And that's where we get to the point of view, you know, like this next generation of where we need to be.
• 18:23
  SPEAKER_00: Absolutely.
• 18:24
  SPEAKER_00: No, it totally makes sense.
• 18:26
  SPEAKER_00: And I think one of the things that we need to do is we need to monitor the digital body language of the user, right?
• 18:32
  SPEAKER_00: What are they actually doing across these different channels and devices?
• 18:36
  SPEAKER_00: And we need to model the use cases, the scenarios, and the actual tasks and then see how well we can support those.
• 18:45
  SPEAKER_00: Organizations should be building libraries of use cases that become the source of truth for really any of the systems.
• 18:52
  SPEAKER_00: And they and and they can come from job descriptions, they can come from there's lots of different things internally and externally.
• 18:59
  SPEAKER_00: But but the organization should be building you know customer uh-facing use cases as well as internal use cases that they can support.
• 19:08
  SPEAKER_00: Because again, it's an ecosystem, and you can't do these things uh seamlessly externally if you have you know uh lots of friction internally.
• 19:17
  SPEAKER_00: We need to improve those internal processes, and that's where when we see these disconnects, that's where some of that uh happens.
• 19:25
  SPEAKER_00: We do have a quick question here, and um, I I was going to we could we could uh just start touching on this.
• 19:31
  SPEAKER_00: Um, what business skills do we need to develop or hire internally to evolve further with increasing functionality?
• 19:39
  SPEAKER_00: What guidance do you have to prepare internal stakeholders for the scale and duration of the journey?
• 19:45
  SPEAKER_00: That's a there's a lot in there.
• 19:46
  SPEAKER_00: Um uh what do you why don't you give your take on that, Chris, and then I'll I'll chime in and then uh we'll we'll continue with the uh uh discussion.
• 19:56
  SPEAKER_02: Yeah, I think I think like you said, there's there's so much packed into that one question.
• 20:00
  SPEAKER_02: And I think you could cover that in terms of the skill sets needed around, you know, I I always, you know, and and AWS, this is actually one of the things that we do, I think really well is do, you know, is work backwards, right?
• 20:13
  SPEAKER_02: So what is the end goal?
• 20:15
  SPEAKER_02: Now, I would say that the end goal is never the end state of what this needs to be because it should be an ever-learning process and evolving, right?
• 20:22
  SPEAKER_02: It's just a the natural.
• 20:24
  SPEAKER_02: So what's the first hypothesis, right?
• 20:27
  SPEAKER_02: So that first hypothesis should be centered around, you know, when we look at it from, let's say, a um a you know, public relations in an FAQ document, this is how we actually do our working backwards session, and we we say, listen, here's what this thing needs to be able to do first.
• 20:44
  SPEAKER_02: And we give it the you know, the description as well as you know, the FAQs, and then we you know we we give that out to everybody in order to poke holes in it.
• 20:52
  SPEAKER_02: Part of that is that we're also looking for additional data cues around what we didn't think about.
• 20:59
  SPEAKER_02: And put and then so we can actually uh really get a holistic perspective of what needs to be prioritized first in order to actually accomplish that first hypothesis goal.
• 21:09
  SPEAKER_02: And so, in terms of the skills needed, it's just a matter of thinking backwards, but also understanding where the data is and what data is not there, that you're gonna be able to facilitate this first piece of, let's say, a more reactive type of a chat bot or IBRs, whatever it is.
• 21:26
  SPEAKER_02: And then we'll get into more of this later, but um, this gets into more of the operational deployment types of things of how to not shoot ourselves in the feet centered around trying to do too much too fast.
• 21:38
  SPEAKER_02: Because the the idea here is, and this is what we also get into is a mismatch of utilizing folks like business analysts to derive our customer or sorry, our conversational experience.
• 21:52
  SPEAKER_02: Much like we did back, you know, when I went to school, I did psychology and human factors.
• 21:57
  SPEAKER_02: And those human factors were really all about UX.
• 22:00
  SPEAKER_02: We didn't call it UX at the time because it wasn't the user experience, we didn't call user experience.
• 22:04
  SPEAKER_02: Well, now we're getting into the derivatives of user experience of behavioral analytics, fit and finish, right?
• 22:12
  SPEAKER_02: In terms of the actual user, the end user Chrome and how things are designed.
• 22:18
  SPEAKER_02: But the colloquial terms still apply where user experience is really my interaction with something else.
• 22:25
  SPEAKER_02: So now when we look at that, we're almost needing to get people that can actually truly understand the goals of what this needs to be and design the conversation in that way.
• 22:36
  SPEAKER_02: And that's a really hard thing to think about because I can't add a binary thinker that's a maybe a business analyst and say, hey, I need you to design my conversation.
• 22:44
  SPEAKER_02: I need somebody that can get in and really think about it from that perspective of when I interact with my customers, what is it that I'm actually really trying to solve?
• 22:53
  SPEAKER_02: And how can I speak to them in a way that's going to help them solve that?
• 22:57
  SPEAKER_02: Knowing that there's, like I said, there's a demographic, there's the um you know cultural needs, as well as now just compliance needs that I need to need to make, you know, follow in.
• 23:08
  SPEAKER_02: So that's you know, just just a very, very high level tip of the iceberg type of a thinking, because what should happen with this is indicative of what the end all be all goal is.
• 23:19
  SPEAKER_02: I should also be mapping that back to who are the key personas that I'm actually tying into, what are their needs?
• 23:24
  SPEAKER_02: How do I address those and actually get to almost almost a psychological profile of who I'm going to interact with in terms of my customer and customer types, and then design the conversation centered around what that looks like.
• 23:38
  SPEAKER_02: But where I see the biggest mismatch in terms of skills is we we do this all the time where we take, you know, folks that are actually looking for the data and trying to find the data and come up with the data responses, and we naturally take them over and apply them towards what the conversational experience should be.
• 23:54
  SPEAKER_02: And they're not, they don't think that way.
• 23:56
  SPEAKER_02: So what we need is is to is to say, listen, what is it that we need to do and design the conversation in a way to say, well, what information do I need?
• 24:04
  SPEAKER_02: Let's go find the information after I can ask those initial questions, you know.
• 24:07
  SPEAKER_02: So it's it's not a um an easy answer or an easy question to answer, but it's one of those things like if you're applying the folks that are actually the data scientists and the data, you know, um researchers and all that kind of stuff, and you're letting them design your conversation, you're gonna miss 100% of the time.
• 24:25
  SPEAKER_00: You know, I I I agree that it is a different way of thinking.
• 24:29
  SPEAKER_00: And I would also add to that that we need very strong skills in information architecture and uh content uh operations because really content and knowledge drive these things, right?
• 24:42
  SPEAKER_00: We're if we're in a regulated environment, we're not gonna just turn our our you know uh algorithm loose on lots of information and see what it comes up with, right?
• 24:51
  SPEAKER_00: We have to be able to be very clear and specific with what those instructions are, and and that requires that we intentionally curate this training data, right?
• 25:02
  SPEAKER_00: Training data is is content, it's it's knowledge, it's information that the customer needs.
• 25:08
  SPEAKER_00: That's the only way that the bot will understand it is if we train it, we give it that information, it doesn't get it out of thin air.
• 25:14
  SPEAKER_00: And so the idea of expertise and skills around knowledge processes, around information architecture, around metadata, and those conversational skills.
• 25:26
  SPEAKER_00: So we have to have that back-end system uh structured correctly, we have to have the information in the right format, and then we we have that, we have to design that retrieval mechanism and that and even and and whether it's proactive, you're listening to a conversation, and then saying, oh, wait a minute, they mentioned these words, let me surface this for the agent, for the human agent, to make them more effective, or give them some documents that they don't have to start looking for, they'll have it at their fingertips based on the conversation.
• 25:55
  SPEAKER_00: Or it can monitor sentiment and say, wow, this this customer seems really upset.
• 26:01
  SPEAKER_00: Let me let me uh cue the uh agent to to use this uh approach to kind of calm them down or de-escalate it, right?
• 26:09
  SPEAKER_00: So there's a lot of things that can be done in being more proactive, and and the skills that are necessary are those that you mentioned, the conversational design, the behavioral understanding, the use cases, scenarios, uh, and and and really that understanding of how that information needs to be structured, and of course going out and doing it.
• 26:29
  SPEAKER_00: Now, the other part of this question was how do you prepare people for the long journey?
• 26:33
  SPEAKER_00: And as you said, Chris, it's never done, right?
• 26:36
  SPEAKER_00: When are you done with you know with with content or knowledge or you know, tooling, whatever it is, when you're when you're done with sales, right?
• 26:44
  SPEAKER_00: When you're done with manufacturing, when you're done with it, it's it's a constantly evolving environment.
• 26:50
  SPEAKER_00: And what organizations are missing right now is the governance and the structure around change management and and uh and the authority and the mechanisms for changing things because there's too much sprawl.
• 27:03
  SPEAKER_00: This is leading to knowledge fragmentation, not knowledge consolidation.
• 27:08
  SPEAKER_00: And so these are large-scale information management problems and and organizations are going to start running into these big, big information management problems and find that the direction they went in with you know, uh dispersing it throughout these bot mechanisms really is not going to work.
• 27:24
  SPEAKER_00: So let me ask you a question.
• 27:26
  SPEAKER_00: I'm sorry, go ahead.
• 27:27
  SPEAKER_00: Sure.
• 27:27
  SPEAKER_02: Let me ask you something.
• 27:28
  SPEAKER_02: So, how of the of the companies that you consult with, how many of them, going back to the information um architecture piece, how many of them do you find that actually even have an ontology or a taxonomy for their data?
• 27:41
  SPEAKER_02: I'll bet it's significantly low.
• 27:44
  SPEAKER_00: Well, it it and and it really depends on how you're defining that because they have something, but they don't have something that is very useful.
• 27:51
  SPEAKER_00: It's usually fragmented, right?
• 27:53
  SPEAKER_00: There's a lot of taxonomies.
• 27:55
  SPEAKER_00: There's no formal structure that puts these into an ontology.
• 27:59
  SPEAKER_00: And for those of you on the phone not familiar with that term, you know, we we build mechanisms to organize information.
• 28:04
  SPEAKER_00: Those are taxonomies or hierarchies, parent, child, whole part relationships.
• 28:09
  SPEAKER_00: There's also uh controlled vocabularies, there's there's uh thesor structures, these are things that help us manage terminology.
• 28:16
  SPEAKER_00: And then when we have multiple sets of those taxonomies, because we don't build one grand galactic Uber taxonomy, uh, we build multiple taxonomies for products, for services, for solutions, for customer types, for document types, all those things.
• 28:31
  SPEAKER_00: Then when we relate them, when we say, here is the solution to this problem, I have a solutions uh uh uh hierarchy, and I have a problems hierarchy, here's the solution to this problem, or here's the here are the services that go to this product, or here are the risks that are in this region, or here are the products that this um uh that this provider would like, or or here are here are things that you're you're building knowledge relationships between the different concepts.
• 28:59
  SPEAKER_00: And and what that does is it allows you to get a much greater control, much greater uh visibility to all of your data, because there's knowledge graphs and there's uh uh there's graph data, and those things together will give you a very comprehensive picture.
• 29:16
  SPEAKER_00: But organizations are very early in that learning curve.
• 29:19
  SPEAKER_00: A lot of those are science projects, not producing ROI.
• 29:23
  SPEAKER_00: What you really need to think about is you know, how do we apply uh our organizing principles to solving a specific problem, right?
• 29:32
  SPEAKER_00: To doing personalization, to doing all of those uh the those other things that you want to do, not just building the ontology for the sake of having a knowledge graph, but being able to apply it to specific problems.
• 29:44
  SPEAKER_00: But you're right, you know, most organizations do not have this under control.
• 29:48
  SPEAKER_00: You know, very few have installed the change management, the metrics, and the governance to do so.
• 29:54
  SPEAKER_02: This is the you know, kind of the overall goal as well, you know, how to get started is how many have already tried and are having to rethink, reinvent, reinovate against what they've already done, where they put a huge initial investment.
• 30:09
  SPEAKER_02: And now what does that look like?
• 30:10
  SPEAKER_02: You know, maybe moving over to let's say utilizing some of these you know new neural nets and all that kind of stuff to actually come up with some of these deep learning algorithms.
• 30:19
  SPEAKER_02: You know what I mean?
• 30:20
  SPEAKER_02: Because I'm sure there's space in there too.
• 30:22
  SPEAKER_00: But yeah, I think that that is a good point.
• 30:24
  SPEAKER_00: It's like how many times have you restarted this?
• 30:26
  SPEAKER_00: And everybody, it's it's a journey for everyone, and it is a long journey.
• 30:30
  SPEAKER_00: And I think preparing stakeholders for that has to be about quick wins and and showing ROI early in the process.
• 30:38
  SPEAKER_00: And I'll show you that we need to really be careful about our use cases uh and and our tasks that we're trying to enable in order to show that.
• 30:46
  SPEAKER_00: But this is an interesting one.
• 30:47
  SPEAKER_00: Uh, this is the Gartner hype cycle, and what's really, really funny, and this is this I don't get at all, knowledge managed for customer service is in the trough of disillusionment.
• 30:59
  SPEAKER_00: Like, why?
• 31:00
  SPEAKER_00: And you know, we we just have not seen companies make significant investment in it.
• 31:06
  SPEAKER_00: And it's one of those things that makes me berserk because it's like what an obvious way to get ROI is to build knowledge for your call center.
• 31:15
  SPEAKER_00: And there's there's lots of metrics out there that shows that, you know, we had one that shows 50% reduction in in time per incident and and uh first call resolution, and you know, we say we've seen projects save millions of dollars doing this type of stuff.
• 31:31
  SPEAKER_00: So this is kind of really surprising.
• 31:33
  SPEAKER_00: And and the fact is, a lot of the AI projects and the and the and the bot projects are taking funding and resources from the basics, from the investment in knowledge architecture.
• 31:45
  SPEAKER_00: And that is is is really a shame because that money should be invested in that foundation, and then building on that foundation uh is where you're going to get a lot of value.
• 31:56
  SPEAKER_00: And then the other virtual customer assistance are are in the pro uh uh trough of disillusionment.
• 32:02
  SPEAKER_00: Chris, do you want to uh add some color commentary on this one?
• 32:06
  SPEAKER_02: I think you nailed it earlier where you know folks that are that have created a or invested a lot of um you know dollars into knowledge management without a goal, right?
• 32:17
  SPEAKER_02: It's one thing to have it a goal to collect all the information because you know most companies that we we you know interact with have thousands of data sources, and whether it's structured data or unstructured data, you know, coalescing that together for let's say a specific you know goal or objective is one thing.
• 32:35
  SPEAKER_02: Um, but if the overall goal and objective is just to gather the data, great.
• 32:41
  SPEAKER_02: You know, now what?
• 32:43
  SPEAKER_02: And so what we also see here is that it looks like it's parallel work streams, right?
• 32:48
  SPEAKER_02: Where we actually see you know knowledge management for customer services happening and then virtual customer assistance, yeah, there's such a great observation, yeah, right?
• 32:57
  SPEAKER_02: Isn't that interesting?
• 32:59
  SPEAKER_02: Yeah, so it's just it's kind of missing really, you know, we're looking the the glue that needs to come together to say, well, the the reason why we're pulling all this together is we're trying to make it functional just like an extensible API for all these other disparate systems to call and to pull back.
• 33:15
  SPEAKER_02: And I think to your point, it's like what we used to get into when we pivoted from you know transactional data sources to like BI structures.
• 33:25
  SPEAKER_02: And then everybody went and said, Oh, kind of left the transactional systems there and said, Yeah, that's great, we don't need that anymore because everything that the new you know gold standard is BI and snowflake architectures and all that kind of stuff.
• 33:38
  SPEAKER_02: And then that moved to columnar databases, and then that moved to graph systems, and then that moved to data lakes, but now we have all these independent systems and nothing again pulling them together where it should be the feature and function of this thing is to do X, and that's what it's there for, and that's what I needed to do.
• 33:55
  SPEAKER_02: And then we pull all these other pieces together, you know what I mean?
• 33:57
  SPEAKER_02: So this looks like parallel work streams.
• 33:59
  SPEAKER_00: Uh yeah, I think that's a great observation.
• 34:01
  SPEAKER_00: I didn't really think about the fact that they're even defining us separately, and they really should be part of the same uh work stream, the sort of part of the same processes.
• 34:10
  SPEAKER_00: Um, let's go ahead and jump into some of our questions that we were talking about before.
• 34:16
  SPEAKER_00: So here's a bunch of stuff on this slide.
• 34:18
  SPEAKER_00: Voice and text, how are they different?
• 34:19
  SPEAKER_00: How are they the same?
• 34:20
  SPEAKER_00: And we can kind of talk about any of these points.
• 34:23
  SPEAKER_00: Uh, interactions and workflows are interesting because uh voice is very different in having a conversation.
• 34:29
  SPEAKER_00: Uh, regional, cultural, generational gaps, what do we need to support?
• 34:33
  SPEAKER_00: And how do we capture that additional information that is lost in voice?
• 34:39
  SPEAKER_00: So maybe I'll just let you riff on um any of these topics for a little bit, and then we can add uh discussion.
• 34:44
  SPEAKER_02: It's it's funny how the last uh question and the first question are really, really functionally tied together.
• 34:50
  SPEAKER_02: We see this all the time.
• 34:51
  SPEAKER_02: I can't, you know, just because I can see um the only thing I can see verbally and non-verbally, let's say through text, is if you capitalize something, right?
• 34:59
  SPEAKER_02: Or you use cultural emoji.
• 35:02
  SPEAKER_02: Yeah, exactly.
• 35:02
  SPEAKER_02: Cultural innuendos, but I can't.
• 35:04
  SPEAKER_00: Emojis are not under well understood either.
• 35:06
  SPEAKER_00: There's a whole article uh that the same emoji has vastly different meanings.
• 35:11
  SPEAKER_00: Go ahead, keep going.
• 35:12
  SPEAKER_02: Yeah, no, you're you're spot on.
• 35:13
  SPEAKER_02: I was gonna say the same thing, but I can't quantify or qualify what even a the hell an emoji means and all the subcultural uh innuendos that people can actually highlight within a textual conversation, right?
• 35:25
  SPEAKER_02: But this just gets into those notions around pulling out the demographic piece of this and understanding how younger generations will communicate and want to solve problems over, let's say, digital immigrants.
• 35:38
  SPEAKER_02: So it's but but that the nonverbal piece is super, super important, especially when you're talking about a com uh an actual audio conversation where I can really get a ton of information out of what's not being said.
• 35:52
  SPEAKER_02: The sentiment, speed, the interruptions, the you know, speakover, you know, what does that all look like?
• 35:59
  SPEAKER_02: And that can actually then set the ability to score it correctly so I can score the outcome and understand what my next action needs to be, whether that needs to be an automated action or that needs to be a manual action, that's where we get into those scenarios too.
• 36:13
  SPEAKER_02: Around, well, if I see, you know, if I can alert and notify that the sentiment is super poor on this call, what alerts do I actually notify automatically to actually barge in or do some things in a call center scenario?
• 36:28
  SPEAKER_02: What does that look like in terms of highlighting and notifying somebody centered around the downstream effects of a poor interaction centered around brand and loyalty?
• 36:38
  SPEAKER_02: You know what I mean?
• 36:39
  SPEAKER_02: Things like that.
• 36:40
  SPEAKER_02: And how do I actually see that score?
• 36:42
  SPEAKER_02: And I think that's the key is first of all, we get the data, we test it, we put the hypothesis, and then we actually put some things there to actually give a grade on what that outcome is very semantic to each organization, but what that outcome needed to be, and if we were able to get the desired outcome, and if that desired outcome ended with positive sentiment and or you know was able to solve the customer's problem.
• 37:05
  SPEAKER_02: Because if we see them call back in five minutes, right, or the next day on the same topic, we haven't done anything.
• 37:12
  SPEAKER_02: Right.
• 37:12
  SPEAKER_02: Right?
• 37:13
  SPEAKER_02: Yeah, yeah, yeah.
• 37:14
  SPEAKER_00: Totally agree.
• 37:15
  SPEAKER_00: I think that the you know what you're getting at is is absolutely you know uh measuring the behaviors, measuring the outcomes, measuring the impact of what we're doing, and what we can uh enable, you know, I believe I'm a big believer in building helper bots before you do anything customer-facing.
• 37:33
  SPEAKER_00: You can do stuff customer-facing that's very transactional and very uh you know straightforward.
• 37:40
  SPEAKER_00: But I I think you can learn a lot by building bots that will help your agents.
• 37:44
  SPEAKER_00: You can train your agents to how to use them, and then you can learn from that.
• 37:48
  SPEAKER_00: You can learn what other types of questions are coming up and what you've missed, and missing content and evolving the content and so on.
• 37:55
  SPEAKER_00: And and part of that is letting that that virtual assistant, that helper bot, listen in on the conversations in real time and being able to do the text analytics real time to be to be able to say, wait a minute, they're talking about this product and they're talking about this specific information.
• 38:16
  SPEAKER_00: Let me get that uh ready for the agent and take that recommendation.
• 38:20
  SPEAKER_00: Or uh, as I had mentioned before, you know, because this is getting heated, what can I prompt the agent with?
• 38:26
  SPEAKER_00: So prompting the agent during those conversations or escalating during those conversations is a really uh critical piece.
• 38:33
  SPEAKER_00: And and our virtual assistant framework, our omnichannel virtual assistant framework should be able to handle that, right?
• 38:40
  SPEAKER_00: Because it wants you want to monitor.
• 38:42
  SPEAKER_00: You miss tone, you miss all that nonverbal um communication or that that uh that uh emotional communication that's part of verbal, right?
• 38:51
  SPEAKER_02: See now you're getting into a scenario too where now we have bots that are specific for the workload, right?
• 38:57
  SPEAKER_02: Or the approach or the job or whatever that is.
• 39:00
  SPEAKER_02: And the the basis too is in a lot of cases, from a textual-based environment, you don't have the ability.
• 39:08
  SPEAKER_02: Um so you know, let me back up.
• 39:11
  SPEAKER_02: So when we when we talk about speech recognition, there's two flavors, right?
• 39:14
  SPEAKER_02: There's long form speech recognition and short form.
• 39:17
  SPEAKER_02: Long form speech recognition is non-structured conversation like what you and I are doing now, right?
• 39:22
  SPEAKER_02: And we can pull out um with high accuracy the speech and the words that I'm speaking, get that into text, and the words that you're speaking, and also separate that out, look at sentiment, look at keywords and all that kind of stuff that is in the speech recognition, and then utilize natural language processing to understand keywords and phrases and all those other pieces, right?
• 39:44
  SPEAKER_02: At the same time, um, we have short form conversations where those short forms are intent-based and they're looking for keywords in order to drive a reaction or an action.
• 39:56
  SPEAKER_02: And so indicative of that, we see also the need to pull those two together.
• 40:00
  SPEAKER_02: Because in that scenario, when you're having a conversation on the phone and you have a helper bot, that is literally that scenario of we're having a conversation, you're trying to solve my problems, I'm the customer, and then that bot is helping you create suggestions, next best action, whatever that is, right?
• 40:16
  SPEAKER_00: Like a real-time coach.
• 40:18
  SPEAKER_02: Yeah, exactly.
• 40:19
  SPEAKER_02: But I have scenarios too where what how do we take that in the form of the utilizing the helper bot in a patient and clinician scenario, too, where they're speaking, and now I can utilize keywords to actually drive action into the short form where I'm capturing the overall long form conversation as well.
• 40:40
  SPEAKER_02: So it's it's also a matter of understanding the ASR frameworks and what needs to happen in the form of the conversation and what I can drive from that perspective of just listening to the words and giving me suggestions and listening to the words and having keywords that can actually drive automated or manual actions into other systems or whatever that is, right?
• 40:59
  SPEAKER_02: Do something.
• 41:01
  SPEAKER_02: So when I say, hey, maybe Sharon is our dirt, our you know, digital assistant, and we're talking about, and I say Sharon, that keyword highlights that intent, and that intent framework is to build us, let's say, a schedule for this time in this period at this, you know, at this juncture, whatever that is, right?
• 41:18
  SPEAKER_02: Things like that.
• 41:18
  SPEAKER_02: So there's also a mixing of that.
• 41:20
  SPEAKER_02: And now we're also getting into the bots that are helper bots that are activity-based and that are also job-based, and that are also you know based on the right type of data.
• 41:30
  SPEAKER_02: Because I'm not going to give customer the same data I'm going to give you as an as an agent.
• 41:35
  SPEAKER_00: Right, exactly.
• 41:36
  SPEAKER_00: And I think that uh that that observation that there will be these different types of bots that will be specialized, and you will have a concierge bot, and then that information has to be managed across all of those systems.
• 41:48
  SPEAKER_00: That's a critical uh challenge.
• 41:50
  SPEAKER_00: Yeah.
• 41:50
  SPEAKER_00: Um, and the workflows, you meant we talked about the workflows.
• 41:53
  SPEAKER_00: The workflows can be very different.
• 41:55
  SPEAKER_00: And you know, when people are having conversations, they'll switch topics.
• 41:59
  SPEAKER_00: They'll they'll go in and something that'll come to mind.
• 42:02
  SPEAKER_00: And and if we're doing this with a virtual system, we have to be able to uh handle that.
• 42:08
  SPEAKER_00: Uh let's see what else do we have.
• 42:09
  SPEAKER_00: I think we pretty much covered uh a lot of this stuff on this particular slide.
• 42:15
  SPEAKER_00: Let me go ahead and uh share another, and then we can use that to ground our conversation.
• 42:23
  SPEAKER_00: So let me do this.
• 42:25
  SPEAKER_00: So this was just a a look at you know the different uh uh the the spectrum of of uh structure and definition and uh and uh ambiguous processes and defined processes versus subjective advice, right?
• 42:42
  SPEAKER_00: So when we start looking at this, you know, we can do things that are very uh that are on the far left much easier, right?
• 42:48
  SPEAKER_00: We use RPA, and I noticed that was somewhere in the trough of disillusioned as well, but people got value from RPA.
• 42:54
  SPEAKER_00: Um being able to support transactions, being able to do other things, and then we start getting to more subjective uh advice.
• 43:02
  SPEAKER_00: And what we need to do is think about how we can uh support somebody providing that.
• 43:08
  SPEAKER_00: I you know, there are robo advisors, but a lot of customers don't necessarily want to use them.
• 43:14
  SPEAKER_00: But if you think about that as sort of sort of a high complexity of the domain of the knowledge and high complexity of the task, we can also think of this on this continuum task and dialogue complexity versus domain complexity, right?
• 43:28
  SPEAKER_00: So if we're life sciences, you know, and and we're trying to troubleshoot a PCR machine that's going to be uh on a complex domain and a fairly complex task.
• 43:38
  SPEAKER_00: And so we have transaction support, knowledge treval, and then judgment based.
• 43:42
  SPEAKER_00: And so we can start to put this on this continuum, you know, where you could have simple processes and simple uh domains for doing, you know, where's my stuff or uh inquiry, and then we can start moving up this curve to say if I have a really complex domain, even if it's a transaction, it might.
• 44:00
  SPEAKER_00: require more uh sophistication but then we're getting into these really uh sophisticated areas and this is where you don't want to start right don't start with high complexity high complexity tasks you really want to cut your teeth on things that are straightforward and and easily modeled but we can build these different types of bots we can build configuration bots we can build transaction bots and we can build helper bots across the spectrum and that's why I think there's so much value because you you're you're you're you're structuring your knowledge and curating your knowledge for helper bots which will help everything right it will it makes that information more more uh more findable it makes that information more usable uh you can go to my tagline more valuable but but the point here is that this is a way to begin that process and a way of looking at your knowledge across your organization because really you should be building helper bots so that people can access their systems conversationally when we say cognitive assistant or cognitive computing we're talking about reducing the cognitive load on the human making it easier for people to do their jobs.
• 45:12
  SPEAKER_00: So when you start looking at that and you start thinking about you know if you're beginning this journey where do you want to focus?
• 45:19
  SPEAKER_00: If you're along this journey, you know what are the roadblocks how do they how can they be mitigated and then when we start thinking about scaling and and really getting to operationalization and I'm gonna stop my share here but um so we can talk about this and what do we really need to do to to to get that at scale to really refine the stuff.
• 45:38
  SPEAKER_02: So I think what we do is we we put together I think the easiest way to do this honestly is to put together like you like you said don't start you know like trying to solve you know trying to boil the ocean with doing everything.
• 45:51
  SPEAKER_00: It's not current cancer which was MD Anderson's Lockin project that failed spectacularly after$78 million.
• 45:58
  SPEAKER_00: Right?
• 45:58
  SPEAKER_02: Yep exactly but I think to your point each one of those sections you know can actually start with what's the um what's the desired state of that first piece need to be and then let's work backwards from there.
• 46:12
  SPEAKER_02: Like even even you know make stuff up like what is it that that the industry will say about it?
• 46:17
  SPEAKER_02: What will customers say about it?
• 46:18
  SPEAKER_02: What will our internal executives say about it?
• 46:21
  SPEAKER_02: Put together that description and then fill out as many as much of those those frequently asked questions that you know can be there like who's the target you know persona?
• 46:31
  SPEAKER_02: What does it need to be able to do?
• 46:33
  SPEAKER_02: What question is it going to answer?
• 46:34
  SPEAKER_02: And then just start with a framework like that and let everybody you know get around it to understand the concept and the idea and get your stakeholders to understand that.
• 46:43
  SPEAKER_02: And what I can guarantee will happen is two things.
• 46:45
  SPEAKER_02: First and foremost you'll have a well vetted understanding of what your MVP or MLP right your minimum lovable product is right for that first piece.
• 46:54
  SPEAKER_02: Second you'll get your stakeholders really understanding and get their buy off quick and get that investment in to understand what that first piece needs to solve.
• 47:03
  SPEAKER_02: Because what's going to happen is before you make that investment if you can understand what that first iteration needs to look like, you're gonna know if it's gonna be successful or not automatically absolutely right.
• 47:14
  SPEAKER_02: And then and then go look for the data to put it together and then it'll help you prioritize what that roadmap looks like and everything else.
• 47:20
  SPEAKER_02: And then start it because you're right it's gonna be a network of bots and it's not a one size itself.
• 47:27
  SPEAKER_02: However, what we can do is put this into a hierarchy of understanding where you can actually then bridge context back and forth between them but there may be some that are specific to a this specific work stream and some specific to this because they're applied to the specific persona in which they're helping or providing information.
• 47:46
  SPEAKER_02: So when we do that I think that's the in my mind that's the best way to start is just to understand what's that first minimum thing need to be and work backwards from and I think part of that is what are you instrumented to measure?
• 48:01
  SPEAKER_00: Right?
• 48:02
  SPEAKER_00: What process are you going to impact and pick a high value process that you can see measurable results around and that so you cherry pick your use case and that should be able to define that the provide value and define you know that that uh that future state when you start expanding this but if you can show that ROI on that narrow set of use cases then that is where you can get to support in the funding.
• 48:28
  SPEAKER_00: If you can't show that then a lot of organizations just aren't instrumented they don't track their metrics or they're so distributed but that's a big part of it.
• 48:36
  SPEAKER_00: It's a big part of what we do in terms of helping organizations build metrics driven change management and governance right because that applies to everything and especially these you know one of the things that has to happen is you have to consolidate your efforts.
• 48:50
  SPEAKER_00: You have to say well what are the best practices?
• 48:52
  SPEAKER_00: Lots of experiments lots of platforms great but what's really working well and can we start centralizing that expertise decentralize the execution at the right level but be able to manage the initiatives uh in some in some centralized way so that you're not reinventing the wheel so that you're not further you know uh uh causing more damage right by by further fragmenting stuff I mean though there are millions of dollars spent on AI content and it shouldn't be it should be content and and AI is one of the consumers of that content.
• 49:27
  SPEAKER_02: I think we cover lots yeah what's that and a data generator it's gonna generate a ton of data yeah absolutely so so when we think about uh and let me go ahead and share this I I think we only have a couple minutes left but I think we're answered um questions as we went along uh but uh Mike Ellsworth had a had an interesting observation about um uh about uh uh emojis he said uh we need a tr an emoji translator for old people is what he was saying yeah yeah the Wall Street Journal had a wonderful article on that topic and it was really really funny because it talked about how a smiley face you know has all of this connotation to to one generation versus another so when you think about scale what's that I'm sorry oh well I learned I learned this week that you know in sarcasm a thumbs up can mean you know suck it kind of a thing right so you you never know so yeah it's spot on because I think you're absolutely right what's the context and the sentiment of the emoji and when it's sent yeah anyway keep going sorry yeah the spender so one of the things we need to do is we need to think about this from a factory approach right talk about a bot factory and talk about virtual factory and what that means because you're going to do this at scale is that you have to have reusability of content you really need to leverage those investments you've already made right and and repurpose and standardize your assets so that they can be consumed.
• 50:56
  SPEAKER_00: And you also can standardize many of the design elements of the box and actually abstract that into an ontology.
• 51:02
  SPEAKER_00: We actually have a patent in the works for doing that to to make these more scalable to be able to reuse elements to be able to update multiple downstream bots when you make a change to your product or service or or even part of your conversation.
• 51:16
  SPEAKER_00: The portability is important because you need to refactor the content that and reuse it in in in different systems because you can't get locked into a single vendor and oh everything is pushed into this virtual assistant.
• 51:31
  SPEAKER_00: It has to be portable and we have to prevent that vendor lock-in to to uh substitute new new components at best of read and then when we start doing this we can manage when we start uh scaling it uh we can manage the uh design elements and even snippets of conversation and abstract that into an ontology so there's some really interesting things that can be done uh you can also uh use machine learning and AI to refactor the content so when we talk about this we talk about componentizing breaking content up into pieces people think that's an enormous amount of effort it's a lot of it it's not insignificant but you can use certain machine learning algorithms to do that there are ways to do that and ways to tag it.
• 52:19
  SPEAKER_00: So again we're taking those documents we're taking that information and we're refactoring it so that we can standardize it we can componentize it we can structure it and and tag it from multiple downstream systems and then we can use it in all of these different contexts.
• 52:34
  SPEAKER_00: So that's kind of the way to think about knowledge and the way to think about a knowledge architecture.
• 52:41
  SPEAKER_00: And then I think I have one more here that is just talking about oh yeah this is just getting into more detail uh you know we we have example deliverables from every one of these work streams but this is the way you can do this and scale it and and bring it to market and reuse it.
• 52:57
  SPEAKER_00: This is powerful this is awesome in my book by the way uh and I think that was it you know I don't think um I if there's any other questions you have we're right a couple of minutes from the top of the hour.
• 53:08
  SPEAKER_02: You know um in closing too from my side I think Seth the key here is is if you go back to your your slide before you know a lot of folks will just help you know ask question well how do I start right and you know understanding and and and even having you know somebody like you that can actually have them guide you know guide them through the process of what this looks like.
• 53:28
  SPEAKER_02: The problem with with off the shelf products is they're predefined.
• 53:32
  SPEAKER_02: And I'd be suspect of a product that comes off the shelf that doesn't have the ability to give me the extensibility I need to support my business.
• 53:40
  SPEAKER_02: I don't want them to define my business for me.
• 53:43
  SPEAKER_02: I want to be able to utilize their their tool right it's a tool and then utilize what I can do to extend that to make this better that kind of thing.
• 53:52
  SPEAKER_02: And so you know I think just having that that that help to walk through step by step is is something you guys are hugely valuable at providing.
• 54:01
  SPEAKER_00: Absolutely and I and I have to say that um one of our uh customers was outsourcing this to a vendor and I said you can't do this you need to own this capability you this can't be a black box this is going to be a source of competitive advantage guess what you can't or outsource your competitive advantage right you have to own your competitive advantage and um and and so that's a really critical piece is that these are skills that need to be built.
• 54:31
  SPEAKER_00: You're gonna find in the next five years companies are going to be caught flat footed on this there's going to be some fantastic uh virtual assistants that are going to provide better customer service lower costs greater accessibility efficiencies and and and I I I'll tell the quick story of a publisher years ago who um was having uh found that the customer uh a a a competitor was leading them to market in their textbook uh sales for K through 12 and uh six months which was very significant lost a lot of business and we found that the uh competitor was using a component repository they're breaking that content into components and then an editor could get a quick um uh outsource thing again uh this you can't outsource your competitive advantage so but what what I was saying is that this organization uh had had said well why don't we have this the CEOs why don't we have this the competitor has it and the team said well this has been the project we've been trying to get funded for the last three years and keeps getting kicked down the road okay well let's do it now how long is it gonna take it's gonna take a couple of years well guess what it was too late they lost that business they lost it it was too late that's the type of thing that's gonna come up and sneak up and bite companies because they are doing digital transformations without regard to knowledge processes.
• 55:54
  SPEAKER_00: I just wrote an article about that it's the missing ingredient to digital transformations it's super important it's super important for this and again customers are being sent down the wrong path by believing that this can all be handled algorithm algorithmically and by the by the box.
• 56:12
  SPEAKER_00: But again it's it's really critical to understand this.
• 56:15
  SPEAKER_02: So there we are uh at the end of our session I think the key this is the first of of a bunch of different sessions that we're putting together too so just to highlight it yes yes we are going to be doing uh so if there's things that you specifically want to cover let us know and here's some additional resources from some of our previous uh uh discussions but I'm really looking forward to putting this series together with you Chris and I we're gonna be diving deep into a lot of these different areas so thank you so much for your time uh thank you to Sharon uh for her um uh keeping the trains running and uh doing all the magic behind the scenes uh so that I can just show up and uh thank you for everybody uh listening and especially thank you Chris you got it pal anytime look forward to working with you on these all right okay guys have a great week next time okay bye now