Seth Earley on setting the groundwork for AI implementation

Next-Gen Technology

On this episode of TELUS International Studios, we're joined by Seth Earley, consultant and award-winning author of The AI Powered Enterprise. Seth has more than twenty years of expertise in data and information architecture, knowledge strategy, search based applications and information findability solutions.

In the conversation, Seth shares advice for business leaders looking to incorporate AI solutions into their support model. Tune in to hear Seth's insights on navigating uncertainty in AI development while keeping customer experience at the forefront, and his take on the biggest challenges when it comes to human-machine partnerships.

To learn more about TELUS International and our digital CX solutions, contact us today.

Click here to read the first chapter of Seth's book, The AI Powered Enterprise.


Intro: TELUS International Studios where customer experience meets digital transformation.

Patrick Haughey (PH): And you are welcome back to TELUS International Studios. I’m your host, Patrick Haughey, and today we have a really interesting conversation for you with Seth Earley, the CEO of Earley Information Science, a consulting firm focused on organizing information for business impact. Seth has a 20 plus year expertise in knowledge strategy, data and information architecture, and search-based applications. He’s also the award winning author of "The AI Powered Enterprise." We get into some pretty interesting questions about AI applications in CX, moving humans up the value chain, and where we see this technology going. Without further ado, we start off today’s interview talking about the ethos behind Seth’s book "The AI Powered Enterprise."

Seth Earley (SE): I wrote that book because there's a lot of stuff that's right at thirty thousand feet, right? It's like, Oh, AI is going to change the world or AI is a threat to humankind or AI's going to do this. And it's all that arm-wavey stuff, right? And that's a lot of it's true. You know, some of it exaggerates. Some of it goes too far. Superintelligence, I'm not sure you know how far we are from a Superintelligence or artificial general AI, artificial general artificial intelligence. You know, which is an AI that can answer anything in any context that we're very far from that. But there's also other books that get into the details and the weeds of, you know, n-dimensional free space vectors and, you know, backward propagation, deep neural networks and all the math and all the heavy engineering. People don't need to know that stuff. You know, business people don't need to know that they need to know about their use cases, about their business processes, about their data sources, about their business objectives, about their metrics. And all of that is what the book is about. The book, you know, a colleague who is a very experienced technical AI guy, machine learning guy. He's deep in the weeds, you know, PhD quality stuff, level stuff, really, data engineering, data science, deep, deep stuff. And he said to me, you know, it's less about AI and it's more about kind of what you have to do to make AI work. And I'm like, bingo, you know, like ding, ding, ding. Yes, that is what it is.

PH: One of the things that organizations have found a very strong use case for when it comes to AI is their customer experience operations. So AI and CX this has proven when used right to be a very powerful combination in recent years, is that right?

SE: Absolutely, absolutely. And it's one of the areas that I really focus on in the book. You know, I focus on the customer journey. I focus on product data. I focus on customer experience. I focus on customer data and attribute models. But a lot of the heavy lifting has already been done, been done by the big vendors. So you, as a business person, as a technology leader. You as a business leader need to look at the capabilities of those platforms and say, OK, what's my business problem? Right? And it's not going to replace a human. It's not going to do the end to end process, and not going to figure out a process you don't understand. You can't automate a mess and you can't automate something you don't understand. If you don't understand it, the AI is not going to figure it out.

SE: So, so it's incumbent upon organizations to understand those processes, understand those data flows, understand the quality of the data, the provenance of the data, the structure of the data, the attribute models. All of those things are basic. Blocking and tackling that should be done. You know, I've seen organizations kind of outsource a whole, you know, virtual assistant capability. Now it's - you need to own the pieces of it that are the differentiators, right? Which is the knowledge, right? What are your products? What are your services? What are your troubleshooting approaches? What's your knowledge about the customer? Because think of it this way right now. Virtual assistants are not great.There's some good ones out there, but a lot of times they break. They stumble on silly things like I had one the other day and I left an apostrophe out of Don't I spelt D-O-N and there's a T. There was no apostrophe. And said, I'm sorry, I don't know what you're talking about. You know, it was a simple. And then when I put the apostrophe in it recognized the intent. Well, that's silly. That's not even basic machine learning, intent, you know, intent identification or utterance classification. So my point here is that there's a lot of basic stuff. And down the road, these virtual assistants are going to be very, very good. They are going to be. I wrote it in my first chapter of the book, and I think you can download the first chapter on our website,, that's E-A-R-L-E-Y.

SE: Don't forget the E before the Y, but I think you can download that first chapter. I'll make sure it's available in your show notes, but it talks about a vision for the future, and it just walks a person through their day where every part of their day, their walk, they're talking to, interacting with virtual assistants that work for him - for them, for this person, and they are doing their work. They're helping schedule flights and checking portfolios and purchasing replacement products and making recommendations and handling scheduling. That is what we'll get to. One day they will be fluent and they will be seamless and it'll be just, we know that's how technology is right. One day we'll get there. Where we are today, between where we are today and where we need to be is a knowledge and data problem, right? It's a knowledge and data problem because it's not about the technology. We can do it the technology, but the data sources are not in the right shape, they're not curated. And it's the codification of the knowledge of the organization because organizations compete on their knowledge. Period. Full stop. Knowledge of customers. Knowledge of markets. Knowledge of solutions. Knowledge of potential products and routes to market. And competitive disadvantages and advantages. And so on. That's all their knowledge and all people do internally. Humans are creative and they work with knowledge and they create that knowledge.

SE: Virtual assistants are chat bots, are a channel to that knowledge. Chat bots are a channel. They need that knowledge. Now, if you don't have a handle on that today and you don't get a handle on it in the next few years, you will be caught flat footed. That means you need to build out those content processes. When you talk about omnichannel, when you talk about Customer 360, all of those things depend on the ability to understand the customer in data terms and to be able to present back to them. The messaging that is presented by the orchestration by digital machinery, the digital machinery that orchestrates these things that takes in customer signals and then presents something back. Now, all of that means if organizations don't do this, it will be existential. If they don't start now, they haven't done it and they don't start now. I guarantee there will be a lot of companies spending tens of billions of dollars trying to catch up, and many of them will not be able to.

PH: Well, based on your experience, what should companies who will set out, let's say they start today to try and take your advice and build on it? What are some of the things they should expect in terms of the big challenges and the areas they need to be prepared for and to concentrate on?

SE: Yeah, yeah, yeah. So one thing is its executive vision and support. One of the problems that we have in most corporations is the quarter to quarter a short term investor earnings expectations. Wall Street expectations. Sometimes when the stock price goes down or they get a new leader, they go, Oh, wait a minute, why are we doing this? What do we need this for? What? What's the ROI? And yet that has been established at the early stages of the project. This is a multi-year commitment. This is a marathon and this will take millions of dollars. People do multi-million dollar, tens of millions of dollars, hundreds of million dollars on digital transformations, and they ignore the knowledge. I worked with one organization spend two hundred and fifty million dollars on SAP, and when it came to the knowledge there was about a million and a half dollar investment they needed to make. Whoa, that's way too much for us, right? In other words, they spent the money on the structured side. They didn't spend it on the unstructured side.

PH: So when you say knowledge, you mean gathering in just figuring out what we have, what unique knowledge we have and data we have and how to gather it together?

SE: Building knowledge bases, building, you know, standard operating procedures, you know, making the day to day operational things so people can't find their stuff. What do they have to do? They have to recreate it. That is inefficient. That causes friction that slows down the process when you don't know, we work with one financial services firm. We had to build the chains of trust throughout the organization for reports that they had to do. What happened was before we built this. People didn't know the accuracy or the validity of the data source, so they would go back upstream, check every data source and it added weeks to their process and huge amounts of inefficiency. So the point is when people don't trust the data and the same thing goes for AI. By the way, if they don't trust the result and they don't understand the result, they won't use the AI. But even in day to day work, unless you are sure about your data and your content, where is the latest strategic plan for this? Where's the troubleshooting guide? Is the knowledge base up to date? What do I need to do my work right? It's a whole ecosystem of information flows throughout the organization when you look at something like personalization. There are so many people in so many departments and so many processes that go into this.

SE: When you're onboarding products, you have a whole product operations catalog operations team. You have to understand what the right attributes are that people need to find there. They're there to solve their problem or find the products downstream. But that's going to depend on who they are and what their background is. So we need to get that from marketing. We need to get signals from marketing. We need to understand their real time behaviors on the web. We need to provision self-service. We need to do. There's all of these information. I have an animation that usually blows people's minds because it shows all of these different flows. But the point is, it's an orchestration process, and you need maturity in sharing knowledge and sharing information up and down that food chain, collaborating seamlessly friction free so that people can ultimately support those downstream customer facing processes. At the same time, again, you have lots of knowledge being created throughout different parts of the organization that all will go into this idea of personalization. And you need the right architecture. You need the right reference. You need to componentize knowledge so that you can serve up the right piece of content and the right product for the right individual. That means understanding their characteristics as they go through their journey. It means understanding that journey in detail.

SE: It means understanding the messaging and what's called a messaging architecture, so you can kind of shift in and out like a different call to action, a different hero image, a different target audience that can be optimized by machine learning. But you need those building blocks in order to try those different combinations. And then you need to understand, OK, what? What's the knowledge of a brand? A brand manager or a merchandiser or customer specialist who knows, well, this is top of the funnel content. This is middle of the funnel content. This is, you know, you know, conversion content, right? For this audience doing this thing with these products, right? That's knowledge that needs to be captured and applied to this digital machinery. So this whole idea of orchestration personalization through orchestration requires that we have those building blocks. Machine learning and AI can help to optimize all of this and can help to interpret signals. But if we don't have those building blocks, if we don't have that data and the data architecture and the content correctly structured and the knowledge captured, none of it's going to work right and think of it. Think of the best salesperson or customer service rep if you've ever worked with. They knew your problem. They knew what you needed. They anticipated what you needed. They could answer the questions.

SE: You trusted them. They can make recommendations that were helpful. And that's what we want to do in our organizations. We need to build that capability at scale. We need to take the knowledge of the best customer service people and be able to capture, codify and present it to the customer. Now it's a long way to get there, but the beginning piece is building things like agent assist types of helper bots, right? Bots that will listen to the call in the call center and anticipate, Oh, they're talking about this. Let me recommend this piece of content to the agent or the agent has a question or the customer has a question. The bot can interpret it. Present a piece of information the agent can say, Oh yeah, that's the right information. Push it out makes their job much more efficient. Or they say, no, that's not quite the right intent, or they didn't interpret intent correctly. So we can correct the frontend interpretation of what the customer is asking, right? Because that's intent. That's utterance classification to an intent. Then the intent pulls back a piece of information. Well, is that the right information? Human judgment can tell you that. Now you make the agent more efficient and you teach the bot and you make the bot better. And then eventually, when it's good enough and it's accurate enough, you can release it to the wild.

SE: You can give it, you can have it directly work with customers unless you have very narrow use cases with unambiguous results. It's not possible. It's not reasonable to build a purely customer facing bot unless it's very focused, very targeted, clear outcome, clear steps like Amazon does a good job of this with returns, right? It's such a defined process that it's repeatable, it's unambiguous. It has a clear outcome. That's what we need to begin with. But if it's anything that requires judgment or human expertise or nuanced interpretation of a complex request, then the human has to be in the loop, right? And we always have to be able to escalate to a human. So there's many, many different things around, you know, 360 view of a customer around personalization, around improving customer service or on call deflection. First of all, why do people call the call center? Because something's broken, right? Why we need to fix that stuff upstream, and we can use these tools to do that. We can make that experience better by using these high functionality tools. So anyway, my point, my overarching point here is there's a lot of foundational work that can be done that will improve processes and efficiencies today while preparing for the future.

PH: It's gathering the data that you will ultimately need to feed into the next iteration of your AI strategy.

SE: That's right. The data and the structure and the content and the use cases. So I think when you ask me the question, what do they have to watch out for? They need to get consistent support. They need to have the right vision. They need to have the North Star, even if they believe that North Star is aspirational. One guy said to me when I showed him the orchestration approach that we've developed that is getting a lot of pickup right now and we have patents on it. And we there's also a lot of open source and there's a lot of techniques that are available to anybody. But this guy said that will never happen in my natural lifetime at this company, right? And I think it's a bad. I think that's not the way to look at it. First of all, things happen really much faster. They accelerate much faster than we would imagine, number one. And you have to still have that North Star, even if it feels aspirational because you can build the building blocks along the way and you can get value from those building blocks. So having that vision, even if it's aspirational, getting the support from executives and then building a roadmap and plan that gets you those incremental wins. It can't be, you know, the gigantic galactic uber AI project that's going to solve world hunger or cure cancer. MD Anderson with IBM and Watson, there's $78 million. Now, they didn't solve the problem, but what they did was they created a lot of other ancillary technologies and benefits and simple things and processes that actually was very, very valuable. These things become transformations when you do lots of small interventions there that are highly targeted around pieces of the process, then they collect together and they become a transformation.

PH: And that's - it's quite clever because what it means, particularly around that executive buy in, if you map it out this way, then you can show results in a consistent basis as opposed to people have to wait 10 years to see the results, so that will continue. That will keep the investment flowing.

SE: Nobody, nobody will wait that long. You're throwing out your stakeholders. Like what?

PH: Yeah, yeah. Yeah, exactly. So the you mentioned the magic partnership between human and machine. In your experience what are some of the biggest challenges that can come to that partnership?

SE: Yeah. Well, there's certainly the threat that people might feel of, OK, I'm going to work with this tool and it's going to put me out of a job. There's a big cultural fit piece. One of the things that has to happen is you have to gamify it, right? You have to give - you have to help the service rep because again, it's going to come from front end. People who are on the ground has to be positioned in a way where they understand that this is to help them be better and to take away the drudgery right to do service. Reps want to walk people through yet another ten thousand five hundred and fifty two times the password reset, right? And that's low hanging fruit that's pretty much solved. But you know, there's some stuff that's just boring, right, that people don't want to do, you know, and it's just drudgery. And so and or - And so to see to help them see that this is a tool to take away the drudgery, it's like the turn of the century,-you know, the turn of the 20th century when steam shovels early nineteen hundreds were steam shovels are being used. People say, What are all those ditch diggers going to do? Well, people don't like digging ditches. Yeah, and you got to build skyscrapers and highways and in cities.

SE: So we have to think of it the same way we're taking away the drudgery. We're taking away the stuff that's boring, that's repetitious, that's mind numbing. And we're giving people more, greater abilities, but we're also giving them tools to handle difficult situations. Customer starts getting escalated. Maybe they get triggered. Ok, let's do a real time intervention. We can do sentiment analysis, identify that hand the call off to someone else, bring in a supervisor, bring in a script to guide them anything right. There's stuff that's going to help make their job more satisfying, but it's also helping them understand that look, you're building skills by becoming a robot trainer, right? That will be in demand. That's going to be in demand. There's going to be a lot of that work that's going to require human judgment to correct those things because they're very, very dumb. And so it's helping to communicate that this is not. And it's also the business model of the BPOs, right, the people that are handling call centers and building call centers and managing call centers, right? What do you do when your call volume decreases? Well, there's you have to kind of cannibalize that business in a way because your job is to add efficiency and effectiveness or a competitor is going to do that right? So now there has to be new business models and pricing models to say, are we responsible for just calls, handling calls and looking at the statistics? Or are we responsible for solving problems and ensuring a high level of customer satisfaction? Right.

SE: So there are some inherent changes in the marketplace. We also have to say, ‘Hey, what's going to happen when we displace more people,’ right? When some of those entry level jobs are automated? While I believe it's incumbent upon the technology providers, the service providers, the BPOs, the organizations to also use the same technologies that are going to cause the problem to help to address the implication of that. So when people have - need to be upskilled, when they need to be trained, when they need new motivation, when they need to find direction, even if it's, you know, even if it's about their own personal value in their own personal motivation, in their own personal meaning in life, there are lots of resources that can help with that, but usually in the form of three hundred dollars an hour coaches or you know or whatever people that are in scarce supply. And we will not be able to scale that at the level that would be required as society goes through displacement. But the same tools can help, you know, take that same time.

SE: It's not going to replace a human coach, but it can really. We can diet. We can assess people using machine learning approaches. You know, you get those assessments and say, you know, do you like to ride a bicycle or eat ice cream? You know, like what? And then you give an answer. And then all of a sudden it comes up with all your personality characteristics, right? The same things can be used to assess people's hard and soft skills. In fact, customized e-learning, customized training, coaching, all of those things are going to be based on the same technologies and be able to scale and leverage and amplify human capabilities to address the displacement because we are going to have to retrain large swaths of the workforce as these things evolve and people may think, Oh, that's a long way away. There's been predictions about job loss. It's going to happen, but it's going to take a little while longer. But when it happens, it's going to happen at a very massive scale. So again, it's looking at both sides of those things and looking at the impact.

PH: Do you think that the human machine partnership will lead to more general AI models that can tackle broader sets of problems and become the norm? Or is specialization the future?

SE: You know, it's such a great question. It's such a great question. I believe that specialization is the future, that you're going to have fleets of virtual assistants that are being highly tuned to specific tasks. And then you will have concierge virtual assistants that will decide where to send people when they have a particular type of problem. Now, could you call that a step toward more general intelligence? Perhaps. But I think it's going to be, just like, you're human. The human brain works the same way. You have very specialized sets of neurons and neural pathways that are higher - that handle very specific functions. They found that in the visual pathways and parts of the brain, they found specific neurons that lit up for specific actors and actresses and celebrities right? Really specific neurons. So it's highly specialized. Human brain is highly specialized. There's lots of different circuits and there's something called the pandemonium theory of language recognition, where you have these different sets of neurons that are responding to interpreting either visual text or speech. And what happens is the ones that yell the loudest or they have the highest signal are the ones that are amplified. So it's almost like you have these little competing sections of your brain that are trying to interpret things and the ones with the biggest signal win. Well, guess what? That same concept will be applied to virtual assistants and to bots. So I'm not making comparisons to the human brain because there's hundreds of neurotransmitters.

SE: I started as a biology major and I went to chemistry. There's hundreds of neurotransmitters and they're all analogue. They're all and a single neuron can connect to tens of thousands of other. Neurons and there's three billion of them, so the number of combinations it's just so vastly more complex than what we can conceptualize right now with technology, but there are some analogies, so I don't think we'll get to human brain capabilities for a very, very long time, if ever. But there are some analogies in this whole idea of very specialized functional bots. So that's why one of our patents is on master data for AI, master data for virtual assistants. And what it's essentially saying is when you have these fleets of tens or hundreds of thousands, tens of thousands of bots that eventually will be built if you need to update them with new products, new services, new offerings, new solutions, whatever, you have to push all those things out to many, many distributed bots. And that's going to be a large scale information management problem. So, so my point is that I believe that we will have specialist bots that handle very narrow functions really well, and then we'll have concierge bots that will route those tasks and those utterances and inquiries and intents to those different because they are specialized. And that's kind of the way the world works, right? Everybody is a specialist. Right? So there's no reason to think that that won't be replicated in that way.

PH: Ok, a couple of quick fire ones to wrap up, Seth. What is the most underrated ML algorithm of the last few years?

SE: Well, you know, it's hard to answer as an algorithm because there's so many, there's hundreds and thousands of algorithms, thousands of algorithms, I think really what we want to say is what application areas are either - are underrated. You know, it's really about, you know, the fitness to purpose. You know, there's I think the basic stuff around knowledge retrieval is underrated, building knowledge retrieval bots. That is a very fundamental thing. It requires you start to get your data house in order, you start to get your knowledge in order. It is necessary for everything else. It is a precursor and a foundation, and they are not difficult to build. They're based on search. They're based on AI powered search is a lot - I just wrote an article that should be coming out very soon on AI powered search and there's a lot of stuff that is in search that you used to have to build. That's much, much less expensive. That can give you very powerful results, building question, answering systems and building knowledge retrieval bots. Those will be part of the functionality of your virtual assistant, so building those will solve problems today. It'll help your employees get to the answers they need. It'll give - start building some discipline in a knowledge architecture and knowledge curation and content processes.

SE: Again, none of that stuff's going away, and we can't do the more advanced stuff without the foundation. So it's building those foundational applications, understanding your intents, you know, understand the utterances, understanding how people are asking for questions. A lot of the problems that we're trying to solve with AI and text analytics is based on our past sins in poor data and content curation and the lack of consistency of architecture. But now we can derive a lot of that stuff, but we still have to curate the sources. We still have to capture the knowledge that can solve problems today that can vastly improve the effectiveness of a call center. And again, that becomes a foundation for the more advanced stuff. So I think there's a lot of different algorithms within that application framework that just need to be exploited and leveraged. So that one is, that's underrated, right? Because it has to be done.

PH: Ok. Excellent, so lovely answer. And I've also got the converse question, what is the most overrated?

SE: Personalization, Customer 360 and virtual assistants that are not using knowledge engineering principles, right? There's a lot of there's a lot of discussion about personalization. I'm writing an article in the Journal of Applied Marketing Analytics, another deadline I have pressing, but it's about personalization, and I've written a few about personalization recently as well. And the point is that everybody waves their arms around, Oh, you have to take in what the customer wants and you have to understand the signals and you have to do this. But there's no detail on how to do it. And it doesn't point. There's very few vendors that are really talking about how you need to architect the information in order to truly make personalization work. And I think the other problem with personalization is there's either simplified approaches like customers who bought this bought this. There can be relationships, product relationships and or cross-sell and upsell, but those are data dependent. They require good curated product data and product data relationships. So they're oversold and they're under-delivered because the foundation isn't there because people try to buy an application to make it work, and it won't work unless you have these other pieces.

SE: There's a lack of understanding of the customer journey and customer needs. We built an architecture for an organization a number of years ago that that enabled personalization, and at the end of it, they're like, Well, how do we differentiate this audience from this audience? Well, we don't know. So they ended up using the same messaging across those different audiences. So at the end of the day, they had the technical capability. They didn't have the process, maturity or enough deep understanding the customer needs. So personalization and Customer 360 - Customer 360 is very dependent upon having attribute models that are consistent across your entire tech stack, your marketing and eCommerce tech stack. And then there's ways of evaluating each of those tools in the context of the customer journey. That's another article I wrote for the Journal of Applied Marketing Analytics. It was almost the whole issue. But there are very clear processes and ways of tuning these things. And unless you get those foundational pieces right, those other - those aspirational pieces of functionality are not going to work. Companies have a hard time with this because they're not building the right foundation. So those are very overrated.

PH: Well, look, Seth, you've been so clear about the importance of this area and the, I guess, the consequences of not doing anything. But also what I've taken is that it's not - it doesn't have to be this big, massive multi-billion dollar investment from scratch. That there's a lot of low hanging fruit. There's a lot of things that can be brought into an organization and implemented from an AI perspective that are small, consistent, give you wins along the way and help to feed that bigger iteration, maybe down the line with the knowledge of the data. And I think you've - you make it very clear in your book that one line jumped out of me. That probably is a good way to wrap. And it's AI only works when it understands the soul of the business. And I think that's it - knowing the soul of your own business and how AI can work hand-in-hand with that.**

SE: Yes. It all has to be understandable. It all has to make sense. If somebody tells you things and they confuse you, walk away, you should be able to understand. Hopefully, today we had a conversation that was understandable, you know?

PH: Thank you so much, Seth. It's been fascinating hearing about you and your organization and also just this this whole area. So thanks very much for joining us on TELUS International Studios.**

SE: Any time I appreciate it, it was a lot of fun. Thank you.

PH: And thanks to you for tuning in. We’re excited to be back with some brand new episodes. So, if you like what you just heard and if you think other people should hear it, share this episode and hit subscribe. While you’re there, give us a review and a rating, it really helps. And of course, if you would like to find out more about TELUS International and the global and disruptive brands with whom we work, check out Until the next episode, take care!

We can help

Be in the know

Get the latest insights and resources delivered right to your inbox.

Subscribe now

Proven formula for results

See how the Culture Value Chain can transform your customer experience organization.

Experience the Culture Value Chain