In a recent episode on Emerj’s AI in Business Podcast, Matthew DeMello interviewed Edwin Pahk, Aquant’s SVP of Customer Success and Pre-Sales, about how AI can help enterprises retain tribal knowledge amidst turnover and expertise gaps. They discussed the challenges of digital transformation, emphasizing the need for AI to be integrated behind the scenes before becoming customer-facing. 

Edwin highlighted the importance of personalizing AI for employees first, especially in complex industries like healthcare and heavy machinery, where understanding the unique needs of different equipment and environments is crucial. Capturing and converting the valuable knowledge of subject matter experts (SMEs) into data for AI use is essential to ensure accurate and practical solutions.

He also addressed the organizational changes needed to adopt AI effectively. With the labor force in technical roles shrinking, companies should focus on hiring individuals with excellent soft skills and customer service experience. AI tools can make jobs more enjoyable by reducing routine tasks and helping to retain employees.

Edwin emphasized the importance of looking for candidates with problem-solving skills, curiosity, and strong communication abilities, as these are becoming increasingly important in evolving technical roles. The conversation underscores the need for a people-focused approach in AI implementation, integrating human expertise to enhance employee and customer experiences.

Check out Matthew and Edwin’s podcast episode: The Importance of Tribal Knowledge for the Success of AI Adoptions – with Edwin Pahk of Aquant

Transcription:

Matthew: Welcome everyone to the AI and business podcast. I’m Matthew DeMello, senior editor here at Emerge Technology Research. Today’s guest is Edwin Pahk, senior vice president of customer success and customer pre-sales at Aquant. Aquant is an AI-powered tech company that builds a co-pilot platform for service workflows.

Edwin joins us on today’s show to discuss how AI can help enterprises from every industry retain their organizational knowledge or tribal knowledge, as it is called, in the face of turnover and other expertise challenges throughout the episode. Edwin draws from his experience in the service industry to emphasize the importance of personalization on both sides of B2B workflows.

He also shares how these same tools can often enhance the work-life experiences of subject matter experts across workflows. Today’s episode marks the first in a series sponsored by Aquant, and without further ado, here’s our conversation.

Matthew: Edwin, thank you so much for being on the program with us this week.

Edwin: So excited to be here. Thank you for having me.

Matthew: Absolutely. I’m a managing editor here at Emerge. So, aside from lending my game show voice to these podcasts, that’s the predominant amount of my day. We had a writer pen an introduction for an article that cited reports from the OECD and APQR, stating that around 70-80% of digital transformations either fail or don’t meet expectations in a way that can’t even be reconciled with a philosophical approach to our eyes. 

I think what we’re seeing across the industry is maybe in terms of the hype cycle that folks were too quick to jump into the pool’s deep end. We want everybody to jump into the pool, but ensuring it’s the shallow end is essential. And once you jump in, maybe take some time. 

But what do you think is going on there regarding the larger dynamic?

Edwin: The last 12 to 24 months have been very telling for us both as an organization and, quite honestly, the entire market. I don’t wanna say the inception, but the emergence of technologies like ChatGPT has created a bit of a feeding frenzy on, “Hey, how are we gonna use AI? Let’s use AI here, let’s use AI there.” It’s created this giant storm of figuring out where we can plug something like this in without a lot of thought around quite a number of different topics. 

So first of all, this would be, well, are you delivering on a use case that makes sense and is going to improve the experience from an employee and customer perspective? Plus, data security should be considered, and if the predictions are accurate or correct. These types of things tend to be the second question you ask after you start to go down this journey for the most part.

The best way I can create an analogy around this is, let’s say you look at some of the best use cases of machine learning and AI—for example, a recommendation engine to tell you what the next best episode or best series for you to watch, or Amazon recommending to you what you should add to your shopping cart. In both examples, if you get it, right, that’s amazing. If you show another video that someone likes, it’s a positive outcome. But if you don’t show something someone likes, it’s not the end of the world either. 

In our space, especially in customer service, and when you’re solving problems and troubleshooting equipment, getting it wrong is really bad. We’re talking about supporting things like large construction equipment, diesel engines, and medical devices that are used in operations. So getting it wrong—or getting a false positive response–is extremely consequential.

As a result, many of those types of outcomes cannot be tolerated. When it comes to digital transformation in our space, there is often a higher bar for the application of AI. That’s the first thing. 

The second thing—a bit more of a standard thing that I think everyone can appreciate and agree upon—is that when you introduce something like AI, many people think of the Terminator or think about their jobs being taken away. 

A lot of the failure in digital transformation also comes from this: how do you show the people who are supposed to be using what’s in it for them? I always like to say is it doesn’t matter if you create the best technology in the world. If no one uses it, it doesn’t mean anything. So there’s a big piece of this that is not just about changing the mindsets of the executives, the leaders, and the IT folks and getting them to use AI. You have to convince the call center agent that it’s something that will help them do their job and improve their quality of life. 

So these are some of the critical things and themes that we, as an organization, see when it comes to why digital transformations fail, especially with this massive word called AI that seems to be sitting out there for everyone to try to pounce on.

Matthew: Yeah, in an industry-agnostic sense. This is the take-home for everybody listening. If you’re expecting a model that works right out of the box and is customer-facing, that’s a dream world that doesn’t exist—that should not be trusted. You want to be skeptical walking into that room and want a long sales process so they can explain it. 

From what we hear across the board, not only from your last answer but with so many of our guests—if you’re not integrating it behind the curtain first… if you’re not acclimating that model to your organization the first before you acclimate it to your customers—doesn’t that kind of also flies in the face of how you train your own human employees? You have to train them first on what the organization’s all about. They’ve got to know what McDonald’s is before they’re working the phones in the aisle, offering people extra fries. They’ve got to know what McDonald’s is all about. They’ve got to know the golden arches. They’ve got to know Ronald. 

You also mentioned how important personalization is, as it tends to get called in a few other industries—at least once you get those models in front of customers and tailor them to their experience. 

Let’s narrow this down to the B2B crowd. In B2B, especially heavy industry field services, if you’re supplying machines that keep people’s hearts beating, those machines can’t fail. Those machines cannot enter a three-week-long customer service and be put at the bottom of the list. They need those machines never ever to break down. You’re forced to come to real grips with the fact that you don’t need a personalized system for your customers first. You need a personalized system for your employees first. 

Are there reasons beyond that why a generic AI model—one that’s not personalized to the employees and the customer—might work for service teams dealing with complex machinery?

Edwin: Yeah, it’s a really good point. When you think about the concept of generic AI, there is an understanding that AI is being fed by a specific fuel. It ultimately drives AI’s decision-making process. 

For us, what it comes down to—especially in our space, and it does apply to other spaces as well—is looking at personalization from a slightly different angle than the way that you might experience it in your daily life. 

As a consumer, you get personalized ads sent to you, given your location in the world. A whole number of different factors are added up to create the experience that you feel. For us, our customers, the machines that they’re supporting, and all of these technologies that we’re supporting, it’s slightly different. 

What is personalization? Let’s explore the concept. For example, an MRI machine in a Southern California hospital versus a rural Texas clinic will be different. The things you need to do to operate and support those machines are very different—even though they are the same machines. 

We like to talk about servicing and troubleshooting problems on many of these devices as a beautiful orchestration of chaos. There’s the call center agent asking the customer’s questions. There’s the machine itself, the service history of that machine, and what it’s being used for. There are the parts that have been replaced in that machine, and whether those parts are high quality or not, or from one place or another. Plus, consider the field technician that goes out there. These different factors create a unique situation whenever someone calls about a problem. 

It’s not as simple as saying, “Hey, my phone’s broken. Did you try turning it on turning off?” — you know how everyone says that. It’s much more about, “What have you been doing on your phone? When did you last go into Best Buy to get it repaired?” These things factor into a decision-making process that AI needs to take advantage of to provide the most effective recommendation.

The concept of personalization is about taking all of these unique events and creating a situation or some assistive technology that allows the employee to make the best decision possible given a specific situation. That’s the critical thing around personalization in our space here. Hopefully, that gives you some sense of some of the different variables that are factored into creating the right experience for our customers.

Matthew: Makes a lot of sense. I think it’s even essential for our listeners. This jargon—bespoke models and foundational models—used to be commonplace a year ago. I’ve since heard these terms fall by the wayside. When we talk about the generic models playing into that old jargon, you know, you do want a foundational model, a more general model for your organization, still tailored to the organization. My question for everything you were saying a moment ago about personalization to the employees is about getting that buy-in from the factory shop and the SMEs on the ground floor. Is that the front door? Is that the only front door, or how best do we tap into that organizational or tribal knowledge through data that avoids these problems?

Edwin: It’s a really good question. In the end, if you think about the value of an organization or a company, the unsung heroes are those subject matter experts who have been in your organization for the last 20 years. They seem able to whisper into any machine or system, figure out the problem, and get it back up to running. 

The thought process and idea is this: we’ve been in business for a few years. For example, about 20-30% of the identified problems and solutions within your organization’s data exist in the experts’ minds. The experts add their specific way of solving a problem. Their specific way of observing how to get things done doesn’t exist in the data anywhere—it exists in the people’s minds.

One of the critical things we realized going forward—and if you remember when I talked about the concept of a higher bar here—is that false positives are unacceptable for us. Our ability to convert expert thought processes, translating what they deem the right solutions, into data that can be interpreted by AI is critical. This separates what we do from other organizations to drive success in these situations.

The entire industry is facing it. People are retiring. People aren’t staying in the same jobs as they used to—you don’t have the 15- to 20-year veterans anymore, the subject matter experts who will be there forever. People typically go in for a couple of years and maybe leave. So, really, how do you institutionalize that primal knowledge and transform it into data that an AI can then sustain through the course of people coming in and flowing in and out of the organization? 

It goes beyond just getting a manual. If you talk to any customer service agent or technician, someone who’s servicing something, they’ll tell you the manual is good, but 50% of the time, it doesn’t represent what happens in real life. As a result, this is the reason behind promoting and creating this bespoke model that is accurate and produces what we call best practice here. In terms of how to solve these types of problems. A critical portion of that is consuming that tribal knowledge and converting that into data that then AI can interpret. 

So, that’s a big part of why we feel strongly about why it’s needed in this industry.

Matthew: Absolutely. Let me ask you a question about that 20-30%. Is that permanent, or does that wax and wane with more advanced technologies? Can you make a dent with personalized systems and AI geared toward gaining that tribal knowledge?

Edwin: Absolutely. You’re right. So, when I’m usually talking about it, it’s the initial instance of getting the model up and running. But once it’s actually in the ongoing improvements, how does the model react to feedback being leveraged within? 

The usage of our tools decreases over time. That’s not to say that there’s always going to be something that someone’s going to figure out on their own that never shows up in the data. We try to make it as easy as possible from a facility perspective to add that in on an ongoing basis. So it’s a living and breathing kind of thing more so than just writing a static manual.

Matthew: To gain this tribal knowledge, as it gets called from the organization, we need to personalize our systems for the customer. Usually, that comes last. You want to personalize it to the employees. First, tell us a little bit about what that process is like to adopt a model that will glean that information and make a product that can be customer-facing.

Edwin: It’s really important to understand how we’ve evolved in our thought process around using AI. And quite honestly, it’s lessons learned from our journey as a company as well. 

We used to think night, almost naively, that we could put all of your historical service ticket data, case data, contact center scripts, chats, and all this stuff into our engine. We tried to create a recommendation or problem-solving engine out of that — this was the original hypothesis. That approach of just using your basic historical data—similar to how a generic kind of AI system or model would use it—did not get the desired results. This was a big reason for what we realized: the need for tribal knowledge and to incorporate this concept into something more specific and approachable. 

I mentioned the context of a false positive as an example. A part of what we do is first examine all of your historical data. We start to understand which problems are being stated in this particular case and what solutions were employed. We make that our foundation.

Then we go through a process of leveraging subject matter experts. We get their opinions: “Hey, when you see this type of problem, how often will this fix it? How often is this going to fix it? What’s the best thing to happen here, given this situation?” We go through the process of capturing that opinion and expertise in a way that translates into data for AI to interpret. So it’s not what you would call a hard-coded manual or step-by-step process. It’s more of a, “Hey, given the seven circumstances of what you’re seeing, what do you think is the most likely solution to this problem?” 

We then combine all of these things and factor in that asset’s history, the parts, and their quality. We factor in the technician who performed the fixes because we all know that there will always be people who are great at what they do—and there are some people who are newer and not so great at what they do. We factor everything in to help you ultimately make the best decision possible. 

You see an evolution here from very basic, generic approaches of taking data and producing the most common result—to incorporating a level deeper of tribal knowledge and inserting the additional context of the work for the worker, the call center agent, the part, the asset…

All of that creates a personalized experience that ultimately allows our customers to get value out of what we’re doing. Hopefully, that gives you some context as to how we think about the process and philosophically approach it.

Matthew: Now, going back to the use case you mentioned before about retaining that expertise—I think even your CEO, Shahar might have mentioned it when he was on the program. That can make an even tougher sell for subject matter experts, right? Because it sounds like, “Okay, so you want to clone me into a ghost that has all my knowledge and expertise, and then you don’t need…” 

So I’m wondering what you find really works to get that SME buy-in, that ground-floor buy-in, to let people know that this is going to enhance their work life, not hold them back.

Edwin: Yeah, it’s a really good question. I would say it’s a challenge for many folks—especially to your point—when they hear AI and need to put a lot of their thoughts into it. It’s like, “Are you just trying to clone me? So you don’t need me anymore?”

It does depend. There are various motivations and desires, depending on who you’re talking to from a subject matter expert perspective. I can tell you a couple of the common themes we have. For example, subject matter experts love to work on difficult problems and love difficult puzzles. To a certain extent, it’s a motivation that keeps them in their job; they just love to work on these types of things. What they don’t like to do is get called up and deal with situations where you try turning it on, turning it off.

So one of the ways that we talk about it with subject matter experts is, “Hey, if you’d like for people to stop calling you about those kinds of things—so you can focus on the real interesting stuff—this is what this tool is supposed to help you with. It can free you up to work on truly unique and difficult fixes. AI is not going to solve everything in the world that you know, but we propose AI to solve all the simple things. That way, you can work on the cooler things.” 

But others have different motivations. We have subject matter experts who are passionate about building it into the system because they see it as creating a more significant impact throughout the organization. Because all of a sudden their learnings are being shared and creating a positive impact in the organization, elevating them as well. We’ve also gotten people promoted by participating in projects like this. So there are various desires and wants. And quite honestly, some people don’t want to do it—so there are those elements as well. It’s an interesting journey but a people-focused one. And I think that’s what seems to get lost in this a lot of the time.

Matthew: Yeah, absolutely. When I was at an AI vendor specializing in global taxes, we used to say that if we could at least take away a lot of those manual tasks, this would feel like an art form rather than a plug-and-chug. You can take that more artisan view of your workflows as you think of yourself as an artisan rather than just a guy who plugs things in. 

Now, let’s assume that that’s a compelling sell. This means an entirely different kind of organization, right? This is a fairly radical change of workflows, especially because you don’t have to worry about expertise being lost with people, retiring people, changing jobs, etc. 

Let’s take it from from the top on down. What is that? What does the organization look like to management or at least from the top? We can even work our way down to the subject matter experts regarding an organization that’s taking full advantage of tribal knowledge. What does that look like?

Edwin: We have a few customers undergoing this transformation process. One of the critical things faced across the board is that the labor force in this space is shrinking. Fewer people want to take on these types of jobs. So from a leadership perspective, the question is, I can’t hire someone who has medical device experience because there’s only a finite pool of those individuals. So, I need to change my thought process about who I need to hire. With the concept of AI-assisted Co-Pilots, it’s less about you figuring out how to turn a wrench and more about other types of skills, like interacting with customers and building relationships.

We often hear from many different customers and leaders that talk about, you know, just find me someone. I’ll be more than willing to hire someone from Chick-Fil-A who has great customer service and has some affinity towards turning a wrench, and I’ll turn them into a technician. That’s one of the significant changes in mindset that our customers are now trying to push forward with because they see the future here. There’s a shrinking labor supply and growing demand, and growing complexity in the products as well. 

If you go further down the pipe, looking at individuals and the technicians themselves, you’ll see a younger generation with an affinity towards using technology. Smartphones and AI are alluring to them because they’re using something that can assist them and help them get better. I don’t think there’s anything worse than being in a position where you’re asked to solve a problem and don’t know how to do it—it’s not a great feeling. These tools make that job more enjoyable and also potentially keep them around.

You’d think it might not be that big of a motivator, but it’s very interesting to see how, in different organizations, this is one of the major attraction points or retention factors. It’s really about the tools and the things you can do and the cool stuff you can work with to make your job somewhat enjoyable.

Matthew: Absolutely. Well, you were bringing up problem-solving before. People, especially the best SMEs, want to stay with the organization forever. It’s not that they’re the best technician, that they have all these metrics of all their wins up on the board. It’s more that they love solving problems.

You mentioned hiring somebody from Chick-Fil-A or any kind of background. Now, it seems more critical to find on the resume whether they have that zeal for problem-solving more than technical expertise. Are there any other skills organizations should seek in their future hires?

Edwin: What used to be a very technical role has become an experience or customer service role. From that perspective, many things you don’t find on a resume are soft skills, like communication, presentation, or social skills. The world is becoming more contactless. The touch points now that these folks have with people end up being the only single human point-to-point contact they have with that organization. They represent a significant portion of the brand experience now, more so than ever before, and are being asked to do things like upselling and increasing wallet share. You’ll also start seeing sales skills come into play here. 

It’s a combination of these soft skills that don’t always appear in a resume but should be factored into your decision-making process. With technology and the ability to train and coach individuals, you can always coach the technical skills. Coaching things like curiosity is very difficult. I think many Human Resources and People departments are trying to figure out hiring to create a competitive advantage.

Matthew: We know that with Co-Pilot technology, having AI-enhanced software behind you guiding what you do means technical technicalities can be worked out. Thank you so much, Edwin, for being with us on the show this week. I think it’s been illuminating for the audience.

Edwin: As always, Matt, I appreciate the time. I’m excited to see what the future holds for us!

Matthew: Wrapping up today’s episode, stay tuned for a conversation with Scott Burdett, Global Division CIO of Measurement and Analytics at ABB, discussing AI’s role in retaining organizational knowledge despite workforce challenges.

Scott expands quite a bit on what Edwin had to say in today’s show about what that will mean for field operation workflows, especially in the field. Very fascinating stuff, especially as we’re bringing more visual mediums into these workflows as an excellent primer for today’s episode. Don’t forget to check out the November 22, 2023, episode of the AI in Business podcast featuring Edwin’s boss and friend of the show.

The episode, AI Solutions for B2B Customer Experiences, features Shahar Chen, CEO of Aquant. Shahar discusses generative AI-enhanced co-pilot platforms’ role, particularly in improving B2B customer experiences in field service across multiple industrial sectors, from heavy industry to healthcare.

On behalf of Daniel Fagella, our CEO and Head of Research, and the rest of the team here at Emerge Technology Research, thank you so much for joining us today. We’ll catch you next time on the AI and Business Podcast.

The post Integrating Tribal Knowledge into AI: Insights from Edwin Pahk on the AI in Business Podcast appeared first on Aquant.