What to Consider When Implementing AI Agents – Techstrong AI Podcast EP54

Hello, and welcome to the Techstrong AI Podcast. I'm Amanda Ani, and with me today is Phil Tomlinson, who is the Senior Vice President of Global Offerings at Task Us. How are you doing today?

I'm great, Amanda. Thank you so much for having me today. It's a pleasure to be here.

Yes. Well, can you share a little bit about TaskUs? What services do you provide?

Yeah, uh, I'd be happy to. So, TaskUs are a, uh, US headquartered multinational outsourcing company or A BPO in other terms. Um, we have, uh, sort of a global reach.

We have, you know, 55,000 people around the world who do a whole bunch of really interesting things for our core client, which is really, you know, high growth tech, I would say is probably where we specialize and have specialized for the last 16 years. We provide customer care, customer support, which we call DCX, uh, trust and safety, content moderation, fraud risk and compliance, um, sales and lead generation services, AI data labeling, and, and a host of other sort of adjacent services to that cohort of customers. Um, you know, we, uh, publicly listed company, um, since 2021 and, um, you know, we, we've really made our name over the last 16 years, uh, working with high growth tech, as I said, providing what we call specialized and tech enabled services.

So this is a super interesting time to be in that space. Right, Wonderful. Absolutely.

We're gonna be talking about AI, of course, and, uh, trends and the global business impact for this coming year. So one of those trends, we're hearing an awful lot about AI agents. It seems almost every company's, um, trying to provide or work with AI agents to automate a lot of different tasks or help make things more efficient.

So what are you seeing from your end? Yeah, and you're right. Um, you know, agentic AI or, or AI agents are the sort of topic of the day with almost every conversation I'm having these days, whether it's with clients or with colleagues or with industry peers.

This, this is coming up. Um, you know, like, like, like most things, you kind of have to cut through a, a little bit of the hype to get to what is actually happening and, and how is it really playing out, out in the wild. Um, you know, task us, as I said, are a, a tech enabled specialized service provider.

Um, we recently announced, uh, our own ag agent AI consulting practice, um, a couple of weeks back where we're going to take a position as sort of systems integrators or, or, you know, solutions architects. Um, but really partnering with best of breed third party ag agent AI companies. Um, we think that's the right move for us.

We don't think we can go and build this tech internally and have it be as good as the best stuff out in the market. These are, these are amazing companies founded by folks with, you know, deep experience in the space and obviously very well funded as well. So we want to, we wanna sort of stay in our lane, as it were, and we think where we can add a lot of value is around that integration piece.

Right? The interesting thing about agen ai, people think it's magic. It really isn't.

It takes a lot of prep preparation, it takes a lot of integration, and it takes a lot of maintenance to keep them, uh, to keep it working the way it needs to be. Um, you know, for example, if, if you're a, you know, e-commerce company and you wanna automate a whole bunch of your customer service workflows, you need to make sure that your, um, your knowledge bases, your training material, your workflows, your customer scripts, your decision trees, all of those really need to be locked tight before you can start deploying automation. 'cause if, if, if you deploy that when you're not ready, you're gonna see some outcomes that you didn't intend and, and, you know, maybe at the worst end of the spectrum, you're gonna cause some real pain for your customers, which is really the opposite of what you're trying to do, right?

Um, so absolutely we're seeing, uh, our clients, um, lean into the idea of automation, but I would say they're leaning in with some degree of caution with one eye on, I think customer experience. They really don't want to just automate for the sake of automating. They wanna make sure that whatever they do brings value and enhances rather than depletes the customer experience, right?

So I think that's, that's where we want to focus on being the enabler for our clients to get from good to great so that they can automate absolutely. We, we want them to save money, we want them to be innovative. Um, and I think this comes to another trend that we're seeing, which is it's sort of the hybrid between humans and technology is really where the magic happens.

Technology's gonna be great at automating your, you know, repeatable, well understood, well documented processes. But when something throws an exception, and I'll give, I'll give you an example, right? Like, you know, if you're, you know, if you're, if you've rented a, a, a car through a, through a car sharing app, right?

That's out in the market and you break down in the middle of nowhere, um, you probably wanna speak to a human right. You probably want to be connected with a human being. If, like, if I'm out with my wife and kids and I'm break down in the, you know, middle of New Mexico or wherever I'm, I wanna make sure that I can contact a human being who is not only, um, empowered to solve my problem, but has like empathy and compassion and human experience upon which to draw.

That's something that is very, very hard to automate, right? But, you know, the other side of the coin is if I want to change the address on my account or update my credit card, or maybe I want to get a refund for something that hasn't gone, uh, super well, that's probably something where automation can, can really help and remove some of the simpler, repeatable tasks that humans, you know, might, might be doing today. And it makes sense what you said about, um, the caution because again, this technology isn't just a magical, you know, technology that can just be left to run on its own and, uh, fix everything.

That human in the loop is very important at this time, for sure. Now, maybe down the road in the future, this is gonna be such high tech technology, it's integrated everywhere, and, um, we're using it like we do other things without even thinking about it. But I think we're pretty far from that right now.

Would you agree? I, I would agree. Um, I think we are some distance from, uh, the kind of wide scale automation, uh, that's been predicted.

Um, it is coming in degrees. You know, we, we do see clients and, you know, think about some of our clients who, who have deployed automation across some of their customer facing flows and, and saw reductions of, you know, upwards of 40% of their volume. But what's interesting is, you know, as the volume went down in, in, in, you know, call it, you know, Q number one, um, there was new issue types and, and new escalation types and new work that was created in QS number two and three.

Now, QS number two and three may need less people, right? You may have had a hundred people working the, the previous line of business, but now you need maybe 75 or 55 or 20. But those folks are more specialized, they're more super agents.

They are, you know, empowered to kind of go beyond the workflow and, and look forensically at a customer's issue and examine metadata from different sources and make sort of judgment calls about an issue. Whether that might be a fraud situation or a, or, or, or a customer care situation. Like I said, maybe even a safety issue, um, like I was describing earlier, that's where I think the real magic happens, right?

Where you're not gonna see, you're not gonna see these, um, workflows go from, you know, a hundred people to zero, but I think you will see some reductions, and I think that's right, right? I mean, if you look back at the history of, of our industry, automation has always been a thing. You've, you know, whether those are IVR phone menus or, um, you know, chat bots in the, in the sort of pre generative AI era, the sort of conversational ai, um, that's always happened.

Um, but you've always needed human beings. Um, I'll tell you another story. Just in the last 24 hours, I, I, I, I, I live in Ireland, I came over to the US for some client calls from some client meetings, and, uh, I left my phone in the Uber coming from the airport to my hotel.

Um, I was completely locked out of everything, all my work stuff, all my personal stuff, all my banking, all my passwords, everything was on that phone. So, you know, stupid, my mistake. But, you know, I had to deal with the ride sharing company.

I had to deal with the bank, I had to deal with, uh, getting a new, a new SIM card and activating a new, I didn't buy a new phone. Um, in all of those interactions, I was immediately first placed onto like a, a, a bot in order to, and, and I had very little, um, success. I should, yeah.

I'll have to say with the first line of defense there, right? I had to, in all three of those cases, escalate to a human being to get what I needed. And, you know, this morning, I thankfully have all my access restored, but it was painful.

It was like five or five or six hours of my time yesterday, you know, so I, I think companies need to be cautious. They need to have one eye on the user experience, and, you know, people will ultimately, you know, vote with their feet, right? If they're not getting the customer service, if they're not, if they're not getting the outcomes that they want, they'll go somewhere else.

And I think that's where companies need to really think about, okay, what problem are we trying to solve? And where is it appropriate to deploy this awesome technology? And then test and iterate, test and iterate, test and iterate, right?

Always just trying to get a little bit better every time. As as you're, as you're learning, as you're going on. Yeah.

And I'm with you. I rarely have a good outcome when I try to solve something talking to a bot. So there's progress being made, but there we're a long way to go, and I'm, I am glad you got it all solved.

Thank you. So another question, you, you mentioned that when they bring in this technology, it solves one problem, but then it causes a cascade of some other problems. Uh, and of course, companies wanna see a return on that investment.

So to that note, how, uh, when they're in that first planning stage, what advice do you have for them when it comes to integrating new technology? How do they go about making sure, uh, to implement the right technology for their needs so that they get that correct outcome? Yeah, it's, it's such a good question.

I mean, there's so much in there to unpack, and we don't have nearly enough time but to, but to give you sort of a, a high level view, you know, as product owners or business owners are thinking about deploying this technology, they need to be considering a number of things. So number one, um, does the underlying data support what I'm trying to do? You know, am I examining the metadata around, um, if it's a customer service flow, for example, am I looking at the metadata around what are the, what are the contact drivers?

Are those contact drivers, um, some things that, that we control. IE am I deflecting volume from one channel to another inadvertently? And, and if so, I'm causing a, a, a spike in volume on, on, on one particular workflow.

Um, is my, is my product, um, is the user experience of my product pro properly configured so that, um, you know, it's optimizing for customers getting to good outcomes quickly. Uh, am I looking at, you know, the existing cu uh, customer satisfaction scores and reading, reading, not just the, the qualitative score, not just the quantitative feedback, but also what customers are writing. You know, what are they, what are they telling me?

Am I running customer surveys and focus groups to understand what the pain points are in incorporating that feedback into my, into my scope? So, you know, I think voice of the customer, both in terms of the data that under that underpins your operations, but also what they're telling you is, is vitally important. Um, secondly, do I have the buy-in?

Do I have my InfoSec team, my tech team, my legal team? Do I have my executive team's buy-in? Or am I just going in a silo and the, the second someone asks me about data privacy, I'm gonna hit a gigantic con concrete wall.

You'd be surprised how often that happens, right? Folks who, who run customer operations are very keen to deploy this stuff, but they haven't, they haven't got their appropriate buy-in internally from their teams that have to sign off on this stuff for, for various reasons of, you know, risk mitigation and, and compliance. Um, so, you know, I would, I would suggest those are really two great places to start.

There are other things, but those are two great places to start. Yeah. So when it comes to employee buy-in on new technology, anytime you have this change management or new tool integration, there is this issue with some of the employees bucking back, um, either sometimes, uh, now it might be due to a skill issue, a skill level issue, or it might be they're just comfortable doing things the way they have been.

So what advice do you have for business leaders for making sure that there's buy-in from everyone? Yeah. Um, again, it's such a big question and, you know, I think you're right.

It, it often you run into challenges around either skill or will, and, um, you know, employees are gonna feel, I threatened if, if, if there's no, if there's no transparency from the executive team, from the product owners about what are we doing over what timeframe are we doing it, what is the potential downstream impact on our business and our operations and on our employees, um, but also I think what's expected of our employees in the new era, right? Like, what, what do, what do we need them to know that they dunno today? What do we need them to do that they aren't doing today?

Uh, so I think it starts with transparency. It starts with, um, some degree of emotional intelligence where, where people are prepared to have conversations and get people in the room, um, you know, representatives from the employee group, people from the different operations teams, people who represent multiple stakeholders and, and asking for their input, right? And, and, um, you know, I recognize not every business decision can be, can be made, you know, via democracy.

Some things are mandated, some things are pushed from, from top down. But if you're doing that without at least consulting and certainly communicating to your employees about the potential impact and or expectation shifts that, that, that, that they will experience, I think you'll, you're gonna run into problems. So are there any, um, um, ethical or moral concerns, um, when it comes to integrating AI and, uh, how do companies manage this area?

Yeah, so actually my own personal background is in trust, trust and safety. Um, you know, so the, so the, you know, I've spent almost 20 years building and leading and, and both, both on the buy side and on the sell side of, of content moderation programs. So, um, you know, ethical and moral concerns are really at the heart of how I think about most businesses, uh, business problems.

It's no different with ai, right? Um, this technology is incredibly powerful. Um, it can be weaponized and exploited by bad actors or even inadvertently by a bad setup or a bad integration.

Um, and, you know, the, the potential downsides are pretty big, right? Like if you, if you've got, for example, a, um, a generative AI model that is integrated with your refund system and a customer comes in and, and is requests a $10 refund, but some are managers to game the system to get a thousand dollars refund, you know, you imagine the implications for your business, you imagine the implications for your, for your brand reputation. Um, so I think absolutely, um, having guardrails in place and having experts who can help you define the policies, define what the exceptions are, define the process of dealing with those exceptions, provide human in the loop maintenance and supervised and unsupervised fine tuning, that's, that work, um, is, is critical to the success of any integration.

And actually, you know, as I was telling you at the top of the call, one of the things task is, is doing and has done for, for several years now, is provide that kind of specialized human in the loop, um, uh, human in the loop kind of, uh, feedback for foundational model developers for large enterprise tech companies that are building their own models or deploying things on top of those models. Um, and we're doing red teaming and adversarial testing and prompt writing and, and safety evals. And, and this, there's actually an, you know, there's a huge spike in demand right now.

So, so, you know, another thing that's that I've, that I've really found fascinating to watch over the last 12 to 18 months is as volumes in certain customer operations flows have gone down or maybe plateaued, we've seen a spike on the other side, kind of the flip side of the coin, which is all the work that needs to go into building and deploying and maintaining those models. Um, and there's, there's a ton of, uh, human effort that's required, right? And, um, I think, you know, coming, again from a trust and safety background, that has always been the case.

The the goal is to automate, but you absolutely have to automate safely and you have to automate with high quality, and you have to automate with, um, the experience and the safety of your end customers in mind. And, and it's no different with ai. Absolutely.

Well, if there was one key takeaway you could leave our audience with today, what would that be? Yeah, great question. I would say, um, don't be scared of this technology, but do your due diligence.

Um, partner with folks who know what they're doing, partner with folks who understand your business very, very well, and who are prepared to roll up their sleeves with you in the, in the integration and deployment of the technology. Um, and I would say don't remove humans just because you can. Uh, I go back to my example.

There are times and there will always be times where we need to speak to another person. And that could be a, an issue of safety, it could be an issue of urgency, it could be one where there's enough complexity or nuance. Um, those are, those are not going away.

And I would you, I would urge all companies to think about where that line is and where across their customer operations it's appropriate for a, a smart, well-trained human being to deal with the issue and where it's appropriate for, for, for technology to pick it up and, and just to constantly monitor where that line is and, and, and, and test and iterate over time. Alright, well, thank you so much for coming on the show and sharing your insights with us today. It's been my pleasure, Amanda.

Thank you so much for having me. All right. And thank you to our audience.

Stay tuned. Here's more.