Two Lanes, One Outcome: Designing Dual Paths for Humans and AI - TTEC Digital

In this CX Today interview, Carrie Brough, Director of Strategy and Ops for TTEC Digital EMEA, argues that a single CX model is fast becoming a liability.

Brough explains why AI-driven interactions need quick, decision-ready responses, while humans still need empathy, context, and brand nuance. Carrie breaks down the operational warning signs that your AI and human traffic are colliding, including repeat contacts, confused outcomes, and frontline teams being forced into a “clean-up” role.

The conversation also covers when an interaction should move from the AI lane to a human agent, and why risk, not emotion, needs to drive escalation decisions.

Transcript

SPEAKER_01 0:07

Hello and welcome to CX Today. I'm Nicole Willing. Customer engagement systems are entering a new phase. For years, enterprises have designed their experiences around a single assumption. The customer on the other side of the interaction was a person. That assumption is beginning to change. Today, companies are starting to interact with both human customers and AI agents acting on behalf of customers. That shift raises important questions about how service systems should be designed, measured, and managed. So to explore what this means in practice, I'm joined by Carrie Brough, Director of Strategy and Ops for TTEC Digital In Mayo. Carrie, thank you for being here. Thank you for having me. Lovely to be here. So let's start with the big picture. What does interaction with AI agents as well as humans mean in practical terms for companies when they're building their customer engagement systems? Okay.

SPEAKER_00 0:58

Well, I believe empathy is at the heart of good service design, or it has been traditionally, especially when we're considering the human customer. And that really hasn't changed. But now we're not just mapping for the human customer. Sometimes it'll be the customer's AI agent taking over. And the day these customer AI may very well be handling the first part of the contact. And they're wanting clarity, facts, prices, updates. This needs to be done quickly, and without all of the effort that you've previously put into service design where you've been looking at tone of voice and brand experience and empathy. We need to build the engagement system that works for both the AI, who wants it quick, and for the people that'll want it to be more tone of voice, more empathy. And when the human takes over, that's where that good design empathy will still come in. And it should be seamless between the two interactions. So I think to do this, you need to get the logic right so that the customer's AI can get through, do its job quickly, but you also then need to know what that handoff looks like and know when you're interacting with a human and need to offer that human empathy. Once you have got that front part right and you can tell the difference, you can tailor the experiences so both types of customer get what they need to get the best outcome. Sure.

SPEAKER_01 2:39

Yeah. And so then as organizations begin adapting to the shift, you know, many of them discovering that the current systems obviously weren't designed for that. This is like a you know a new technology that's come along since then. So what tends to break first when companies try and force humans and AI through the same CX path?

SPEAKER_00 2:56

Um, yeah, I mean that's a great question. And I think, you know, the real challenging part comes when an AI wants to or needs to access more personalized answers, um, or it needs greater context that's based on the individual customer records. So authentication, um, you know, current systems that uh we're designed to block box. We don't want box accessing that type of information. Um, and they're built for humans. So they'll also stop the AI agent from helping the customer and where it is authorised to help. So if AI becomes a normal part of the customer contact, we'll need to rethink identity, how and when we let the AI access customer information without losing those levels of security and privacy that we've worked hard to create. Very much we might need to rebuild authentication frameworks, um, you know, authorising the AI agent to access the customer information and that for them will mean new ways of checking whether that AI does have the consent to do that, whether it has the consent to share data safely. And similarly, we'll need to track actions for compliance and make sure that we're doing the right thing. Um I think that's where things could break first if we're trying to force an AI agent down a service that has been designed for humans. Um, we'll now have to balance that quick, responsive and still secure method of interaction with a way that still works for humans, but will allow AIs to interact with your services as well.

SPEAKER_01 4:36

Yeah, definitely. Um so then you've just described an approach that helps enterprises think clearly about think more clearly about this challenge. Um, and your framework for this is kind of a dual path operating model built around three stages, detect, decide, and route. So can you explain how that works?

SPEAKER_00 4:54

Yeah, so when we talk about dual path, we're really creating um a service where you know humans and AI will will reach the same front door, but then they don't necessarily want to take the same route once they're through. Um, you need to be ready for contacts from both of these services, so from from real human customers and customer-owned AI, and make sure that every interaction gets the balance that it needs to answer it correctly. So, first of all, you need to figure out who or what is reached out to you and what they actually even need. So that's the detect part. You know, understanding is it human um or is it an AI? Is it something simple or is it something complicated? Um, and checking identity and understanding that context before you can move it forward. Once we've done this detect part, we can then decide how to handle things by setting up um what I call like guardrails and guidelines. You know, if if it's low risk and straightforward, then we may let the AI on behalf of the customer continue to execute and even go through our own automated services. But if it's complex or something that we think is a bit more, you know, is more personal, um, we might need to be doing some extra checks when it's the customer's AI, or we might actually need to encourage that it's be a human-to-human contact. And this will need to be customised based on your business, um, you know, reflecting what you know about your company processes and risk appetite. Um once we've decided how to treat it with these guardrails, um, we can then route it based on what we already know. Um, will we let the AI handle it and go through our normal path? So AI to AI route in many cases. Um, do we need our humans to step in and help out the customer's AI so that it's really getting the right specific answer it needs? Or do we actually need it to be a full human interaction? Um so routing and making those decisions is you know the third part of that stack. Um so basically, I mean, what the dual path means is that they're starting at the same spot, um, but they will then, with these three steps, go down different roads. And I think designing it in that way with those steps in mind will keep it efficient, make sure that you can cater for AI interactions, um, but you're maintaining great service and trust in your organization and data.

SPEAKER_01 7:25

Yeah, that makes a lot of sense. Um, so then once that's in place, you know, measurement is another area where assumptions can cause problems. So, why do KPIs need to be different for you know AI versus humans and then what goes wrong when they aren't?

SPEAKER_00 7:40

Yeah, um so humans at AI will have very different expectations about what good service looks like, and the KPIs we give each part needs to reflect that. Um, so an AI's expectation when they're calling you will be efficient, speed, accuracy. Um we wouldn't necessarily want to measure an interaction like that if it were a human that were asking us the same thing because we do want to look at resolution, we do want to look at branding, we do want to look at empathy. Um, if we apply the same measures to both, then we're gonna be getting it wrong. So if we're we're trying to apply human measures to what an AI wants out of us, um, we might be making those interactions more risky, we might be leaving things open to interpretation rather than looking at accuracy and speed to get the best result. Um, you know, I think we've got to think about what the service you're offering needs to offer the person or person's assistant that's calling and play to those strengths. So I think you know, when we're looking at AI, um we need to be offering it a service that is highly accurate, that it's you know clear and that it's speedy. Whereas if we're offering service to the human, then still empathy, your judgment, and personalisation still become as important. So the KPIs need to be balanced and different in each case. And that way I think you're going to get the service right, regardless of who's contacting.

SPEAKER_01 9:12

So obviously, this being a new technology, this idea of AI agents acting on behalf of human customers, you know, um, most companies are still in still early in this transition and understanding how it works. So, what do you think most enterprises are still underestimating here?

SPEAKER_00 9:29

Yeah, I mean, so I don't think many organizations are giving the thought that they need to to this at the moment. Um, and I think they're probably just underestimating how much things will change when it's an AI, making the call on behalf of the customer. And I think from my perspective, there's there's a few areas that will creep up very quickly and could get missed. Um, if I think about demand and how it could shift as a result of the customer's AI contacting you, if you haven't got proper controls in place, you could very quickly get overwhelmed by AI contact making you know lots of requests at an incredible pace, which could easily swamp your traditional routing and systems and you know your actual agent team. Um, and then I guess there's the whole question of over risk. Um, you know, you're not handling human contacts anymore. You've designed your processes to woo work smoothly so that humans can interact with you. Um, suddenly that looks very different when you've got an customer's AI involved, especially if it's been given permission to, you know, make decisions, raise issues, negotiate, or you know, complete a transaction on behalf of the customer. You've really got to think about what you as an organization want to do in response to that. Um, you know, it's about identity and consent. And I think those things are still sticky points, and you know, there's a level of uncertainty about you know how much are we going to let a customer's AI do on their behalf. So it's you know, it's vital for organisations to understand who built the AI, who's running it, what permissions has it got. And we don't have all the answers on best practice right now, and I think it will vary by industry. You know, we're no longer just designing about a great customer experience, we're having to now think about decisions and putting together more policy-driven models so that it can handle that interaction with the AI and still have those boundaries to understand what decisions you're prepared to let an AI make with your processes. Um, so rather than relying on old scripts and guided flows, we really do need to think about designing to respond to the customer's AI. And I think that's being heavily underestimated in a lot of the work that we see happening today.

SPEAKER_01 11:51

Yeah, right. Can I imagine? And then so for leaders watching this who are responsible for customer experience, you know, I guess the the challenge probably feels very technical as well as operational. Um, so if you could offer one piece of advice to CX leaders about what they should be doing to prepare for this idea of the jaw customer, what would it be?

SPEAKER_00 12:12

Don't dismiss it as if this is something that's far off in the future. It's not, it's already happening. Many of T-Tech's clients are already receiving calls from the customer's AI service. Um, so I think you know, the piece of advice I would give you is ask yourself a question first. Can an AI agent easily understand, verify, and act on behalf of your customers today? And if the answer is no, then dig deeper, look to three essentials, make sure you've got those guardrails and you understand what you want to allow the customer's AI agent. How far are you going to allow them to go on behalf of the customer within your organization? Your human team need to know this now because they could get a call from the customer's AI any day. You know, as I say, many CTEC customers are already getting these calls, and so preparing your human agent team to expect a call from an AI agent on behalf of a customer and letting them know, you know, what guardrails are in place, what should they be letting the AI agent do and not do is really important. Um, you know, as with any AI service, we're always telling you, you know, make sure your data is well organized and is easy to interpret at a glance. That's going to become even more important if an AI on behalf of the customer is trying to read your information. So well-organised, well-structured AI-ready data. Um and so I think you know, starting to think about those two simple steps and putting them in place now will make sure that you're ready to handle that. And then you can get more complicated about designing AI to AI services. But guardrails, clear information, how are those in place today? They're going to set you up brilliantly to handle this when it comes and it will come quickly. Sure.

SPEAKER_01 14:01

No, that that sounds like good advice to start. Uh so finally, for people who want to learn more about this topic, uh, where should they go? Um, as we say, to take the first step that they need.

SPEAKER_00 14:12

Yeah, so I've said that, you know, make sure your your content as it exists today is set up in a way that's easy for AI to interpret. Um and if you head over to the ttechdigital.com site, there's lots of information on there that shows you how we're helping brands master that dual customer era. Um, you know, look on there, look at the case studies, look at the examples that we're talking about today. Or by all means reach out to me on LinkedIn and we can talk more specifically about your industry and how we've been helping customers in your industry get AI ready. Great.

SPEAKER_01 14:49

Well, Carrie, thank you so much for sharing your perspective and walking us through this practical framework. Um to our viewers, if you found this conversation useful, stay tuned to CX Today's channels, where you'll find more coverage of AI customers, identity and access risk, and the changing shape of contact center operations. And to keep the conversation going, join the uh the CX Today LinkedIn community and also make sure you subscribe to our newsletter for weekly insights and highlights. Thank you for watching.