Author transcript
Hello and welcome to CX Today. I'm Nicole Willing. AI is now embedded in customer service, but there's a clear gap between what organizations believe is happening and what customers experience when AI needs to hand an interaction over to a human. Nearly every decision maker surveyed by Five9 in its 2026 business CX leaders report said that their organization preserves context during AI to human handoffs. Yet 83% of consumers said that they still have to repeat themselves uh you know at least sometimes after being transferred and that's more than inconvenience. You know repetition can increases effort undermines trust and it can turn a routine service interaction into a reason for a customer to leave. So to look at why AI handoffs still break down and what better looks like in practice, I'm joined by Mat Ladley who is principal solution consultant from Five9. Mat, thanks for joining us.
My pleasure to be here.
Great. So Mat, you know, I referred there to Five9's research, you know, identifying this gap between, you know, the organizational confidence and what the customer is experiencing. So you know, what should a customer reasonably expect when AI is passing a case over to a human agent?
Good question. So, first and foremost, in my opinion, at the very most basic level, a customer should never feel like they are having to start over again when transitioning from AI to a human agent. And that should be obvious, but the reality is it still happens. I've experienced it. I think everybody will have experienced it. And our research backs this up. As you said, Nicole, our data shows that 96% of business decision makers believe they are preserving context, but 83% of consumers say they still have to repeat themselves at least sometimes during this transition to a human. And so we do need to start to rethink about what we actually mean by a successful handoff. It's no longer simply did the call transfer technically successfully. It's more around did the experience successfully transfer.
So the perfect scenario is when a a customer is transferred from AI, they are done so at the right moment and to the right person who can seamlessly continue this conversation without any awkward interruptions or repetitions. They get the transition uh seamlessly with all of the relevant contexts from the AI and they allow the human to do what humans are good at doing. So offering empathy, dealing with prickly situations, making judgment calls, providing true expertise, recommendations, so on and so on.
So I I've I thought about many experiences I've had uh and I want to give an example of imagine that you have contacted your insurer about a claim. It's something everybody has to do at some point unfortunately. And let's say in this scenario you're you're speaking with an AI agent. You've you've already explained what has happened. You provided policy information. You've answered a series of questions. If the AI determines that the claim requires human involvement, then that agent should already understand who you are, why you're calling, what you've told the AI already, and what needs to happen next. The customer shouldn't hear a a question like, "Can you explain what's happened? Who am I speaking with?" Go through all the authentication information. Again, that's a particular book bearer of mine that still happens. Instead, they should be hearing something much more like, "Hello there, Mr. Customer. I see you've already given us the details of your claim. Let me pick it up from here for you." And it's a very different experience for the customer. And and that matters, I think, because as a as a consumer, as a customer of a company, we are not interested in automation rates. People are more interested in the level of effort. actually a resolution if their issue getting resolved and and overall about the experience that they receive.
So the goal here isn't just to keep a customer in automation for as long as possible. It's it's about using AI where it adds value and making the transition to a human effortless when that human judgment or empathy or expertise is needed.
Yeah. Exactly. So then you know as that handoff takes place to the human agent you know what's the the customer context that's being captured and passed on to the agent.
So that's an interesting point as it's not really a one-size-fits-all. I would say I'd say the type of interaction or the level of detail may vary from company to company or even from department to department or even from call to call. For me, the the most important thing is is that you're not just passing over a transcript to the the human advisor. Really, it's better to be passing understanding and relevant information that's going to help that that advisor. If we go back to my insurance claim in my example, the AI may already have established the the customer's ident identity and authenticated them. Uh they've found their policy information. They've identified the reason for contact, the details of the incident, all that information, as well as some of the information, maybe the sentiment that the that's been detected by the the caller at that precise moment and the level of urgency that they need to to get something resolved. All of this information can be brought together and presented to the human agent in a useful way. And even that way itself of how it's presented can vary from business to business. It can vary from interaction to from interaction to interaction as well.
Sometimes the best method may be presenting all of his data side by side in a quick overview for uh an agent to review when they when they answer that call. Other times it may be embedding that information within the the CRM record or the case management system that the agent can instantly see. So in our example of if the caller has already gone through half of the insurance claim information with the AI agent, then it's going to be beneficial for the contact center agent to automatically get a screen pop of that claim with the information prefilled in so that they can seamlessly continue the conversation. And that usually is accompanied by a relevant succinct summary of the the journey or the interaction so far.
So what's interesting to me is that the more routine interactions that we automate with AI, the conversations inevitably that reach human agents are increasingly likely to be the more complex and sensitive or emotionally demanding ones. So actually it's more important than ever that people are prepared for this like higher value work that they are undertaking and in this scenario the AI is is not simply replacing their work but it it's being used to support them whether that's with a a handoff of information or even then continuously supporting the conversation uh as it progresses and for the customer as well it means less repetition from their side. So, there's less friction and it's a genuine feeling that they're having a a single continuous conversation uh regardless of whether they're speaking to AI or a a person.
Yeah, exactly. So then, you know, given all of that nuance that's required, how does orchestration play a role in here in determining, you know, the right step, the right agent and the the right information that they need to carry forward in the interaction?
So orchestration is a big topic at the moment. Uh it's about knowing where to pass a customer to and and when where they should be actually where where they can ident uh correctly uh transition to a person to get their issue resolved and have a good experience. And that may be to a human agent. It may even be to a specialized AI sub agent.
The fundamental thing is to remember that a great customer experience isn't about automating everything you possibly can automate. People do not like feeling trapped in an AI loop. So an 80% automation rate might look great on a contact center dashboard, but if customers are just abandoning calls, having to repeat themselves, having to to phone back in, becoming frustrated, eventually getting through to a person, having that bad experience, then that automation rate is not telling you the whether it's actually successful or not.
So a better question is to look at what's the best next action for this particular customer at this particular moment. So we should never be trying to automate everything. You need to build in business rules into AI agents to know when and where you should escalate an interaction. Just as companies have escalation procedures in place internally for managers or for priority issues, AI needs this too, it it should not be left to chance for an AI to guess these things, we need to build those in the correct processes. So if we look at our insurance example again, if the AI identifies there's a complication with our claim or if a customer's becoming worried or frustrated or the situation needs that human judgment, this should not be automated but should be handed over to a human with the context. uh the saying goes if uh there are any technical issues we we we cannot reach a particular system or the caller is not being understood people should never feel trapped with AI and in fact Five9 research has also found out that when customers can clearly see that human help is available to them and they're not being forced down an AI route always then their trust in AI's ability increases and the their trust that the AI can resolve their issues uh rises significantly as well.
So a well-built AI assistant can recognize all of these signals and determine the the best next step isn't always another human isn't always an automated response. It can be a human who's needed. And so that orchestration layer can be used to then help identify the right agent based on skills or attributes, intent, availability of the nature of the issue, customer preference, whatever it is. And it can carry forward all of that relevant context plus anything else that's happened so far. And that's why we should not be thinking about AI and humans as two separate channels. The opportunity really is is AI and humans working together as one connected experience. And the orchestration is deciding dynamically where each one adds the the most value.
Sure. And you know, as you mentioned there, we're all familiar with, you know, the conventional IVR or chatbot flow that we've had experience with in the past. How does, you know, an agentic voice AI approach change that handoff experience, you know, especially if we're talking about a customer in, you know, with sensitive financial information?
Yeah. So the the big difference for me I'd say is that we're moving from just rooting an interaction which we had in a traditional IVR or a chatbot and we're moving towards actually understanding the the caller's intent and then acting upon them as well. So a chatbot or IVR would always work through a predetermined decision tree. It's a scripted experience. So at the end of it, you know what information is being gathered in in what order and you know then what the best output is to present to a human. But with agentic AI like Five9's voice AI agents, the experience has changed. So you can have a much more natural conversation and that is likely going to be more fluid. It's not going to be linear. Consumers veer off in different directions as people do in conversations and the AI agent can adapt to this and they can truly understand the reason about what the customer is trying to achieve and AI agent can take action where appropriate as well and at any point that's needed it will escalate to a human agent with a completely tailored summary and additional context for that specific unique interaction.
And voice is always fundamentally going to be different to digital channels where the experience is not always fully real time. So on on a phone call, the handover needs to be highly pertinent. It needs to be instant. You can't have a a caller waiting while a specialist is reading through a massive list of information or having to manually search some some context that's specific to their interaction. any kind of latency or silence here can be damaging to the customer experience. So with with a gentic AI that information can be uh kept relevant and succinct for the specific interaction. And if we look back to our out of ongoing uh example around a claim with an insurer, the AI can understand what the the caller is asking about, gathering information, uh resolve uh elements of that inquiry, but critically it can recognize where the conversation has reached a point when the human is a better option. For example, if the caller is is concerned about their financial situation uh because of the claim that's that's going on and impacting any uh their mental well-being or anything like that of the AI can identify that because remember good automation is not automation at all costs of we need to understand when we escalate to a human and and make sure that that transition is is done seamlessly so the human doesn't enter the conversation blind. mind they understand the situation what's been said what the AI has done already and why the interaction has been escalated or whether that is because of a a a vulnerability or a particular sentiment that's been detected. So that's where agentic AI becomes really powerful in my opinion. We're not creating a barrier between customers and people. we're actually allowing AI to handle what it does well and allow the humans to handle what they do well and we make the transition between the two uh practically invisible to the customer.
So for me that's really the the real measure of of success and the successful AI in CX. It's not how much did we automate but it's really did we resolve the customer's issue and did we make the overall experience better.
Absolutely. So can you share with us now a demo of of how the Five9 system works in this way?
Yes, absolutely. So I have I prepared a a short demonstration here of really talking through some of the themes that we've spoken about already. Uh we're going to carry forward this customer uh example in the insurance space where I'm a caller wanting an update on my claim. The conversation will start with the Five9 voice AI agent answering the call and actually being able to answer some questions, authenticate the caller, but importantly understand where there is real cause to escalate to a human agent and and do so in a seamless manner. And we'll then see that hand off to a a human advisor with all of the relevant context and see how that information can be maintained as well throughout the conversation and the AI can even continue to assist where appropriate. So if I share my screen here and what I'm going to do, I'm going to place a call in.
Hello and thanks for calling Five9 insurance. I'm your AI assistant here to help with your claims, questions, or anything else you need. How can I assist you today?
Yeah. Hi there. I need to check what's happening with my claim.
Sure, I can help with that. I've matched your phone number to a claim, but for security, I'll need to authenticate you. Could you please provide your email and the first line of your address?
Uh, yeah. It's [email protected] and it's 7 Pattern Street.
Thanks, Mr. Ladley. You're now verified. I can see your claim here. There was an update yesterday and the status is currently on hold with the reason awaiting underwriting document review. Can I help with any more information?
Uh, okay. How long do I have to wait still?
Let me just check that. So, the current review process for claims of this nature can take up to 72 hours for additional approval. The latest update will be updated. So, expect
Okay. Uh, I don't know if if I can wait that long. This this whole process is just stressing me out. I don't know if I can afford all this. I'm not really sure what to expect to be honest.
I'm really sorry to hear that, Matthew. Don't worry. We're here to help. I can connect you with a specialist right now who can guide you through the process and address your concerns. Would you like me to do that?
That would be really helpful. Thanks.
Okay. So on screen now you can be seeing the caller being transitioned to a human agent and that orchestration layer there has has taken all of the relevant context and identified the correct agent to deal with that interaction.
On screen here you're seeing an example of how that AI information can be presented through to a human agent. In our scenario here, we have a an agent working within the Five9 agent desktop and I'm using an agent script to present this AI information alongside uh pertinent customer information about the the ongoing claim itself of identified policy information of anything uh that has been captured during that AI conversation itself can be presented in a number of ways alongside traditional contact center data as well. We can see that the caller has been identified and each step that they have taken to get here and even a full AI summary of this interaction so far. So the idea is that the human adviser can start the conversation in a positive way rather than having to ask the caller to repeat themselves. So for example, hello and welcome Mr. Ladley. I can see that you have spoken with our AI assistant and you have got an update on your claim. You still have some concerns about the uh waiting time. Is that right?
And what we're seeing here as well, if I look at agent assist, this is where AI can be continually monitoring and supporting the human adviser during the interaction itself. So in the same way this is a different format. We can present information through to the advisor. We information can be pushed through to a CRM record if we want to do a screen pop of that automatically as well or embed the agent experience within to a case management system. There's many possibilities as we discussed earlier around what the best experience is for a particular company or a particular type of call. Uh, in this scenario, the AI is presenting guidance information, a task list of of what is being done. And all of this information can dynamically update depending on on what's being said by either the caller or the the human agent. For example, if the caller says something like, "Ah, yeah, that's right. I'm really worried about my financial situation." We can see the full transcription coming through here. And u what else is is any information that has been identified can actually trigger off some actions as well. I'm really concerned about my call.
And for example here we can see that we have suggested that there is a possible vulnerability here based on what's been said and the information that's being captured so far. So we're presenting this guidance information to the uh contact center agent and likewise if the agent has to do a search in a knowledge uh base here for example in our case it's going to say how to escalate this claim. Now this is going and and using an AI search to to bring that information. But once again this is keeping the context from the AI agent conversation as well because we've already identified the caller and we have identified their particular claim type that context can be passed through to this search as well. So that the response here that we get is highly relevant to the uh particular question this particular customer and the example here as well. So we're supporting the the human in this process to make everything uh much simpler for this uh for this interaction and ultimately providing a better customer experience.
And the final piece, we when this call finishes, we of course get a an AI generated summary of this interaction. And we're not just summarizing the human to human part of the conversation. We're again looking at the overall journey here. So we can be taking that AI agent transcription summary, any context that's been captured and making sure that that is being presented within the call summary as well. And of course this can of be automatically pushed into a a CRM or case management or in our case the the claim record as well. So that context is maintained throughout the uh conversation itself and is stored for longevity. So we are showcasing here that AI to human handoff in that seamless way and ensuring that the advisor can continue that experience to provide the the best uh CX for the customer.
Well, thank you Mat. I think it's really valuable for our viewers to be able to see that to that demonstration and how it might differ to what they might expect about a voice AI interaction. So thank you for for joining us and sharing your insights. Kasha and to our viewers um for more on Five9's 2026 business leader CX report and its findings you can go to five9.com and you can also find other resources and um information about the product there and you can go to CXToday.com for more interviews, news and analysis on the latest in customer experience technology. And don't forget to join us on LinkedIn to continue the conversation. Thanks for watching and we'll see you next time.
