#733: Amazon Connect - So Many Cool New Capabilities For You to Use!

0:00This is episode 733 of the AWS podcast, released on August 18th, 2025.

0:09Hello, everyone, and welcome back to the AWS Podcast. I'm Les Sheehy with you. Great to have you back. I'm joined by a special guest today. I'm joined by Douglas Park, who also goes by Doug, depending on which country he's in. We'll talk about that shortly. And he is a senior specialized solutions architect for Amazon Connect. If he doesn't know about Amazon Connect, no one does. Welcome to the podcast, Doug. Thank you very much for having me, Simon. I I want to say that I'm a long-time listener, first-time caller, as they say. I actually, prior to doing this, went back and checked when I began listening to the podcast.

0:42And I thought it was right at the beginning, but you've been doing this a long time. So I checked in at episode 218, which was the re-invent update show from, you were in Vegas back in 2017. and the reason I bring this up is obviously I've been listening ever since then for all the reasons that people at AWS do, which is how do you keep up with all the things that AWS does? But I was amused to find it took me several months to realize that the Simon I sat three desks away from was in fact Simon Alicia from the podcast that I'd been listening to for several months because you recorded that from Vegas, so I assumed you're an Australian living in America.

1:23Ah, they're in Vegas. Yes, but you were Simon from three desk, I was like, this is how it works. So as I mentioned, Doug is one of our long-term experts in the world of Connect. And Connect is something that on the weekly shows I've often talked about, that they just release so much new stuff all the time. And so I thought it was well overdue that we got together and helped folks understand, firstly, why contact centers are actually important. If you're not in that world, you kind of don't understand that world, but it is a really fundamental component of business. But also all the new stuff that's been coming out and why it's coming out, like what the reason behind it.

2:05So we're going to go pretty deep. But firstly, I'm going to call Doug Doug because Doug and Douglas are used differently depending on which country you're in. Doug was born in Scotland and in Scotland, apparently you get called Douglas, but in Australia you get called Doug. Doug, and I'm assuming that if you get called Douglas now, you think you're in trouble from your mum. Well, I'm bilingual. Let's just leave it at that. I can answer to both. So, Doug, tell us about contact centres in the context of not just business, but also public sector organisations as well. Every organisation actually pretty much has one of these.

2:41What are they and why do they have them? Well, no matter what you're doing, and as you say, both across public sector and commercial, if you have an activity and you're performing some kind of service at a certain point, someone's going to want to talk to you about it, whether that is questions about the service you're delivering or the products, or indeed, whether they have trouble with what you're doing or reasons to improve it. And I think in the case of public sector, oftentimes it's a duty absolutely critical to the functioning of that organization is that people can call up and understand their plans, understand any changes to legislation and things like that.

3:17And I think where it becomes really interesting is where you recognize that for a long time, it was, maybe this is overstating, it was something that was sort of a grudge requirement for organizations. They had to have this to provide the service they were delivering. And it was effectively seen as a cost center. And with what we're doing these days and what we've done over the last few years, and Amazon Connect is absolutely the forefront of this, is how do you actually pull that into the main drivetrain of the organization and use the data that's being generated and the information about when realistically, and this is, I think, one of the phrases that really focuses people's mind on the importance of the contact center.

3:57It's often the only place where your customers or citizens are calling you up or messaging you and telling you what they want from you, telling you what the problem is. It's all just sitting there in these interactions, how you can improve product, how you can improve marketing, how you can improve communications. And I think that's where it becomes so important when you start to think about the contact center as the key interface to the outside world for an organization. It's true. It's true. It's interesting when we think about it. And I think particularly from an IT perspective, we think about the systems we build and the apps and the customers and citizens, They're interacting with the app and the web and it's all great and everything's fantastic.

4:39And we kind of forget that as good a job as we can possibly do, there'll always be exceptional cases. And we've all experienced them ourselves where, you know, I'm trying to do this, it's not working or I can't, or I've got to contact someone or jump onto the chat or I've got to, I have to interact with a human or an agent acting as a human or something. And you're right, that experience A is very important because usually I'm pretty frustrated by the time we get to that point. But also, you know, we talk about customer service and customer experience. It's about taking feedback and having that feedback loop.

5:13And from what you're saying, for a lot of organizations, that feedback loop is just a feedback sync. As in, you've got all these people talking to folks and gathering all this information and just sits there and never gets back to the people who can make any difference. So how does something like Connect help us close that loop? What's it actually doing to close that loop? Well, I mean, we can jump all the way forward to the really exciting stuff we're doing now, which is what people want to hear about, I think. But to level set, it is a complete contact center as a service. A large number of my customers, I would say upwards of maybe 40%, and talking to my peers, this is consistent, really just get the benefit of moving to the cloud.

5:55They get some cost out, but they effectively rebuild what they had before. Hi, welcome to the organization. Press one for this, press two for that. and you get through to a human who then has to work out who you are and what you're doing, all that kind of thing. From day one, Connect has focused on moving the contact center closer to the data and the data the organizations have about the person who's calling. And so that's the first step in changing that arrangement. So rather than a call where it is that high press one for this, press two for that. And ultimately, if you ask yourself the question, when was the last time I had an interaction with an organization where that wasn't the case, just shows how much work there is still to do.

6:33Like still when you call most organizations, that's your experience. That's your starting point. Welcome aboard. Yeah. And the job's all on you and the agent to work out how we're going to navigate this situation. So what we did from the outset is by, and I hope we can go slightly geeky here and I can refer to services such as AWS Lambda and say, we can invoke Lambda functions directly from the contact flow. So, that's the first step. So, we go from using no data to immediately being able to say at, you know, because Lambda is so compellingly cost effective. So, it's all, by comparison to historically, it's effectively free integration to data where you get to start to say, hi, Simon, you know, I see you called yesterday and spoke to Doug about this issue.

7:21Is that why you're calling? And all of a sudden, Simon has a sense that, well, well, this person actually cares about, this organization cares about me and is using the data it has about me. And I don't have to re-explain myself for the 47th time. You don't have to re-explain yourself. That's one of the key ones. But that was what we were doing. I mean, we launched with that seven and a half years ago. Well, actually, we just crossed eight years. Oh, on the 1st of July. I missed the anniversary. I also missed my wedding anniversary on the 4th of July, but that's a bigger problem. That's a whole other issue.

7:48It's a different podcast. But let's get into the really fun stuff now because what you spoke about there and what you touched on was the next layer, which is what are we doing with the data we're generating out of the contact center? And there's two prongs I want to tackle here. The first is the use of generative AI at AWS generally and specifically for Amazon Connect. And we've taken the approach at AWS, and I don't think this is a controversial statement, to be the adults in the room when it comes to generative AI. We're going to make sure it's secure. We're going to make sure it gives business outcomes.

8:23We're going to make sure it's scalable and it's not just a tool that allows your kids to cheat on their essays for school. It's actually a meaningful, reliable, you're not having your information bleed into the underlying foundational model. I mean, I know you've been speaking to the listeners about this since DayDot. So, how does this apply to the contact center? Well, what I've found is that a lot of organizations have a little bit of FOMO when it comes to generative AI. They say, we want to start using generative AI. How can you use it in our business? And the contact center provides some pre-built use cases where we can deliver instant business value.

9:04And with Connect, we took a step back and thought, well, we don't want to have our customers build that. because in the contact center, it's very different from the usual AWS user kind of profile in that our customers, contact center managers, tend not to be builders. They tend not to have app developers or developers on hand who have spare time and spare cycles to develop. They're using stuff out of the box. 100%. And Connect is reasonably different to most other AWS services in that it is a complete application as a service. It's not an infrastructure service. It's not a building block. It's the thing.

9:44It's the thing. And so the customers who use it expect the thing to do the things. And so the three big ones we launched with that have made an immediate impact, contact summarization, single tick box in your contact flow to say, hey, I want this call recorded. I want this call analyzed. And bang, I want a summarization of what happened on this call. There is a direct line to agent productivity improvement. Agents are not writing notes afterwards. They're now back on the phone quicker to take the next call or take the next web chat. But on top of that, so not only getting that productivity gain, you now have a succinct summary of every single interaction saying what is happening on this call.

10:23Now, you can start to then use a large language model if you want to ask questions of it. And the thought experiment that makes it come alive for most customers I talk to is, if you were to say to a C-level executive at your organization, would you like to know what our customers say to us when they call? Would you like to know what our highest spending customers say to us when they call? Would you like to know what the response was to that$100 ,000 you spent on that marketing campaign or that communications campaign? Would you like to know what's wrong with that product? Why people are calling?

10:54The C-level executive would not be able to sleep at night knowing that this information was available to them in terms of how they could improve. And they're not getting access to it. Yeah, exactly that. Now, the next - Just to round back on it, because this is all the data that's there that was always, you know, this is being recorded for training and, you know, other purposes. It's just been sitting there in S3. But it wasn't always used because you physically couldn't do it. Like, you just - Yep. Just reams and reams of information sitting in S3 unanalyzed, you know? And the scales to which large language models have changed how you can navigate that data and surface trends and things like that, as has changed so many things in the IT space and in the business space, it's really had a massive impact here.

11:40And this is what I mean about moving the contact center into the main drivetrain of the business. All of a sudden, the contact center can say to product, can say to service delivery, can say to marketing or communications, hey, this is what's going on. This is the problem. You know, if you fix this… Or here's what's working well. Or here's what's working well. In all, in full disclosure, people don't often call your contact center to tell you… It's usually negative feedback, isn't it? It's going well. It happens occasionally. Sometimes the absence of evidence can be evidence in itself in as much as, you know, if we've done something different and we're not getting a lot of complaints or people needing guidance, then you may say, well, actually, that was successful.

12:18Whereas, you know, every other time we've released something new or done something different, we get a wave of calls. So, that's also potentially a feedback. Absolutely right. Absolutely right. And that's where the data analysts and the people who are looking into this stuff can really, I mean, they can lose themselves in this. Like, they can go as deep as they like. Now, the only issue here is the contact center is often like Cassandra from Greek mythology in that they know all this information now. They can say to the product team, hey, you need to fix this, or the marketing team, hey, stop sending out that big mail out on Monday because we get swamped on Tuesday morning with calls.

12:56But the product team might say, hey, we've got a backlog of 12 months worth of stuff. Contact center, go whistle. So the second one is the one that really I think is getting contact centers excited. And I was working with some of the health customers over the last few weeks, turning this on for them and helping them configure it. And that is the agent evaluations using generative AI. because this is now, the reason this is important to the contact is you have control over the contact center and agent behavior. So, you can start to get all this feedback. And when you think about how poor and outside of the contact center, this might not be widely known, but the situation is dire currently in terms of agent quality.

13:32That is, you're listening to two calls per agent per month, or if I'm going to be generous, two calls per agent per week. And your chance of finding a skills gap, someone not following process or something that's not working is pretty low. And so, having generative AI listen to every single interaction and you can start to say, okay, our process is that you greet the customer by name, you ask for these pieces of data, you recite Mary had a little lamb and count backwards from 10, whatever your process is. Yeah, whatever it is. And then you say, so did the agent follow this process? And you'll get not just a yes or no, you get a paragraph of information back saying, hey, Simon did this well, he missed this.

14:10And then all of a sudden you can start to say, well, why is it when Simon answers a call, the customer never calls back and is satisfied and it's all done in three minutes yet when Doug answers a call, it takes 10 minutes and the customer calls back and is confused. Then we can start to, you know, all these kinds of, and you can do this at a quantitative level, but all of a sudden you're getting this really good feedback on what's working for your agents and what's not. The last bit, and I realize I've been talking nearly nonstop, but there's so much to talk about. The last bit in this generative AI will be familiar because of the name of it, which is Amazon Q and Connect.

14:45So that is Amazon Q, which is familiar to everyone who has anything to do with AWS for Q for business. And of course, Q developer, which has changed everything, certainly for my coding, is Q and Connect, which allows you to both make recommendations to agents, ragged only with the documents that you provided it. so you can say, hey, whether that knowledge store is in Salesforce or it's in ServiceNow or any other location or indeed just in S3, which I quite like because I think oftentimes people don't realize they have bad documents sitting in their document store. But you say, these are my documents in S3, my process documents, and it will generate based in real time on what the customer is saying, recommendations to the agent.

15:30Over and above that, we've then now turned this outward so you can now give answers to customers who call in or message in. So, you've now got a generative AI chatbot that is able to provide information only from the documents you've provided and lock down to just that information. So, the potential hallucination is almost eliminated. Where it gets even more exciting is the fact that you have complete control over the AI prompts. So, those AI agents, and you can tell it the persona, you can tell it what you want it to do, how you want it to do it, how you want it to behave. And there's two prompts here.

16:07One is obviously the answer generation prompt, but then the pre-processing prompt I think is almost more important. And I find this has a lot of purchase in public sector where oftentimes organizations need to be very, very careful about what they say. Very careful. Yeah. So using generative AI, not to necessarily generate the answer, but to better understand what the caller is asking or what the question is. And you might say, okay, well, these are the 200 or 500 things that legal and everyone else has said that I'm, and security said I'm allowed to say. Generative AI based on what Simon is asking, which of these is the best answer to give?

16:40And you're really locking it down. So, if you want to go to that, wherever you find yourself on the continuum of dynamism versus risk versus, I don't know, how you want to generate the answers, you can place yourself absolutely securely. And this inherits all of the guardrails we've got bedrock as well. So all the, you know, those hate speech and all that kind of thing can be very quickly eliminated. And all of this is wrapped up in our unlimited AI for Amazon Connect, but we'll talk about that in a second because I do want to pause and give you the chance to ask any questions about that dump of information.

17:18There's a lot, you're right, there's a lot to unpack there, but I guess this is about how do we make the experience for customers better, as in the customer calling in or also chatting in. I should point out that Connect also offers that same interface or a chat. So if anything like me, you actually don't want to talk to a human being, you'd much prefer to talk to the chat bot in your own time and get it done. But this is all about providing way more context in the call than ever before. And I guess potentially as we roll forward into the future of agentic ai etc is actually doing doing the things that need to be done on your behalf automatically rather than having to you know i don't know if you've had this experience lately but i i was on a call center call somewhere the other day i can't remember who it was was i won't i won't name the guilty but it was that classic you know i rang then they had to put me on hold because i'm going to connect to another system then they came back and then the system wasn't available so then they had to transfer me to the person you know it was one of those diabolical situations, which is basically the way I see it is it's an organization exposing their system integration failure to their customers.

18:27And I don't decry that in as much as, yeah, it's hard to integrate all these systems. There's often good reasons why, but as a customer, I shouldn't have to care about that. Well, and I think we've all accepted because of some legacy application cludge that's soaked up the market for so many years. And I'll talk through a quick thought experiment here to explain what I mean by that. If you think of mobile banking or web banking, which nearly every human uses these days to interact with the bank, a phenomenally good application. You can do everything. You can move money between your accounts, apply for loans, increase your credit limit, and more.

19:06More than you can do if you walk into the branch. If you walk into the branch, you can't do any of those things. You've got to ask someone behind Bulletproof glass who works for the bank to do it for you, but you can stand in the bank with your own little piece of bulletproof glass in your phone and do all of these things. And yet, if you click through that same device, the call me or contact us button and call the bank, all of that disappears. And the weird thing is when you tease it out, that application stack that you're calling into, the old contact center application stack, is often sitting in the same data center as that web application stack that has all this integration.

19:37And so, one of the key things we've tried to do with Connect and successfully done with Connect is to actually treat the contact center application similar to a web application with those same integration points, that same ease of integration that, you know, I mean, people used to talk about service-oriented architecture a few years ago, gone slightly out of fashion as a phrase, but basically that's it. Why is the web application stack different from the voice application stack? You've already got that integration. Your mobile banking is magnificent. Why doesn't the phone system have that same interaction.

20:09And the reason has been the old application stacks, the cost as well of integration. You're going to need CTI license, you're going to need an application, you're going to need a database. And did you get the enterprise edition license or the standard edition license, all this stuff. DynamoDB is my second favorite Amazon web service because of the number of years I was just desperate for a database nearby just to store some stuff. It's changed everything. So if we think about then, I mean, there's a lot of dimensions to what Connect has to work into, which I think is probably a reason why we see such fast iteration because there's so many inputs from customers coming into it.

20:49But if we've got, obviously, there's the customer experience itself. So me as a customer dialing in, we haven't even touched on the fact that there's this whole telephony piece that has to get done. Let's just assume that, but that's not a small thing, including some excellent pricing rates for inbound and outbound as well. But also then there's, so there's the experience from a customer standpoint, but there's the experience from an agent standpoint. So as the person taking the call, what does my job look like? Is it easy to get access to information? Is it hard? Am I going to be dealing with lots of people who are annoyed at me because I can't get the information?

21:20Or am I getting really good information to answer questions and getting great coaching? Then you've got the agent supervisors who their job is to make sure the whole thing runs smoothly and is optimized. I'm getting the right amount of people going through, et cetera. And then ultimately you've got that, as you mentioned, that sort of product data feed coming out as well saying, hey, here's what's working, here's what's not. There's a lot of pieces here at play. And one of the reasons why we're seeing such rapid iteration on the Connect service is because all these people are sending back ideas to the service team to say, hey, as an agent supervisor, I need to be able to give far more targeted training and I don't have any more time to do it.

22:03Help me out. And these types of questions. And I've said for a long time with generative AI that it's going to basically disrupt any knowledge-based value chain. And this is a great example of that happening where we're taking what we know how to do, but we're compressing it and we're also doing it for everything rather than that subset. So maybe help us understand a bit more about, you sort of touched on where AI helps. Where do you think it's going to continue to change things, but also where do you think it's not going to continue to change things? What are the areas of that value chain that you think just will continue to happen and need to happen, but we can augment them?

22:46Understandably, I spend a lot of time thinking about this. And of course, no one has the, you know, no one has the - There's no right answer. There's no right answer. So I think there's a lot of players in the market now who are thinking, let's reduce the number of humans possibly to zero involved in the customer service that we should be able to do everything. And I think as we've all experienced with generative AI at present and people always jump in with a, well, that's now, it's only going to get better. But I think we've all seen that there are limitations to where you see something happen or you feel you have an interaction with it and you think it's good.

23:26And it's kind of like, it's nearly human, but it's missing something. It's missing the spark. That thing that humans can do that robots can't do. And I've been hearing for a long time, and I think this is a good analogy, I mean, maybe for the last 25 years, in fact, I've been doing this for 26. So for the first year, it wasn't spoken of much. But after that, it's been no one will want to call, everyone will want to web chat. Yeah. The millennials only want to chat. And the data that's come back time and time again is everyone wants to talk when their anxiety is high. if you're getting a mortgage, if you're in trouble, if you're on the wrong side.

24:03Something's gone really wrong. Really wrong. I want to talk to a human. I want to call and I want to speak to Simon because I want Simon to have a human response to say, I understand you're in trouble. Yeah, you want empathy. And I don't think that'll ever go away. No. No matter how good our generative AI voices get. And look, the speech-to-speech models, like, I mean, Novasonic, yeah, I mean, most of the time I spend presenting to customers, I like to get the messaging right and make sure that they understand exactly where we're coming from. With the Nova Sonic demo I do, it's like, well, you just do the demo and it speaks volumes.

24:36It's so compelling. But even that, it doesn't quite have that empathy. It doesn't have that moment. And I think that will always be there. So, I think where I'm really excited about Generative AI in the contact center is all the things that we've spoken about, but just getting better and better and better, which is helping agents, superpowering them. I mean, I didn't even touch on the fact that with Q and Connect, you mentioned one of the personas there being the agent desktop, you can actually not only make recommendations of knowledge answers to agents, you can recommend entire flows of what are called step-by-step guides, the information or the process that the agent needs to follow to satisfy an interaction.