ACP: The Amazon Connect Podcast

13: AI in Amazon Connect

July 08, 2024 Episode 13
13: AI in Amazon Connect
ACP: The Amazon Connect Podcast
More Info
ACP: The Amazon Connect Podcast
13: AI in Amazon Connect
Jul 08, 2024 Episode 13

Send us a Text Message.

In this episode of ACP, Tom and Alex dive deep into the exciting world of AI in Amazon Connect. They discuss how AI is transforming the contact center landscape by empowering agents, enhancing customer self-service, and providing supervisors with crucial insights. Key areas of focus include AI-driven agent assist with Amazon Q, comprehensive speech analytics and sentiment analysis through Contact Lens, and the integration of AWS Lex for advanced chatbot functionality. Tune in to explore how AI tools are optimizing contact center operations and delivering improved customer experiences.

Find out more about CloudInteract at cloudinteract.io.

Show Notes Transcript

Send us a Text Message.

In this episode of ACP, Tom and Alex dive deep into the exciting world of AI in Amazon Connect. They discuss how AI is transforming the contact center landscape by empowering agents, enhancing customer self-service, and providing supervisors with crucial insights. Key areas of focus include AI-driven agent assist with Amazon Q, comprehensive speech analytics and sentiment analysis through Contact Lens, and the integration of AWS Lex for advanced chatbot functionality. Tune in to explore how AI tools are optimizing contact center operations and delivering improved customer experiences.

Find out more about CloudInteract at cloudinteract.io.

Tom Morgan:

It's time for another episode of ACP and I'm joined in person by Alex. Hello, Alex. Good to see you. Good to

Alex Baker:

be back again. Yeah, definitely.

Tom Morgan:

It is. This episode we're going to be talking all about ai because it's very much ai season. And we're living in the ai era and And obviously ai has infiltrated all sorts of bits of tech And amazon connect is kind of no exception to that and it can sometimes be a bit confusing about you know Where the ai is and what it's for and what it's doing so I know Alex, you've kind of split down like some of these big areas where AI is kind of making a difference in how Amazon Connect is sort of delivering for people.

Alex Baker:

Yeah. Well, it wasn't, it wasn't really me. So this is, I've seen Amazon suggesting some focus areas. Apologies if we miss any that people think are really important. You know, it's like you say, it's kind of a growing and exciting place to be at the moment, always happy to have the suggestions of other things we can talk about. But the, the broad areas of focus as it comes to Amazon connect are around agent assist. First of all, empowering agents to be able to handle more complex queries. By giving them some, some really useful at their fingertips, contact sensitive information. Yeah. Then end customer self service, being able to give customers a more personalized, a richer journey through the call or contact flows, allow them to more easily self serve their queries. And then the, the third area of focus is around the sort of supervisor or manager or quality analyst type, you know, more managerial type persona allowing those, those types of people to get a deeper understanding of how the contact center is doing. So maybe looking at things like speech analytics and the trends within it, or being able to query the contact center data using natural language, rather than sort of having to use SQL queries, and we're seeing that loads with things like chat GPT, right? So that kind of the chat, chat bot type application. Why not extend that to the contact center data? You might be able to interrogate your data in a much more natural way. Ask a chat GPT like service. How's my contact center doing? What are the pain points? Where, which cues do I need to allocate more resource to? That kind of thing. I can see it being. That's

Tom Morgan:

really interesting. So it's very much about. Empowering, giving existing job roles the tools they need to do their jobs better. Like on the front end, it's, it's giving, helping customers be sort of more self sufficient, but then when they get through to agents, those agents are better empowered to help them, and then the supervisors and managers are better empowered to understand what's going on in the contact center. Because they've got not just access to data, but access to kind of that data processed, if you like, or you know, the information that comes out of that data.

Alex Baker:

Yeah. And having it having the right data at the right time, I guess, is really key to it, isn't it? Yeah. Yeah. We, we know how much, how much data is generated by things like contact center platforms. And we know how much data is probably there in the background around. We've talked about CRM, customer relationship management systems. how complex and massive they can be for an agent to be able to navigate to the right bit of data at the right time, I think is powerful. It's, it's, I can see it as being a big time saver and it just allows them to do what they're doing a lot more efficiently.

Tom Morgan:

Yeah, absolutely. Okay. So there's various kind of product names that have been kind of flying around in the Amazon connect space, delivering AI. I'm going to start with Amazon Q. Because I feel like it's the one I understand the most. It's a high level kind of what, what is Q? How would you describe Q to someone who'd never heard of it? Because it sounds like, I mean, James Bond, right? It's the, it's, it's the assistant. Yeah.

Alex Baker:

Yeah, definitely. And, and exactly that. So Q is. AI personal assistant effectively. And I mentioned chat GPT previously, but I guess think of an Amazon flavor of chat GPT kind of under the scenes, but baked into the, the agent application. And it is, this is working mainly in that agent assist area of focus. The first one that I mentioned and it's. The personal assistant is giving the agent access to that, the contextual information based on either them searching a connected knowledge base. So it's, it's possible you, you get, I should say this, you get in the agent application in connect a Q sort of sidebar on the right hand side. And that's where you see the, the information surfaced. You can. There's a search bar in there, which you can enter a search into, and it will query that connected knowledge base. But the other thing that it does, which is, is really cool is if you have that the, the real time speech analytics capability, which we'll talk about in a bit more detail, if you have that turned on, then Q will react to what's being said in the call and it will try and surface the most relevant information. I guess the, the key phrase is next best action. A lot of people refer to it as, so this thing's happening in the call or in the contact. What's the best thing that the agent can suggest in that particular context?

Tom Morgan:

Got it. Okay. So it, and it knows, so it's, it's connected to some corpus of data that you've previously set up. So you're describing your products, like your policies for returns or I don't know, the things that an agent does on the call. Right. Whatever that looks like. Okay. So it's a. It's a sort of rag. So in AI terms, it's, it's a rag based LLM presumably, cause it's then grounded. It's not just kind of making answers up. It's grounded in your, your data. But then additionally, it's got this kind of layer on top where if it's got access to the transcript, then it can kind of be a bit proactive about kind of suggesting things proactively, which is kind of smart because I think a lot of hurdle we have to overcome with AI is remembering to use it at the right time. Like it's a new tool, like none of us had it three years ago. Yeah, it's, we have to remember when it's the right time to use that tool. And so actually a bit of proactivity is, is probably quite a good idea, I think.

Alex Baker:

Yeah. Yeah. If it's there in your agent workspace that you're using on a day to day basis anyway. Then it makes total sense, doesn't it? If it's just pushing that information to you rather than you actively have having to go and search for it.

Tom Morgan:

Yeah, definitely. And I think you can connect it to various different sources of data. Can't you? I'm not gonna make you list them, but I know SharePoint's in that list. I

Alex Baker:

was going to say, yeah, I'm sure I probably can't list all of them off the top of my head because there are, there are lots definitely we've experimented. With Salesforce, for example, as a source of data or SharePoint, as you mentioned, SharePoint's a great one for us when we're doing demos and things, because we have quite a lot of information held within SharePoint. So we can point you at a particular area in SharePoint or at your whole SharePoint set of folders, if you wished, and just tell it to go and find the information that it needs within there.

Tom Morgan:

Yeah, I could see it's been really useful for like, you know, it help desk tickets or something like that, where, you know, the ticket probably already exists somewhere else. It's just finding it quick enough and, and all that sort of stuff. So that's really cool. And in your experience, like, how well does it work at that? How good it, you know, real world usage, like, what are you seeing?

Alex Baker:

I think. Probably a lot of people will say this, and it hopefully doesn't sound like a cop out answer, but it really depends on how good the data is underneath it, you know, if it's, it's not going to magically solve a problem where if you've got a whole load of junk in the connected knowledge base, it isn't going to make that better. So you, you need to have the right information in there, it needs to be somewhat structured. And yeah, you, you don't want it to, you don't want lots of information that probably isn't relevant to the, the tasks that are being carried out, I guess. So when I said that you, you could point it out your whole SharePoint set of folders, you probably don't want to do that. You might want to, you might want to filter it down a bit.

Tom Morgan:

Yes, absolutely. And this management of data is a reoccurring theme That people, you know, we come across people readying for AI and it's It can be a blocker like, you know Companies are turning off AI everywhere because they don't have a good data management policy right now. And so they can't confidently say Yes, this thing is, you know, just looking at the right data or not I think this setting up queue is slightly easier because it's very much an opt in system. It's different from you know, the I mean, if you were to compare it against a Microsoft 365 co pilot, which is, you know, turned on on your tenant there, it's going to look across your whole tenant, whereas this is very specifically, you are telling it which knowledge bases to look at, because it's, it's a very contextual, it's serving one need, right, which is to help agents in that particular contact center on that particular queue, do a particular thing. So it's much easier to say, well, these are the documents that serve that, you know, here is the data that helps serve that role. So one big thing about Amazon queue is the cost and it's not so much that it's expensive, but it's not consumption based. It's fixed cost, right? It's 40 per month per agent, whether you use it or not. Yeah, it

Alex Baker:

goes on the first point. It goes slightly against the general consumption based billing of Connect as a whole, which, probably was a bit of a surprise when that was released. I guess maybe I can see why that is, you know, that there's probably some pretty hefty sort of investments in the infrastructure. Yeah.

Tom Morgan:

Technically I can totally imagine like they, as soon as you specify that knowledge base that, you know, it needs to go and build like a vector database on that and it needs to keep it up to date. It needs to scan it. There's a ton of compute that needs to happen there.

Alex Baker:

Yeah. We know that. If you're, if you're attempting any of those things that the vector database thing, for example, you can build that, that kind of database yourself, but actually it can get quite expensive quite quickly to do it and maintain it. Yes. The one thing just to mention I think it is kind of consumption based in that it's, you don't sort of pay an annual license or anything. It's if an agent is. An agent needs to use it basically to, to trigger the, the 40 per month. If there and you know, it's probably made quite easy to use it as it's there. And it would, if you had the real time speech analytics capability turned on, certainly it would automatically be, be using it, but what you could do potentially is you could maybe set up a percentage of your agents to, to use it or. Only enable it on a percentage of calls, maybe and try and evaluate it before rolling it out across the board.

Tom Morgan:

Yes. So if you're an IT admin or a connect admin, so it's, it's 40 per agent per month that is triggered on the first time the agent uses it. And there's no, there's no grayish period. It's like literally one search. That's 40. That's an expensive search, two searches, 20 each on off you go. And then, but you can, you can turn it off presumably or cause I know the default experience when it's turned on is that it's fair. It's, it's fairly prominent in the, in the connect UI, isn't it? And it, it fairly wants to be used. So it, it kind of easy in your face a little bit. So yeah, can it, can admins turn that off or hide it?

Alex Baker:

Yeah, well, there's a couple of, a couple of points there. So you have in the, in your inbound contact flows, you have a, a node where you can enable queue. So you, you can do things like, as I mentioned, you could put a maybe a percentage split before that and just turn on 10 percent of your contacts to go into queue, for example. Yeah. How that would actually work with, because it's on a per user basis that it's billed, I guess you could, you could only have it on 10% of calls. Mm. It's still gonna get split across all your agents, isn't it? Yeah, exactly. So you, you, you need to think quite carefully about it. Maybe ring fence and an entire small team to evaluate it. The other thing I was gonna mention is in security profiles as an admin, you can. Enable or disable the, the, the queue applicant part of part of the application.

Tom Morgan:

Hmm. Okay, cool. I mean, the other thing is, you know, it it's 40 a month, which sounds like a lot. But you know, that's 2 a day for an agent. Is an agent going to get 2 worth of value? I mean, maybe like if you, if you've got good data and you, you know, if it's saving agents time, if it's stopping them, transferring the call out to someone else. Absolutely. It proves its value, doesn't it?

Alex Baker:

Yeah, and it's probably something you can fairly easy put a value on. So whether, if it's time saving, you probably have a good idea of how much your agent's time is worth. If you can prove that it's just by giving them access to the right data more quickly. If you're saving 30 seconds or one minute per call. And you're saving that across 500 agents, you can quite quickly put some sort of number on it. Definitely. And then if you're, if you're using it to suggest next best actions to, to try and up your, your sales again, if you can prove that it's having that, that tangible effect on improved sales, that's quite easy to quantify. Yeah,

Tom Morgan:

absolutely. Yes. Amazing how I don't know. Yeah, good, good sales comes from just saying the right thing at the right time, doesn't it? So as I was reminded, yeah, I was in a shop yesterday and I went to buy, I went in with the express purpose of spending about two pounds on something and came out having spent like 10 pounds just literally because it wasn't even a hard sale. It was, Oh, by the way, this is a sort of related thing that might be useful. And I was like, actually, that's perfect. Like, and I wasn't planning on doing that, but it's just, yeah, that's, yeah.

Alex Baker:

Yeah. That's a great example though. You know, if you, if you have, if you keep things like your, your offers really up to date in the underlying knowledge base, we've set up some POC type things around this for financial organizations. As an example, you know, we, we, we have this certain mortgage deal, or we have this certain rate on our, our savings accounts that you can take advantage of right now. It's really good. It's a limited time offer. If that's being served to the agents automatically, they may not know that they may not have read it in a an email or something if it's being served to them right at the point of the contact with the it's literally

Tom Morgan:

easier to read it out than to say anything else. Isn't it? So yeah. Yeah. Okay, cool. So that's that's Amazon queue. The other big area is contact lens now. There was contact lens around before this AI explosion happened, or is it quite new? Because I'm not quite sure I know.

Alex Baker:

Yeah, it is. And. I guess to some extent because contact lens covers sort of everything workforce optimization, I guess you'd probably call it. Refer back to our discussion with the guys at Calabrio for some, some more eloquently put details about that, but it covers from a base level contact lens covers call recording. So something that you would expect out of the box with a contact center solution these days. And something that isn't inherently sort of AI driven. But it also goes through to kind of full speech analytics that I mentioned, kind of the trigger for the real time the real time next best action stuff. It also includes screen recording and it also includes quality monitoring. So you sort of evaluation forms that you can evaluate your agents with and all of that. Workforce optimization type tooling is being bundled in under the contact lens umbrella. So like I say, to some extent, you'd say maybe that's not AI, but when you get through some of the rest of the tooling, yes, it does bring AI into it. I

Tom Morgan:

see. Yeah. And can you. Enable those things all separately. Are they all separate features that you can kind of toggle within that contact lens box, if you like? Yeah, yeah,

Alex Baker:

exactly that. You hit the nail on the head. So it's another another box, another node within a contact flow where you've at the top, you've got kind of turn on call recording. You can do that for either both parties, or you can have recording only for the agent or only for the customer. Yeah. But within that same box, you can turn on your screen recording, you can turn on your analytics, you can, at the point of turning on the analytics, you can say, okay, I'm not too worried about the real time analytics right now, but I do want to get the, the call transcribed and analyzed after the event. So you can have post call versus real time analytics.

Tom Morgan:

But I guess if you do have the real time, that's where we're, what we're talking about with the Amazon queue stuff, because it's got access, those two things can talk to each other. And that's how Amazon queue can be smart about, you just said this, why don't you do that?

Alex Baker:

Yeah, exactly. Yeah. You've got it. So yeah, if you, if you want, if you want that next best action type thing in queue, then yeah, you want to be turning on real time analytics. The other thing to mention, and this, this is where more of the, the AI Tooling under the hood is coming in you can do things like turning on PII redaction So if you're in a contact center that that takes maybe card details or health care information That kind of detail you can ask contact lens to automatically try and detect and then filter out Okay, and it will filter it out Both in the the audio recording if it's a voice call and also in the transcript of the you know The analytics transcript

Tom Morgan:

got it is this is contact lens also where the sentiment analysis comes in because I've sort of seen reference to that Is that another sort of feature of contact lens? Is that it it can work out the sentiment of the call and is that also that can be post and real time? I think as well. Yeah,

Alex Baker:

absolutely. Yeah. Yeah, so you potentially got some You Interesting use cases. If you're doing it in real time you can do things that there's quite and the capability is growing. But there's some quite good capabilities for setting rules up around contact lens. So, for example, real time rules where you have a negative sentiment, for example if you if you have a call where the customer is, you know, Is negative throughout, you might want to trigger a rule, which then notifies a supervisor just to get them to take a look and maybe start live monitoring that call and see if there's something they can do to interject.

Tom Morgan:

Yeah, we're all going to be like, putting on our best sad voices next time we need to use a contact center just to try and get that extra discount code or get the supervisor on to escalate the case.

Alex Baker:

Yeah, or throwing in, I guess, things like. You know, complaint, for example. Yeah, yes, right. Trying to get all the keywords. As soon as those keywords hit.

Tom Morgan:

I'm very well today. I wouldn't dream of making a complaint in a sad voice.

Alex Baker:

Yeah, there's all sorts of scope there for messing with the system, I guess. Social

Tom Morgan:

engineering, the the tech. Yeah, yeah,

Alex Baker:

yeah. Getting yeah, because you're right. Potentially you could, if you were If you were sort of routing calls off the back of that, which is a possibility, you know, assigning a high level of priority to somebody you think has a complaint, for example, you could possibly game the system in a way that's an interesting, interesting use case there.

Tom Morgan:

But yes so that's, that makes sense. And that's much more really on the kind of helping the supervisors and the managers as well, isn't it? Like the, there's so much stuff that comes out of that, like, yeah, the, the trends, the sentiment trends, you know, what's how different cues are performing. And yeah, there's lots and lots of scope for, for AI in, in all of that.

Alex Baker:

Yeah. Agreed. Yeah. And, and things like, The whole QM, the quality monitoring part of it is interesting as well. So increasingly you can automate your, your QM process. So using the analytics, you can have a QM evaluation form set up in Connect, which traditionally. It probably wouldn't have even been native to the call recording tool. It would have been in a third party system. You can automate the, the filling out of that evaluation form with these rules that I mentioned. So sentiment might be something that you might include in that. Was the, Was the agent sentiment broadly positive throughout? Was the customer sentiment positive or was the customer sentiment turned around from negative to positive at the end of the call? And another thing that we've we've, we've set up for customers is things like compliance monitoring where it may be in financial services, for example, or healthcare. where the agent has to say a particular thing to a customer. You think of mortgages, for example, if they have to say a certain compliance phrase, you can get the, the analytics tool to detect whether that's been said and immediately flag it up to the supervisor, but also maybe automatically fail your evaluation

Tom Morgan:

form. Yes. I'd be fascinating. So once we get some really good data from, from some of these kind of Bigger organizations and some of our bigger customers to just dive in and I know we're doing some of this kind of Analysis of data as well, but even things like sentiment analysis over the course of a day You know are there happy times and sad times and what can you do? Like is is that an action that you could you know? bring some extra resources to bear on the sad times to try and you know, bring them up a little bit or Yeah.

Alex Baker:

Yeah. And any other correlations. Yeah. And that's the great thing, isn't it? Because all this data is produced. People are only scratching the surface with what you can do. Absolutely. Correlation between all sorts of things at a basic level. Maybe have I spent half an hour in the queue? That's probably going to drive my sentiment down a bit. Those, those kind of correlations I think are really interesting. Yeah. It's a great scope for analyzing the data. You know, having a sort of unified contact center data platform that you can then run queries off with all this data.

Tom Morgan:

Yeah, definitely. Because there's so much data, as you said earlier, there's so much data being produced. And it's only really now that we've, we've got a chance to do Some interesting things with it. I think, I think we've always been able to obviously do the reports that we've always been able to analyze the data from produce the reports. I think our challenge has been, we haven't always known what reports to produce and I say, we very broadly, but like, you know, contact center service owners, you know, technologists haven't had, you know, you could make so many reports and you just don't really know where to go. I think the, the, having this layer of like, you could just sort of describe the thing you're trying to solve and it can really help with some of that as well, as well as producing better data as well. Like the, you know, the analysis, the sentiment, the summaries, the, it can do a good job of producing better quality data as well.

Alex Baker:

Yeah, great point. We haven't even, haven't even touched upon the. AI or gen AI generation of, of post call summaries, for example, that was something that we covered it so, so

Tom Morgan:

useful for agents who have to kind of update the CRM or whatever, send an email, summarizing

Alex Baker:

the call, all that sort of stuff. Absolutely. And also for, it's really useful for the agent and a great time saver, you know, maybe you can. Knock two minutes off your wrap up time. Cause you're not sort of typing up detailed notes, but also for the, the supervisor or manager or QA, who's going through and reviewing calls, they can see that couple of sentences of summarization and get a real feel for what's going on without having to actually listen through and fast forward to different bits of the

Tom Morgan:

course. Yeah, definitely. So where does Lex fall into this then? Cause we've had Lex for a while. Like I remember I played with it. Remember first playing with it a couple of years ago when you sort of write your first skill for you know, I'm not going to say the lady in tube. So there's one in here, which is a terrible idea for a student. Yeah. So so where does, where does that fall into this is, you know, because it's not really part of connect, right? It's more part of the broad AWS ecosystem.

Alex Baker:

Yeah, it's, it's a standalone service, Lex. So. Yes, you can, you can use it for creating chatbots, but it doesn't have to be linked with connect but Handily, it does, it does integrate really well with connect as you would expect. How, how has Lex come on recently with the advent of gen AI and large language models? What I would say is that previously on Lex was, was great. And like many of these conversational interfaces, you could design a really, really rich self service experience. It was probably a bit more prescriptive though. So you had to have a really good mapped out idea of. what the intents were that you're, you were hoping to, to solve, so what a customer's calling for and the types of things that they were asking to trigger that attempt. So the utterance is quite

Tom Morgan:

boxed sort of. Yeah.

Alex Baker:

Yeah. That was, that was certainly, it was certainly used to be the case, but now They're doing some, some great work with Lex where you can, I guess slightly before even the whole gen AI hype thing started there was Lex Q and a bot which, which kind of, it put a sort of search capability behind Lex, basically, which expanded a bit away from that real prescriptive, you know, it's got to fit in within these five different boxes, that approach. And that, that's now been improved even more by Gen AI because you can kind of get Lex to, to fulfill your standard intents, but if it doesn't recognize it, you can then send the, the utterance to a, a, a Gen AI, a large language model in the background and get that to, to try and figure out the best way of serving it, whether it's via a knowledge base or something else.

Tom Morgan:

Got it. Yeah, that makes sense. Very cool. All right, we should stop talking because we've been talking for over half an hour. Which is which is great. But I mean this, and there's so much talk about this AI stuff. I feel like we need to do another one. We need to put something in the calendar for six months time just to kind of circle back and see how things have improved, what's worked and what hasn't. And, and there's tons of stuff. Yeah, there's lots and lots of potential in the air stuff and we'd like to hear from you as well. If you're doing cool things with connecting AI and you want to show them off, or you've got questions about them, do get in touch either on LinkedIn or you can email us at podcast at cloud interact. io. I think it's time to bring this episode to an end. Thank you very much, Alex. It's been great chatting with you again. Thanks, Tom. And thank you all for listening. Today we discussed AI in Amazon Connect. Be sure to subscribe in your favorite podcast player and that way you won't miss it. Whilst you're there, we'd love it if you would rate and review us. And as a new podcast, if you have colleagues that you think would benefit from this content, please let them know. To find out more about how Cloud Interact can help you on your contact center journey, visit cloudinteract. io. We're wrapping this call up now and we'll connect with you next time.

Podcasts we love