The Role of AI in Contact Centers and Regulatory Considerations with Philipp Beck

Tom: 0:00

Hi, welcome back to the podcast. This week we have Philip Beck. Philip recently announced he's stepping down as CEO of Luware. He was founder and CEO for 14 years. He's been a really great supporter of everything I've been doing. And Luware have been a really big supporter of empowering cloud. So really thank them for their continued support. In this conversation, we got into the Luware journey, the journey from on prem to cloud and some of the benefits of being in the cloud, SOC 2 compliance and going through that process. power automate and power platform and how they integrate there. And also some of the new features coming to Luware. And of course we got into AI, AI in the contact center, some of the opportunities there, but also some of the risks and the compliance considerations. Really hope you enjoy this conversation. If you've got any comments, questions, or feedback, do let us know. Thanks Luware and on with the show. Hi, everybody. Welcome back to the pod. Excited for this one. Philip is a long term friend and has been a big supporter of empowering cloud and always gets a good to get his perspective. Philip Beck, do you want to just give the say a bit of an intro and background, please?

Philipp: 1:08

Sure. Thank you very much, Tom, for having me here today. A quick one to my person. I am CEO and founder of a company called Luware. We do customer experience or customer service application on top of Microsoft UC stack since about 14 years now. Always been purely on Microsoft, strongly related to Microsoft, try to get the maximum out of what Microsoft is able to give away through the API functionality and make the whole Microsoft ecosystem. kind of work with the use cases our customers are actually asking for, right? There is a big focus on customer service contact center scenarios, but also a focus on compliance recording. So for all the big regulated industries there's always a big thing about, yeah, we need to get that under the regulations and how can we actually make that work? And there we help as well. That's a short summary about me and what we are doing here.

Tom: 2:08

Thanks. Yeah, I like that's where you sit. Obviously, I've worked with you guys pre empowering cloud as well. And it's interesting because it's a cloud first journey, but also your background is highly compliant customers, highly regulated customers. So it's that fun journey of regulation and the cloud at the same

Philipp: 2:23

Yeah, you can say fun, right? I would say

Tom: 2:25

can.

Philipp: 2:26

it is. It is definitely an interesting journey. We started with the on prem business as I said, 14 years ago, doing it all on OCS Lync Skype for Business at the end. And then there come the cloud around and about five years ago, we started to build our solution again from scratch. That was a heavy investment. We started until we have been able to sell it was about 2. 5 or 3 years time until we actually got there. But we really believe that the cloud is the next big thing and we need to make sure that we do not have this kind of old bags with us, which drags along and we cannot really do it. So we try to get the newest technology at that point in time, making sure it's going to scale. Making sure it's adoptive to new needs and we can easily replace some stuff or build new some stuff. And also, if you go and look now about 3 years back with all the Covid people from home, the stuff where data centers come less and less important over time as regulations now allow as well to run things in. public clouds. That was a massive push for us. And we also see that gratefully on the last year's numbers. So that was really an interesting but a hard journey. But yes, when you say we started as non prime business selling licenses and now our really core business, and yeah, more than 70 percent of our revenue comes from recurring revenue that was a really interesting time on that shift to get there where we are today. Yeah.

Tom: 4:02

Talk about last year and what you've done with the cloud platform. So I know you've dropped some new features, some omnichannel capabilities. Where's the best place to start in terms of things added last

Philipp: 4:12

yeah, the good thing is when you are on the cloud and you have the bundle or you have built it the way that it is running with all the principles you have in the cloud, then it's easy to just update feature by feature when you are done developing and also easy to being able to update the platform, which then enables all the customers who're running on that platform using the new functionality. So we have pushed out about 30 to 40 features over the last year. And yeah, you said it right, it was the year of Omnichannel. It was a lot about chat, external task. email is for now just around the corner. And there, yeah, if you have that on this multi or this cloud platform that's really great, right? You can update it and customers can actually use it. On the next day when that feature is available, we wouldn't have been able to do that on the on premise side. To be really fair, with about 7800 customers now on the public cloud infrastructure, if you would have to update all the on premise infrastructure, no chance

Tom: 5:19

It's a non starter, isn't it? And the customers don't want to patch either the particularly with contact center that they're nervous, so the fact you take care of it for them service side is amazing because they inherit that benefit. The other thing is you inherit the benefit of the bigger Microsoft cloud platform, right? So as Microsoft add capabilities. you can leverage those capabilities into your solution.

Philipp: 5:39

Yeah that's definitely true. So when you look on chat on how we build chat, we are using ACS and other technology, right? All the text to speech stuff is, of course, done by Azure's text to speech services. So that's building it as we started five years ago, one of the main goal was building a solution which is embedded within the Microsoft ecosystem as close as possible right wherever we see something which is within Microsoft and already available within M365 or Azure, then we're going to take that. We do not have to rebuild this so I can use the developers here we have to actually build Build the new features around the contact center rather than do baseline, which has been done already by twenty, thirty people so that's really the advantage as well from building into the Microsoft ecosystem and was example. I can tell you that we are just around the corner to get officially pushed out and the power automate certified connector and that's again a statement on how this goes and how tight we actually are integrated within the Microsoft ecosystem and with the power automate. It's fun of what we sometimes see. The customer is actually able to build in a contact center scenario, connecting their CRM's, connecting other databases or information services he has and really cool use cases, and I'm pretty happy that we actually, it was a painful journey, it took us about four or five months to actually get there because of the Microsoft processes. But hey we got there finally, and I'm really happy we have been able to do that. Yeah,

Tom: 7:27

Yeah, we did a webinar on that earlier late last year, actually, and some use cases and examples and showed in real time how it works and it does unlock all sorts of value, but it also unlocks, as you said, the customers do it themselves so they can then extend with power platform into all sorts of scenarios because you're providing that connector, which is really nice.

Philipp: 7:46

That a lot in the low code space, right? Everyone will do something low code. You have seen that in the contact center already. Back in the days, it was and amazing when there was a graphical call flow editor, right? You have to workflow and that was 14 years ago, on leading edge, right?

Tom: 8:04

I remember UCCE is like selling those big back when I was a Cisco side, and it was like 12 weeks of scripting, you'd sell to configure the contact center, and then we got those drag and drop builders. That was

Philipp: 8:16

That was really cool and now you're going to think of you can do all the integrations you have still to code in many of the contact centers into CRM webhooks. It's all open platform, but you need to write still code in most of the cases, not all of them, but in most of the cases. And what we have seen over the last two years and the adoption of the power platform to actually do simple stuff and get parameters back and the routing decision back on data, and that's fairly easy. And as you said, yes, that's that the customer can actually do it himself. And that's the great value you're going to see there. Sometimes a little bit tricky on the support side, but it definitely unlocks a lot of value for the customer. And I'm really happy

Tom: 8:55

It's also a great partner opportunity. I've seen certainly some Europe partners doing this where they will sell Luware. Obviously, they'll make some margin on the Luware but then they make more opportunity out of the extensibility because they'll help the customer with that

Philipp: 9:08

Yeah, completely agree. Especially if a partner focuses on that, right? There is sometimes where you have a partner, yeah, we do the telephony and the UC and some of them have already identified, oh yeah, there is where we can bring value to the customer. And we have very successful partners who are doing exactly that.

Tom: 9:28

Yeah I have this conversation with partners all the time and like particularly new partners into the space that are like telephony is just not as hard as it used to be like, I used to do it right. It was racking and stacking gateways and it was complexity and like you've still got porting, but like it's so much easier, particularly knowledge workers, whereas that contact center space like I think we're up to 24 certified providers by Microsoft, which basically means That's not even a short list anymore so you have to help the customer understand all these differences, all these capabilities. But then with contact center, it's either business impacting revenue, generating service impacting. So you're into reporting, you're into workflows, you're into AI, which I know we're going to get to, there's lots of value and excitement in that space.

Philipp: 10:10

Definitely true. I see. I see an important factor more and more as you say, the regulated industries, of course, as we are working as well with regulated industries, it gets very important for us and our customers that the data stays secure, right? And data secure there are various different certification we have found out for us and for our customers. SOC two type two is actually the best we can do where we can guarantee our customer data confidentiality, making sure we have the security stuff in place. And that was a big shift as well or a big lift last year

Tom: 10:46

I was going to say, you've just glossed over that. That's a massive effort to get that certification. It's non trivial.

Philipp: 10:51

it was really a massive effort. For those who are knowing what ISO 2701 means and SOC and you want to compare it, it's pretty easy by saying ISO. You have to convince the auditor that you're doing it right. SOC you have to prove the auditor that you're doing it right? And it just needs a lot much more proof. And all of the evidence you need to collect to make sure that they actually believe you that this is how it should be, which is good for our customers because it's not something we try to do, it's something we actually do and have proof that we are actually doing it. So I'm really happy we got into that. There was a lot of process changes internally, but of course it looks now or we are as well proud of ourself on those processes and have those implemented as a company with about 130 people, right? It's still I would say midsize and not large, but having being able to push that through beside everything else we had last year and the growth, of course, was a big thing as well. I'm really happy about the team that we have been able to push that.

Tom: 11:53

yeah, that's amazing. And you yourselves are in that range, but your customers obviously are a lot bigger. So things like SOC to that, that, that's some of their baseline requirements,

Philipp: 12:02

Yeah, they use that to actually, as you said there's where we're coming from. When you look on the list of the certified vendors, yeah, look on the list and then look for in the enterprise segment who is having SOC two type two and then the list shrinks to a short list again, which you can actually use. Yeah.

Tom: 12:20

Yeah. If you have a look so plug for our research, which people can get to in the community we have a comparison of all the vendors for recording all the vendors for contact center. And as you say you take a connect versus extend that immediately drops the list massively, but people don't often understand that. Then you're into like, what's the minimum seat count? Is it true cloud? Can you self-service? Things like this start to separate away. The other one that's hard for us to quantify is reference ability of genuinely Teams first customers because there's some really big players on that list by brand that we all know, but I'm like how many teams integrated teams native solutions actually have versus just getting the badge. So yeah, check out the research if you want to start comparing what your options are.

Philipp: 13:03

It's just an easy answer for us. It's 100%.

Tom: 13:06

I have to play the even card for it. But

Philipp: 13:08

Sorry. yeah,

Tom: 13:10

up there, of course.

Philipp: 13:11

No.

Tom: 13:12

So let's talk about because I want to get onto AI, right? And we haven't really prepped this for people listening in because I wanted to have this debate live. So I know you've started on the AI journey, but what have you got so far

Philipp: 13:23

So on the AI journey, AI, especially if you look on the Gartner hype cycle is on the top of the hype when it comes to the contact center, right? So yeah, you can do everything. So now it comes down to what brings you actual value? I go first a little bit back into last year. We have built over the last quarter of last year and the automatic transcription. So we have built in again from, took information, not doing ourself the speech to text. Now we have the speech to text, we have the mood in our system we get that data out of the call. Nicely distributed. What was Tom saying? What was Philip saying? So we can easily have those streams and get a good text representative. As I believe being a Microsoft ecosystem, getting that out of Microsoft is probably the best choice anyway. Again, it falls on the SOC, making sure that not all not different vendors are in the play or we need to write to Google or whatever.

Tom: 14:22

going? Is it staying in the single vendor?

Philipp: 14:24

So we have that transcription now, and funny with the possibility we have on the Power Automate, we can actually today make that in a summary and write it into CRM. And there is where I actually see for now and the next probably one or two quarters, the biggest value. Also had a talk with Forrester and they accidentally agreed on that. But the biggest value is on the efficiency of agent for the next couple of quarters, let's put it this way. You can also use that data then to create FAQ or to use the FAQ to give the agent three possible FAQ articles. That helps, especially at the moment when you there are new starters to get them up to speed really fast and as well to try and try to train the AI on that discussion, which is the right answer that when the learning there is at a certain point, you then can go. and move it in front where you're going to say, okay, look, my opening hours are between eight and five. When you're going to call it at eight PM that then probably I do not have agents around, but you can talk to a virtual agent if you want and try to figure out if this person or the virtual agent can actually sort that problem and then get some first experience around that and if that works well yes, there is a possibility to put it in front of the customer at day one.

Tom: 15:57

Yeah, that's that's what I'm excited about. So we're definitely into kind of 2024 vision now, aren't we? But like it right now, that agent coaching is a really safe first step because you've still got the person with the customer. I don't know. again, the kind of customers you work with, it the information can't be wrong, so right like that. That's a risk with AI. We all know hallucinations are risk right now. But I can definitely see a future, particularly with businesses like simple use cases like you say, open hours. Where's my order? Where I as a caller in or I guess a chatter or a viewer of video potentially can self select. Say I'm happy to engage with an AI agent. I know what I'm getting myself into. I've elected to do that. And I would I'm a bit technology first, obviously, but I would do that every single time because I want the instant answer, probably the instant question, I know. there's some very skilled agent scenarios where you're always going to need a person, but a lot of contact center traffic is just the person like you say, working on FAQ, working off a scripted answer. The AI in theory should be able to do quite a good job of that.

Philipp: 16:59

Yes. So now the theory is right. You're completely right on that. In practical terms, that looks slightly different. So how do you train the AI? It should, right? You said it right. It should, but it doesn't. What I believe is there needs a lot of AI needs a lot of training data, and to get to that training data, you need to figure out how to get or how do you produce that training data you can use knowledge base. you can use internal documents, whatever it is, but it still needs on the interaction side, some stuff, right? And therefore, I believe it is good if an agent, if you use the agent at the beginning to say, okay, yes, that's the right article from the FAQ, that's the right article, and if you hit there a certain amount, then you can actually use that and say, okay, I'm 99 percent sure that this is the right answer and you can let the AI run. But you always need to think AI is only as good as the training data. And we are not talking generative AI here because it needs to have company internal knowledge and getting that data inside the AI. Of course, it can work correctly and it figures out what that comes from the generative part. But getting the point where you use that data and put it in training into a training exercise for the AI and then use that to get it out.

Tom: 18:28

Yeah, you're going to have to build your own LLM with your own data because you don't want it and we've seen I'm sure you've seen it as well. We all seen gimmicky demos where someone slaps the open AI API in front of a conversation and it does something right and it looks impressive at a glance because it's engaging, but that's where your business risk is highest because it's just that general model taking a stab at an answer based on a giant training set. So you're going to have to train it with your FAQ your data and you want a smooth transition. Like you want to train the A. I. If it doesn't know the answer, transition to agent and don't try and make up an answer here, just transition.

Philipp: 19:02

Yeah that's where we see a lot of customer actually asking for, can you build this? Because we already have a chatbot with a bot kind of thing, but, there is always a difficulties to hand it over to an agent and that stuff. So we rather go down a little bit in functionalities, but we have it in one system. We see a lot of that consolidation going on right now. And another thing which I would say is a difference of what we have seen before inside AI. Bye. In the past to the new one is because of this chat GPT, you need less training data, right? You definitely need less because the whole semantics, how a sentence is built, how it's worked nicely and so on. That knowledge is already in general. Now you need to treat the specifics on that, so you can say, the little kid has grown, so you do not need to teach the first five year of a kid growing up, you need to give him a new job and probably train them for two weeks very hard on that new job. That's about the time frame and the scale I see it on that. And I really believe it's much faster. You need much less training data, but you still need to. And one big thing I would like to mention is especially in our industries where things change and are regulatory, need to be regulatory compliant is when a regulatory change at time at the 1st of January, let's say 1st January 2024 regulatory change. How do I train the AI in before that at that day at midnight, it doesn't respond in the old way, but the new way. You cannot learn from training data you had throughout the last two years of experience. You need to have a model somehow on how you can replace that information

Tom: 20:50

Just gonna say, I think the only thing I would trust today is to build a whole second model, have it ready, cut it over because I wouldn't trust, I'm not expert on AI by any stretch, but I wouldn't trust anything other than I've tested this model 17 times 17 ways 17. It always comes up with the new right answer.

Philipp: 21:07

And then you can still have something getting the wrong answer because you only tried 17 times, right?

Tom: 21:13

I wonder if we'll end up with AI in customers that are big enough AI center of excellence is within the contact center business of like their job is not anywhere near being an agent or a contact center manager, their job is to run this understand and help feed the engine, run the model, test the model, because it is a skill in the it's own, right?

Philipp: 21:32

Yeah, this is skill in the same one. And you can discuss if that's inside the company or with a partner. and it it depends on how often you're going to change and how dynamic you would like to have it, then there's always this battle.

Tom: 21:45

And who you want to hold the risk. Some companies quite like having a partner hold the risk.

Philipp: 21:49

That's the other point. Yeah. And for us, it's just important that we integrate with that model, which we believe has the biggest biggest structure in the market. And for now, it's for us, yeah, the Azure OpenAI stuff.

Tom: 22:04

Yeah. It's fortuitous or however you want to look at it. You obviously so aligned to Microsoft. Microsoft have bet on open AI so you can run all that in Azure, but also Microsoft are offering other models in Azure as well, so you've got options there, which is great.

Philipp: 22:19

Definitely. And I think over the next kind of 12 months, we will see, some frustrated contact center operator out in the market because what they hoped will actually not materialize and we will see some very happy and in about 12 months time, it will get to Let's say commodity more and more. It's still on the early adoption phase, and some will probably have a hard time there., but it will pay out at the end, I strongly believe that's actually the way to go.

Tom: 22:52

Yeah, I think it's inevitable on some time. I don't think anybody can bank on the exact timeline. I think the two verticals are immediately interesting for AI is education. I think if you look at that model of certainly traditional classroom education of one person to 25 children, generic goes at the slowest pace, that's really interesting where AI can add value of one to one experiences and then contact center where it's very people heavy and it's very data heavy. And let's be honest that the average user experience as an end user often not fantastic. There's lots of room for improvement there, so like those two industries are going to be disrupted. And then when I look at context and more specifically, what are the big players going to do that are on the traditional big cost per agent model? Are they ready to disrupt themselves? And I'm not sure they are. So it'll be interesting.

Philipp: 23:42

Completely agree. And I believe in the 1st phase, we're going to see a lot of the agent efficiency as I said, right agent turnover or staff turnover is a huge problem. Typically in a contact center, you would like to get them as good and far skilled up as possible, and you would like to get them as efficient as possible. And especially if you talk summary, everyone writes it slightly different. Some forget about it, then if you get that automated, that just helps the business so much. And then, yeah, I see the first customer that was an interesting discussion very early this year was about, hey, if I can get that automated by by an AI, and I would like to use my people then to Or I would like to open AI trigger when there is an upsell potential and having my people actually then got pulled in and try to do an upsell because AI doesn't make upsell but really on the, yeah,

Tom: 24:50

That's a really good use case. You still need that humanity of feeling out what the customer's not saying and what questions I should ask. You're right. AI is nowhere near that yet. That's really interesting

Philipp: 24:58

But but this we have seen a couple of companies moving from a cost center to a revenue center and in the customer service, right in the contact center, and therefore, I see it's interesting to see where that goes. If that was just a one off or if that see to get a trend in the industry that they say, Hey, yeah, okay, I do not need Probably that many people anymore, but I would like to have the people to get triggered by the AI. Hey, look, there is a potential up sale potential, we would like to have it this way, that was the discussion before a quick summary, now please join the call. That would be an interesting one.

Tom: 25:32

It's also unblocking. There's a big there's a mass amount of data in the contact center. And we've been recording for years. We've probably been transcribing, but nobody really, not many that I see really efficiently mine that data, particularly when you get to smaller contact centres, and it's like the very big ones probably have a data team, but the AI unlocking, like you say, if you could take all the upsell calls over the last six months and have AI say, what are the commonalities of these successful upsells? The value you can unlock from that data is amazing.

Philipp: 26:03

And you can take that back, right? When we started 14 years ago, it was about the work for workflow editor. We just talked about at the very beginning, which was transformative, right? And now we have moved into the cloud, which was transformative and also gives. The contact center, when I started contact center were typically deployed in customers with about a thousand people, right? Before that, no one was actually able to afford a contact center. Now with cloud, it moves into, okay, it gets affordable for everyone because you only pay what you actually need, there is no huge project at the beginning. Great. And now we're going to talk about the other piece. Oh, yeah, only the big ones have this kind of data and their eye department who are looking into that and data mining it, and I really believe what you said is over the next one to two years, this will come down and this technology will be available on affordable costs as well for midsize and smaller customers.

Tom: 27:03

Yeah, and sometimes the midside customers are quicker to make the jump because they have the agility, right? So we saw this with cloud like the midsize went first because they saw all the upside and they didn't have the draggy heels of, massive environment. So yeah, it'll be interesting to watch. Cool. Philip, thanks for taking the time. We've definitely got a little bit longer than we intended, but I wanted to get deep on that, so I appreciate you taking the extra time. If people want to find out more best thing hit the website, I

Philipp: 27:29

assume Hit the website. Join us on LinkedIn. Follow us on LinkedIn. Yeah, there's where all the news gets ripped or sign up for the newsletter on the website as well. There's where we get all the news out. That's for sure.

Tom: 27:41

Awesome. Cool. Thanks for it. And again, thanks for the support of everything. Empowering cloud. Appreciate you guys being on early and supporting us and everybody out there, you've got any comments, questions or feedback, do let us know. Thanks a lot.

Philipp: 27:52

Thanks Tom to having me here. Bye bye.