Why Conversational Voice Design is key - no matter the voicebot platform.

Today with Steve and Tom, we discuss conversational voice design. The key requirement for a successful voicebot. No matter what the platform from Genesys to Avaya, wherever there is the facility to build voicebots, the conversational design is so important. We talk about why voicebots are not being used as effectively as they should be, even now, and how to get some starting point and expertise to get you off the ground.

A voicebot should sound like a human agent, so how is that achieved?

Our latest series of podcasts, concentrates on voice and how that is going to impact the next few years with tips along the way. Find out more about voicebots here and if you have any subjects that you would like us to discuss then email info@disruptionworks.co.uk with the subject Podcast and we will see what we can do ;-)

Transcript

0:0:5.950 --> 0:0:14.40
 Steve Tomkinson
 Hello everybody and welcome to another exciting edition of disruption works chitchat. I've got Tom Houwing with me today.

0:0:14.940 --> 0:0:19.700
 Steve Tomkinson
 Our voice consultant. Good morning, Tom. How are you? You're alright.

0:0:19.630 --> 0:0:21.280
 tom (gast)
 Good morning. I'm fine. How are you?

0:0:21.190 --> 0:0:28.290
 Steve Tomkinson
 Yeah, not too bad. Not too bad. I'm glad it's sunny outside for a change and it is over here anyway. I mean, what's it like in The Hague?

0:0:28.670 --> 0:0:31.940
 tom (gast)
 It's a little bit foggy, but the sun is trying to get through the clouds but.

0:0:32.380 --> 0:0:32.680
 tom (gast)
 Yeah.

0:0:31.430 --> 0:0:35.930
 Steve Tomkinson
 Ohk. OK, well, you might have some sun coming over from our direction, so that's good news.

0:0:35.260 --> 0:0:39.50
 tom (gast)
 I mean it's it's almost more English whether here than with you in English.

0:0:41.140 --> 0:0:50.440
 Steve Tomkinson
 Yeah, thanks for that. Yeah. I'm always very jealous of you to go. Yeah. I've just been down to the beach and, you know, I took the dog for a walk. And what time?

0:0:49.440 --> 0:0:50.730
 tom (gast)
 Yeah. Yeah, but.

0:0:50.810 --> 0:0:54.300
 tom (gast)
 So with the weather could be better actually, but OK.

0:1:9.140 --> 0:1:9.570
 tom (gast)
 And we.

0:0:53.980 --> 0:1:10.190
 Steve Tomkinson
 Ah, OK, alright. Yeah, yeah, yeah. Well, it because we're in Plymouth and I'm at the nice thing about it. Plymouth is that we are by the same. But you have a beach and a beach. The beaches aren't walking distance from Plymouth which is a real shame, you know. That's that's one thing we're missing.

0:1:10.760 --> 0:1:15.370
 tom (gast)
 Yeah. When I when I woke up my door, I like one minute I am on the beach actually.

0:1:15.580 --> 0:1:18.160
 Steve Tomkinson
 Uh, I hate you. I hate you already.

0:1:21.320 --> 0:1:39.710
 Steve Tomkinson
 Well, like thanks. Thanks for doing this. Uh, join us on the podcast day. I I think we're going to try and do this as as an ongoing thing. So anybody is listening we we'll we'll be doing a series of these podcasts with Tom. But what we wanted you to try and talk about today was.

0:1:39.870 --> 0:2:6.780
 Steve Tomkinson
 Uh. The fact that there's a lot of big systems out there. Uh like, UM, uh, genesis, nice aviral the rest of them that have voicebots built in, but there's one, not a great take up of that technology internally on those big platforms and and two, it's like well, why, why, why why is that and what are the challenges and?

0:2:7.500 --> 0:2:11.290
 Steve Tomkinson
 At what's the approach that should be taken with these things? Uh, cause.

0:2:12.220 --> 0:2:32.200
 Steve Tomkinson
 There's a there's a huge gap in the fact that we need voicebots in the market. The technology has moved on, or else generally we have our own platform that uses all the cloud services and the power of that, the cloud services that are around, you know, like Google for.

0:2:32.900 --> 0:2:41.100
 Steve Tomkinson
 Uh and Lu and stuff like that. So it's it. How can you engage with your big platform and that's kind of what I'm trying to talk about?

0:2:48.960 --> 0:2:49.190
 Steve Tomkinson
 Yeah.

0:2:57.590 --> 0:2:57.960
 Steve Tomkinson
 Yeah.

0:3:5.410 --> 0:3:5.740
 Steve Tomkinson
 Yeah.

0:2:42.220 --> 0:3:5.810
 tom (gast)
 Yeah, you're right. I mean, first of all, we need to respect one very important issue here. I mean, you're talking about those big systems and Europe actually, right? But those big systems were actually created as telephony platforms and they were created already for some time ago on the go. And of course, there is an ongoing development on those platforms and.

0:3:19.20 --> 0:3:19.410
 Steve Tomkinson
 Yeah.

0:3:27.290 --> 0:3:27.590
 Steve Tomkinson
 No.

0:3:6.150 --> 0:3:36.560
 tom (gast)
 Uh, the industry recognizes the added value of voice, so they would be working on an integration trying to build in something that could be a voice bot functionality. But in the basic those platforms were made not especially because they were not voiced, focused and now you see now you see the problem here a little bit. I mean there are players out there who have been creating software.

0:3:36.810 --> 0:3:45.630
 tom (gast)
 And that is exclusively dedicated to voice and everything that has to do with that and the other ones are trying to integrate it. And that's the difference.

0:4:10.920 --> 0:4:11.300
 tom (gast)
 Yes.

0:3:46.130 --> 0:4:17.60
 Steve Tomkinson
 Yeah, and having the interesting bit is, is that they know they need it because they've been including it. So it's a, you know, Genesis aren't gonna go and buy a voice, but provider or try and integrate it into their stuff. If they don't think it's a thing, you know, they know that it's a requirement and this is only gonna get more important as they go. But like you said, they're trying to shoehorn this in trying to force it into their system.

0:4:17.140 --> 0:4:26.580
 Steve Tomkinson
 Wasn't like you said originally. Inspired by that and and it wasn't thought through to have that stuff in there. So it is a challenge.

0:4:27.20 --> 0:4:30.740
 tom (gast)
 Yes. Then you're totally right. And I mean, let's also face it, I mean.

0:4:53.670 --> 0:4:53.940
 Steve Tomkinson
 Yeah.

0:4:30.830 --> 0:5:2.310
 tom (gast)
 And a working with speech technology and speech recognition and later on in the context of a voice bot, you really need specialized technicians that are that know the the way of how to program software that really works well with the cloud, for instance, how we work with speech to text and how we analyze what we get back from the cloud. So this is a skill set that is specialized on voice.

0:5:8.490 --> 0:5:8.820
 Steve Tomkinson
 No.

0:5:2.390 --> 0:5:9.940
 tom (gast)
 Which those great big providers like Genesis, they do not have voice specialists in that that context.

0:5:32.700 --> 0:5:33.110
 tom (gast)
 Correct.

0:5:10.150 --> 0:5:39.160
 Steve Tomkinson
 No, and. And that's and that's kind of part of this uh podcast is the fact that this is the challenge. So you know the challenge we see at the moment is that there's two sides, one that the platforms aren't really providing support about the technique of providing a conversational flow or any design within their own voice bot platforms, but also the the client, the end user of the platform.

0:5:39.560 --> 0:5:58.640
 Steve Tomkinson
 They also, you know, inside their businesses, they're not gonna have the expertise as well. So what do you do? How do you get started? And I think that's the that's then the challenge they've go why? We've spent a fortune on this big system we spent. We spent a lot of money with keep coming back to this. They're gonna hate us for it, but.

0:6:20.440 --> 0:6:20.690
 tom (gast)
 Yeah.

0:5:58.750 --> 0:6:25.310
 Steve Tomkinson
 You know they have this system. Well, it should we switch to a via. So a via, right? There's been a fortune on it and the game, but how do we use the voice bot elements of this? You know, how do we do that? Because we know we need it. We've got a contact center that's overrun. We can't. The queues are getting longer and longer. How do we remove those queues? And of course Voicebots is the solution for that. Everybody understands that but.

0:6:26.110 --> 0:6:27.30
 Steve Tomkinson
 How do they start?

0:6:32.700 --> 0:6:33.420
 Steve Tomkinson
 Yeah.

0:6:38.470 --> 0:6:38.990
 Steve Tomkinson
 Yeah, yeah.

0:6:46.690 --> 0:6:47.60
 Steve Tomkinson
 Yeah.

0:6:55.580 --> 0:6:55.850
 Steve Tomkinson
 No.

0:6:26.180 --> 0:6:55.980
 tom (gast)
 Yeah. You're. So you're so right. You're so right. I mean, the other day, I was talking to one of my my business customers who literally told me I I, I, I, I showed them the solutions but possible with voice bot. And they really show you. They literally told me you have. But we invested so much in it on the Genesis platform. So we don't. We are not going to invest anything more. And now they're stuck because they do not. They are not in a position to quickly create something that is working with speech technology because.

0:7:9.600 --> 0:7:9.950
 Steve Tomkinson
 No.

0:6:56.310 --> 0:7:15.570
 tom (gast)
 Those platforms do not provide a toolbox where you can easily build a voice bot, and secondly, an even more important, they don't have the skill sets that is required to create a conversational style. I mean, the conversational style in fact has nothing to do with the technique. It has to do with the way of thinking.

0:7:15.880 --> 0:7:20.10
 Steve Tomkinson
 Yes, that's right. And that I suppose that's the that's the point now is that.

0:7:21.520 --> 0:7:28.490
 Steve Tomkinson
 We there's been a this has been around for a long time as as you know a time because you've been doing this for decades.

0:7:29.510 --> 0:7:32.100
 Steve Tomkinson
 I'd hate to say sorry. Sorry you have.

0:7:32.360 --> 0:7:32.630
 tom (gast)
 Good.

0:7:32.220 --> 0:8:4.70
 Steve Tomkinson
 I'm, but you know. So it's been around for a long time. But of course the technology now is caught up at to make this it a proven system. Now it works. It's not a, it's not something that's experimental anymore. It's actually it works and it's and it's a thing, but there's still isn't enough of it used in the marketplace to help customers, but also to help the teams that are help also helping customers and something has to break. And I think the.

0:8:4.160 --> 0:8:13.300
 Steve Tomkinson
 Problem we now have is that there's a whole space where people who've got these big systems need to consider we have to invest.

0:8:13.990 --> 0:8:19.810
 Steve Tomkinson
 Time and effort into the design process of this we have to we have to start looking at.

0:8:20.940 --> 0:8:50.770
 Steve Tomkinson
 How we design AA voice bot rather than just even just using the tools if they've got even the skills to do that to go OK, we've made a voice bot, but it was really bad and you just go. Well, of course it was because you didn't think it through you. You never had that experience. You don't have the conversational flow design. You don't have. We make bots that sound like an agent, but that's a skill, you know. And I think that's the point. The bits that are missing in this equation.

0:8:51.670 --> 0:8:59.910
 tom (gast)
 Yes, I think you're right. We should not forget some, some, some things here. We're coming from a.

0:9:5.160 --> 0:9:6.230
 Steve Tomkinson
 Yeah, yeah.

0:9:17.590 --> 0:9:17.870
 Steve Tomkinson
 Yeah.

0:9:28.430 --> 0:9:28.770
 Steve Tomkinson
 Yeah.

0:9:0.990 --> 0:9:31.780
 tom (gast)
 From a former period of voice recognition development where it, let's say, let's go back ten years, then we we there was no cloud solution and I mean let's face it, Google gets millions billions of inputs every day. They are in the position to tune immense correctly compared to proprietary systems where people handle their own language models and try to make it work as possible. So in the old days.

0:9:34.710 --> 0:9:34.990
 Steve Tomkinson
 Yeah.

0:9:39.510 --> 0:9:40.230
 Steve Tomkinson
 Yeah.

0:9:32.70 --> 0:9:47.200
 tom (gast)
 The accuracy was also not that high, so we were not always sure if we really got what the caller said. And we also work with what we call one word recognition so that the system was tuned on only single words.

0:9:47.490 --> 0:9:49.610
 tom (gast)
 And some.

0:9:56.910 --> 0:9:57.460
 Steve Tomkinson
 Yeah.

0:10:4.420 --> 0:10:5.870
 Steve Tomkinson
 You might as well, yeah.

0:10:26.510 --> 0:10:26.800
 Steve Tomkinson
 No.

0:9:57.520 --> 0:10:28.90
 tom (gast)
 Are today say today now if you rent a car tomorrow, press one goes faster. I mean this. Yeah. This type of one word recognition kind of design is not providing any added value and furthermore it does not connect with what we call language instinct. I mean people use their language in communication but we create with these old school designs we create something that has absolutely nothing to do with the dialogue it has to do with trying to get.

0:10:30.150 --> 0:10:30.660
 Steve Tomkinson
 Yeah.

0:10:28.400 --> 0:10:48.140
 tom (gast)
 Somebody to match your word. And now today we are in a totally different situation. I mean everything is GDPR compliant. Today Google is compliant, everything is safe and we have a unbelievable accuracy in speech recognition. So when we send something spoken to the cloud speech to text.

0:10:50.180 --> 0:11:0.950
 tom (gast)
 Transcription is working so well today that we are really in the situation to work in a total different dimension compared to what we had to do for that.

0:11:10.680 --> 0:11:11.430
 tom (gast)
 Yeah.

0:11:0.540 --> 0:11:28.830
 Steve Tomkinson
 And then that that gives us then. So the natural language understanding is there, uh, and then you've gotta do the processing of that language. So you get your text stream and it's then to say, well, OK, we now know exactly what they're saying. We can design the flow around what they're saying and respond accordingly in a very natural fashion. So the flow of conversation is, like me and you talking, and I'm going, I need to rent a car tomorrow or I need to hire A car tomorrow.

0:11:37.900 --> 0:11:38.260
 tom (gast)
 No.

0:11:29.390 --> 0:11:59.980
 Steve Tomkinson
 Tomorrow I could do with a car and you know all those different ways of of, of asking the same question doesn't really matter. We're we're still matching it up and we go, that's fine. No problem. So OK, you need something for tomorrow. We've got these available. What do you want? And that negotiation is so natural that look, we never fool and we've always say this. We never fool anybody that this is not an automated service. It's not a voice bot that's calling them or.

0:12:0.60 --> 0:12:3.50
 Steve Tomkinson
 Or is is taking the call but they forget.

0:12:4.390 --> 0:12:14.930
 Steve Tomkinson
 So that's the point. If we can make them forget that it's a, it's a voice bot and they just get it done. Whatever they're trying to do, rent a car or what, or change their account details or whatever it is.

0:12:15.790 --> 0:12:27.290
 Steve Tomkinson
 That's where we've won, but that takes the skills of a conversational design, not somebody that's just going. Ohh, I'd got this Avaya platform and I've got I'm doing some stuff.

0:12:39.450 --> 0:12:39.810
 Steve Tomkinson
 No.

0:12:42.30 --> 0:12:42.630
 Steve Tomkinson
 Yeah.

0:12:50.140 --> 0:12:50.380
 Steve Tomkinson
 Yeah.

0:12:27.860 --> 0:12:56.730
 tom (gast)
 Yeah, you're you're totally right. And OK, my in my experience, the emphasis was always the technique and the software design, never the design of the flow. That is always that came always 2nd place. Now we finally start to understand that if you wanna make a voice that's very successful, the voice bot has to be very conversational. And I mean to make a flow design very conversational, you need to have.

0:13:17.380 --> 0:13:18.40
 Steve Tomkinson
 Yeah.

0:12:57.570 --> 0:13:27.610
 tom (gast)
 Expertise in how language works, how how you, how you react on somebody's verbalized message and how you can direct that message into an answer that would be maybe guiding the caller in a place where you want to have him or her and where he, he or she says what you want. Finally wanna hear. But to get there you need to be specialized in a human dialogue.

0:13:41.50 --> 0:13:41.390
 Steve Tomkinson
 No.

0:13:27.700 --> 0:13:46.510
 tom (gast)
 Structures. And I mean, let's face it, when you come to a big company where they want to automate, they do they do not directly have linguistical experts who help you design A flow that's always covered by IT people. And it's always technically dominated and that's the wrong thing.

0:13:46.810 --> 0:14:17.580
 Steve Tomkinson
 Yeah. And I and that kind of brings us back to these plate, the platform things. So they go all we've got all the tools. So we've got this thing they say it's a voice bot platform and then then they look at it and maybe a tech can look at it and figure it out and program it up and all that sort of stuff and make something. But like you said, there's no design in that process. There's no flow design. They will be on one words or they'll be just, it'll be a very robotic story and that's just not how you'd like. I said you make it.

0:14:17.650 --> 0:14:25.390
 Steve Tomkinson
 That a successful but in addition to that you've got different language flows. So if you're in a talking in a different language.

0:14:39.920 --> 0:14:40.340
 tom (gast)
 Yes.

0:14:25.470 --> 0:14:46.560
 Steve Tomkinson
 Do you know what we're and talking to English in the moment? But you know, if you say, oh, well, in a different language, the tone and the intonation and the phrasing is actually quite different. So to then understand that language. So if you're doing something in French, saying something in English totally different, you know, there's a different flow.

0:14:53.240 --> 0:14:53.610
 Steve Tomkinson
 Yeah.

0:15:0.50 --> 0:15:0.440
 Steve Tomkinson
 Hmm.

0:15:7.80 --> 0:15:7.350
 Steve Tomkinson
 Yeah.

0:15:10.450 --> 0:15:10.740
 Steve Tomkinson
 Yeah.

0:14:45.910 --> 0:15:11.970
 tom (gast)
 Yeah, well, you're right. I mean, it is both cultural and language structure. So if you, if you would say for instance in English yesterday I bought a car and in German we have a guest on in out of the Gulf. So the car in English comes lost and in German we have the verbs around the car. And so that is easy.

0:15:23.390 --> 0:15:23.770
 Steve Tomkinson
 Yeah.

0:15:32.170 --> 0:15:32.400
 Steve Tomkinson
 Yeah.

0:15:33.470 --> 0:15:33.750
 Steve Tomkinson
 Now.

0:15:12.60 --> 0:15:41.790
 tom (gast)
 But if you are going to talk about variables and you're going to talk about amounts of money or dates, it's a very important in in what place the date in the sentence comes to make the concatenation right? So for instance, when you make a design in English, you can just press a button and let Google do the translation because the translation will be incorrect and will not work. So that's why. And that's one thing. So the structure of the language is one thing, but also the culture of people, I mean.

0:15:42.170 --> 0:15:48.870
 tom (gast)
 And in in the northwestern European countries, we have a tendency to be quite directly in our communication.

0:15:49.300 --> 0:15:49.580
 Steve Tomkinson
 Yeah.

0:15:54.570 --> 0:15:54.860
 Steve Tomkinson
 Yeah.

0:16:3.570 --> 0:16:4.690
 Steve Tomkinson
 Yeah, right. Yeah, yeah.

0:15:49.440 --> 0:16:12.400
 tom (gast)
 And we would say that I do not agree or something or I don't like the solution you go when you go to the Latin countries in Europe, they would never go that direct, they say, well, I like the solution, but maybe I have another idea about it. And so this is also something we need to respect. So it's not only it's not only the language structure but also the culture of the country.

0:16:13.0 --> 0:16:13.250
 Steve Tomkinson
 Yep.

0:16:27.50 --> 0:16:27.400
 Steve Tomkinson
 Yeah.

0:16:34.310 --> 0:16:35.120
 Steve Tomkinson
 Yeah, yeah.

0:16:38.70 --> 0:16:38.520
 Steve Tomkinson
 Yeah.

0:16:13.400 --> 0:16:43.690
 tom (gast)
 If you're going to be direct in a in an application that does collections and you wanna escalate the message to the people who are not going to pay and you are only the slightest rude, it will not work. You have to be very polite all the time and you can be a little bit more direct, maybe in English or in Dutch. I'm sure about that, but that's the difference. Cultural difference and we the voice needs to be designed by a designer who understands that.

0:16:44.190 --> 0:16:45.10
 Steve Tomkinson
 Yeah, yeah.

0:16:44.350 --> 0:16:46.240
 tom (gast)
 It doesn't do it doesn't do it itself.

0:16:47.0 --> 0:17:18.390
 Steve Tomkinson
 No, and and and that is, you know it's it's it just highlights the importance of the whole design structure because you know most of us, uh, you know, we trade in English. Fortunately English is the business, the business language for us. But there's so many cultural differences that it doesn't matter if you're even talking English to an audience. The tone and the directness and and all that kind of politeness, even across the Atlantic. So if you go to the Americans, the Americans will deal with the English.

0:17:18.630 --> 0:17:23.940
 Steve Tomkinson
 It's, you know, it's their primary language, if you like. And and it's the same difference, you know.

0:17:26.860 --> 0:17:27.130
 Steve Tomkinson
 Yeah.

0:17:31.160 --> 0:17:31.800
 Steve Tomkinson
 Yeah, yeah.

0:17:24.950 --> 0:17:35.400
 tom (gast)
 Think about humor, for instance, making jokes. How how, how, how cultural, determined that is. I mean, a German joke is really not that funny.

0:17:37.90 --> 0:17:42.420
 Steve Tomkinson
 Alright, we just lost them. Well, I'm not sure we had a German audience, but we've lost them now anyway.

0:17:47.860 --> 0:17:48.110
 Steve Tomkinson
 Yeah.

0:17:53.870 --> 0:17:54.170
 Steve Tomkinson
 Yeah.

0:17:40.920 --> 0:17:56.90
 tom (gast)
 No, I'm just joking. I'm just joking. Yeah. No. But you know, you, you get my drift, right? It's it's it's it's not only language structure. It's also the way people do interact with their language. And that's very much a cultural thing.

0:17:56.470 --> 0:18:29.460
 Steve Tomkinson
 Yeah, at night? Yeah, absolutely. Right. I'll look, I I suppose the the whole irony of this and the the reason why we why, you know, I wanted to get you on the podcast and actually start talking about this is that the requirements is there we've got investment in the voice box being part of big platforms. That's huge. That's a lot of money that people are paying for that genesis of it and all the rest of them they are spending big monies in here but I don't feel that the clients taking them up nearly as much as they are because they just don't know where to start and it's a blank sheet of paper.

0:18:29.640 --> 0:18:47.580
 Steve Tomkinson
 And they just gave no real support from Unspecialized platform people who are more interested, like you said in the technical stuff. And I think, you know, just wrapping this up now it's it's about conversational design. You wanna make something right where you know we can provide.

0:18:48.580 --> 0:19:18.70
 Steve Tomkinson
 If you get the right people, then like ourselves, of course you know you can get, you know, the design, but also Technical Support to integrate into things as well. But it means you know that is a required thing. Now you've got to get somebody in to start this process off for you. Maybe it's we upskill people in the business, whatever it may be. But the point is it's very unlikely that that's the scenario. But the demand for voicebots.

0:19:18.250 --> 0:19:23.700
 Steve Tomkinson
 And taking the cure away for customers, clients and all the rest of this massive now.

0:19:33.390 --> 0:19:33.900
 Steve Tomkinson
 Yeah.

0:19:36.450 --> 0:19:36.740
 Steve Tomkinson
 Yeah.

0:19:39.930 --> 0:19:40.250
 Steve Tomkinson
 Yeah.

0:19:47.450 --> 0:19:47.920
 Steve Tomkinson
 Yeah.

0:19:52.790 --> 0:19:53.50
 Steve Tomkinson
 Yeah.

0:19:24.150 --> 0:19:53.170
 tom (gast)
 Yeah. I mean in, in my experience, the last 20 years, I have met very few people who are technicians with a feel for language and designers with a feel for the technique, that work together on creating speech technology applications. So this is rare. And but I mean to me it's the only requirement to be become successful because the technique needs to be fully voiced, focused and.

0:19:57.50 --> 0:19:57.280
 Steve Tomkinson
 Yeah.

0:19:53.270 --> 0:20:0.140
 tom (gast)
 The design needs to be fully languaged connected and if you bring those together, I think we have a formula.

0:20:0.640 --> 0:20:8.500
 Steve Tomkinson
 Yeah, OK, well, let's. I hope that was interesting for everybody. And thanks very much for your time. Again, Tom was superb.

0:20:9.10 --> 0:20:24.390
 Steve Tomkinson
 Umm, uh. If you want us to talk about subject, then please e-mail in info@disruptionworks.co.uk and we're more than happy to pick the subject up and have a chat about it. But until next time, thank you very much and thanks again, Tom. Appreciate it.

0:20:24.290 --> 0:20:25.540
 tom (gast)
 You're welcome. Bye bye.

0:20:25.710 --> 0:20:27.230
 Steve Tomkinson
 Alright, cheers now thanks.

0:20:27.510 --> 0:20:27.670
 tom (gast)
 Yep.