Disruption Works Chit Chat

Voicebots and Voicemail Detection - A sticky problem

Disruption Works Season 2 Episode 2

Today we are talking about voicebots and the fact that they are smart enough to detect voicemail, a crucial part of any outbound campaign, with increasing accuracy due to the AI incorporated in the logical understanding of this part of an overall dialogue.

It is a tricky one when you consider the amount of different shapes and sizes that voicemail comes in with.

Listen in as we talk about how the Ai voicebots are managing the human fallibility in a badly crafted voice message.

if you want to find out more about Voice bots click here

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 ;-)

0:0:3.900 --> 0:0:11.350
 David Antwis
 Welcome to another exciting episode of Disruption Works chitchat with Steve and me. David. Hello, Steve.

0:0:13.470 --> 0:0:15.800
 David Antwis
 Umm alright thanks. Yeah, how are you?

0:0:17.30 --> 0:0:17.820
 David Antwis
 Yeah.

0:0:11.660 --> 0:0:19.300
 Steve Tomkinson
 Hi, how you doing? You're right. It's. Yeah. We've let you back up again. Let's say you're. You're back on the back on the top floor.

0:0:19.780 --> 0:0:24.970
 David Antwis
 I look forward to these so much now. The windows, the panoramic views.

0:0:26.910 --> 0:0:31.290
 Steve Tomkinson
 The the fact that you have to weigh special glasses because the sunlight, after coming out of the basement.

0:0:32.280 --> 0:0:36.230
 David Antwis
 Well, the company paid for them. It's just a bonus, isn't it? Yeah.

0:0:37.330 --> 0:0:39.200
 Steve Tomkinson
 Yeah, bless you. Can't sucks well.

0:0:44.570 --> 0:0:45.530
 David Antwis
 Yeah, yeah.

0:0:39.840 --> 0:0:47.650
 Steve Tomkinson
 Thank you for coming up from the basement as as far, far away as there's been. So I'm glad. I'm glad to, to have your company today.

0:0:48.100 --> 0:1:0.390
 David Antwis
 Uh, Mr earlier one. Right, well, what we're here for today, we're looking at one of the challenges of outbound voice box communications. I'm gonna talk a little bit about voicemail.

0:1:0.920 --> 0:1:1.220
 Steve Tomkinson
 Yeah.

0:1:1.330 --> 0:1:5.770
 David Antwis
 And the opening of conversations by the voice bot.

0:1:6.390 --> 0:1:10.370
 Steve Tomkinson
 Yeah. Well, it's it says a tricky old one.

0:1:10.550 --> 0:1:10.990
 Steve Tomkinson
 Uh.

0:1:10.930 --> 0:1:11.350
 David Antwis
 Never.

0:1:27.650 --> 0:1:28.70
 David Antwis
 Yeah.

0:1:12.590 --> 0:1:42.260
 Steve Tomkinson
 The thing is, with a lot of outbound kind of dialers is that they're not particularly smart at detecting voicemail and and it's because it's quite a challenge, really. There's so many varieties, you know, voicemail. There's a tricky old thing. You know, if you leave messages around, you'll realize there's 101 different variations of somebody's antiphone voicemail message or whatever it might be. You know, there's huge amounts.

0:1:42.920 --> 0:1:43.370
 David Antwis
 Right.

0:1:42.730 --> 0:1:47.890
 Steve Tomkinson
 And uh, and that's the that's the problem that you have when you're trying to.

0:1:49.210 --> 0:1:56.830
 Steve Tomkinson
 Get something to smartly identify whether a a voicemail is voicemail and somebody's answering the phone you know.

0:2:2.240 --> 0:2:2.530
 Steve Tomkinson
 Yeah.

0:1:56.770 --> 0:2:9.260
 David Antwis
 Only the solutions gonna be you. You're gonna say that you're gonna look for some patterns in. Yeah. So I'm you've analyzed a lot of phone calls, I imagine and and you know all about it, don't you?

0:2:7.970 --> 0:2:20.710
 Steve Tomkinson
 Well, absolutely, yeah. Yeah, let's say and. And The thing is the bot does now as well, you know, so our stuff's built into platform. So you, you know the whole modules just jammed on the front of.

0:2:21.650 --> 0:2:38.680
 Steve Tomkinson
 A sequence. But what happens is we're looking for a particular language in those in that kind of voicemail. If you think about, you know, you very, very rarely answer the phone and go.

0:2:43.280 --> 0:2:43.750
 David Antwis
 No.

0:2:53.980 --> 0:2:55.520
 David Antwis
 This is clear, isn't it? Yeah.

0:2:56.230 --> 0:2:56.420
 David Antwis
 Yeah.

0:2:39.320 --> 0:3:10.190
 Steve Tomkinson
 Hello I'm not in at the moment and all that stuff saying you know or you that's not why you answer the phone. You know there is a certain language and and also you ramble on a lot to start with the voicemail. So all these are patterns that you start getting an idea of why. Yeah of course it is. And the key that we now have with the power of NLU or natural language understanding is that the voice bots are now much smarter. So the IO involved in understanding what somebody is saying.

0:3:30.130 --> 0:3:30.570
 David Antwis
 Yeah.

0:3:10.280 --> 0:3:41.40
 Steve Tomkinson
 Is really strong, so that means then we also know what that language is and voicemail language is particular. So if somebody is saying I'm not in, I'm not available, leave a message, can't take your call right now. All those different phrases that you have, then those are all managed and the managed quite quickly so that we instantly know that that's a voicemail. So we can let that voicemail go through to its fruition.

0:3:44.620 --> 0:3:45.50
 David Antwis
 OK.

0:3:41.460 --> 0:3:47.470
 Steve Tomkinson
 And then start leaving the message when we're ready and go down that particular thing. So.

0:3:48.570 --> 0:4:2.780
 Steve Tomkinson
 Yeah, even to the point where you start having to manage gaps in the messages, I mean, my voicemail surrenders, and now I was told off right and soundly by the the platform team that testing against mine was horrendous because.

0:4:3.720 --> 0:4:24.220
 Steve Tomkinson
 Because I've got like a four second gap between my record, my personal recording of you've got through to Steve Tomkinson. Please leave a message and then it goes. Burr for agents and then the telco kicks in. So then you have a message for me going if you want to rerecord your message, you can't do at any time. Plus on, you know, and all that sort of stuff.

0:4:22.940 --> 0:4:25.330
 David Antwis
 Ohh yeah, I know the ones, yeah.

0:4:25.10 --> 0:4:38.250
 Steve Tomkinson
 Yeah, and and of course that the big gap means that's really hard to deal with. So because you're expecting it to be then and and even people and you know, like I've been told, even people then start talking.

0:4:39.410 --> 0:4:40.10
 David Antwis
 That's true.

0:4:39.290 --> 0:5:8.820
 Steve Tomkinson
 Because they're not expecting such a big gap, you know, so it's badly formed. So I've used mine as the exceptional rule. So that's that's our managing that. But yeah, it is tough. You've got all sorts of different telco messages somewhere, all right. And some are straightforward in their dad's easy to detect. And there's no problem. But a lot of tricky. And that's why we've been working really hard on the voicemail detection side.

0:5:9.280 --> 0:5:14.380
 Steve Tomkinson
 To make that so much more effective than our competitors and that you know that works really well.

0:5:14.970 --> 0:5:16.600
 David Antwis
 Where the goal, yes.

0:5:16.660 --> 0:5:22.340
 Steve Tomkinson
 Yeah, absolutely. I mean The thing is you can tolerate all these gaps and all the limit they they kind of.

0:5:23.640 --> 0:5:35.250
 Steve Tomkinson
 At the the language understand that get that right and that just makes the whole thing much more effective and a better customer experience on the other side as well.

0:5:35.930 --> 0:5:36.400
 Steve Tomkinson
 You know.

0:5:36.500 --> 0:5:38.80
 Steve Tomkinson
 I sorry.

0:5:44.140 --> 0:5:44.480
 Steve Tomkinson
 Yeah.

0:5:36.680 --> 0:5:46.200
 David Antwis
 Yeah. You you wanted to, you wanted to seem smart and just use conversational smarts, don't you? That's that's what you want out of a bot it.

0:5:45.990 --> 0:5:46.850
 Steve Tomkinson
 Yeah, that's right.

0:5:47.710 --> 0:5:51.880
 David Antwis
 Otherwise you kind of look at it oddly and they think why? Why?

0:5:53.340 --> 0:5:54.360
 David Antwis
 Ohh yeah.

0:5:52.490 --> 0:6:22.230
 Steve Tomkinson
 Yeah, no, absolutely, absolutely. And and The thing is the messages that if you're leaving a voicemail, this can be quite different than the main conversation. So to reduce the amount of kind of clipped conversation or not the full, not the right messaging and stuff like that, there are a lot of autodialers will be doing then using the smarter voicebot means that then that's managed quite successfully. I mean, look, the limitations, we're not saying that this is perfect. There are limitations even with.

0:6:23.320 --> 0:6:23.830
 Steve Tomkinson
 A.

0:6:25.80 --> 0:6:25.390
 David Antwis
 What?

0:6:26.160 --> 0:6:26.780
 David Antwis
 Human.

0:6:25.480 --> 0:6:31.250
 Steve Tomkinson
 A human, you know? Yeah. Like, if you if some. If you do your voicemail and you go.

0:6:32.640 --> 0:6:32.980
 David Antwis
 Yeah.

0:6:32.370 --> 0:6:33.600
 Steve Tomkinson
 Hi, it's Dave.

0:6:37.150 --> 0:6:37.510
 David Antwis
 Ohh.

0:6:35.520 --> 0:6:49.150
 Steve Tomkinson
 Please leave a message after time. You just telling me that somebody isn't going high, Dave, and then they gone. Ohh. Hi Dave. I'm trying. Ohh. She's voicemail. You know those all those things are real catchy at. And so if you're filling the human.

0:6:49.980 --> 0:6:50.470
 Steve Tomkinson
 Ah.

0:6:47.660 --> 0:6:52.450
 David Antwis
 But is the example. Yeah, that example occurred to me. I was, yeah, yeah.

0:6:53.860 --> 0:6:54.440
 David Antwis
 You talk.

0:6:55.790 --> 0:6:56.50
 David Antwis
 Yeah.

0:6:52.750 --> 0:7:0.400
 Steve Tomkinson
 It's just rubbish, you know, because you've done it. You know you've you've put that in the loop and it and it's just badly formed. So there are limitations.

0:7:1.840 --> 0:7:13.480
 Steve Tomkinson
 But we're all we're trying to do is get as as far to perfect as possible, you know, and and we have done quite a lot of work on our voice detection. Even then the last month or so.

0:7:14.410 --> 0:7:25.170
 Steve Tomkinson
 And I mean some of the solutions like capacitors like Vonage and stuff like that. They're they're looking at about 65% detection rate.

0:7:25.870 --> 0:7:28.880
 Steve Tomkinson
 From what I understand is is there in the industry average.

0:7:29.370 --> 0:7:30.700
 David Antwis
 Detecting voicemail.

0:7:31.750 --> 0:7:32.10
 David Antwis
 Yeah.

0:7:31.100 --> 0:8:1.210
 Steve Tomkinson
 Yeah. So they'll detect it 65% of the time and successfully leave a message at cetera, et cetera, which is pretty low. You know, 35% is a serious amount of calls to voicemail that you're missing out now. Weird. Gough already got it up to about 80%, but with the work that we've recently been doing, don't we get, we're up to about 90% now or over 90% now, which is we're getting up to as much tolerance as you can give to make sure that you're actually.

0:8:2.80 --> 0:8:20.960
 Steve Tomkinson
 Putting through voicemails correctly most of the time, you know, and if you're drilling down and you go down through the numbers and you're doing, you know, 10,000 calls and of those calls, X are gonna be voicemails. You don't connect and then X of those like.

0:8:21.820 --> 0:8:25.0
 Steve Tomkinson
 Up to maybe 10% of those you're talking about?

0:8:39.60 --> 0:8:39.370
 David Antwis
 OK.

0:8:46.300 --> 0:8:47.280
 David Antwis
 Yeah, yeah.

0:8:26.170 --> 0:8:48.340
 Steve Tomkinson
 Very few numbers, of course. You're not really connecting, and of course you don't really pay anyway because we're not. We're not charging you for that kind of conversation anyway, so it's just that we didn't get connected to the person or leave a message, but that is significant difference from 65%, which is the typical for an autodialer and one of the biggest challenges that they have.

0:8:49.820 --> 0:8:57.490
 Steve Tomkinson
 Mainly because they're not smart, you know, they're not using the same NLU power that we are, and because that's the the crux and the.

0:8:58.250 --> 0:8:59.880
 Steve Tomkinson
 The core of a voice bot.

0:9:0.560 --> 0:9:2.330
 Steve Tomkinson
 And I said that's why it's so important.

0:9:8.40 --> 0:9:8.290
 Steve Tomkinson
 Yeah.

0:9:3.810 --> 0:9:8.830
 David Antwis
 Yeah, we do want exciting results. That's that's why we're in this space, isn't it?

0:9:9.740 --> 0:9:10.20
 David Antwis
 Yeah.

0:9:12.160 --> 0:9:12.430
 David Antwis
 And.

0:9:9.190 --> 0:9:15.300
 Steve Tomkinson
 Well, absolutely. And what you know, I suppose the point is, and we always say it is, we're trying to make the voice bot.

0:9:16.250 --> 0:9:18.640
 Steve Tomkinson
 Be assuming as possible or not full if.

0:9:17.670 --> 0:9:19.340
 David Antwis
 Taking the robot out-of-the-box.

0:9:21.230 --> 0:9:21.920
 Steve Tomkinson
 Let's try it.

0:9:20.330 --> 0:9:24.260
 David Antwis
 It's still be some cool way of saying, yeah.

0:9:25.140 --> 0:9:28.880
 Steve Tomkinson
 No, I don't think that's it. We're not using that strum like Dave. I'm sorry about that.

0:9:29.720 --> 0:9:30.690
 David Antwis
 That's all one for free.

0:9:31.70 --> 0:9:36.530
 Steve Tomkinson
 Thanks. Yeah. Yeah, that's right. Yeah, we can have that one free. Doesn't means we don't have to use it.

0:9:37.920 --> 0:10:0.500
 Steve Tomkinson
 So I suppose the the only other thing is that a voicemail isn't the only the challenge we've got, and some of them are like the the large side tool that people do. So we have things like barge in where somebody talks over the bot and the bot will still be listening at that point. And certainly in production where it's applicable and where it makes sense.

0:10:10.510 --> 0:10:10.970
 David Antwis
 You would.

0:10:1.750 --> 0:10:21.260
 Steve Tomkinson
 So you'd have things like. I can't take hold on a minute. Hold on a minute. I'm just on another call. You know that type of thing, which would typically you may get as an agent. Well, those things you can also manage where it's applicable to do that. And the style of the calls. Right. And things like that. But we don't necessarily start from that point because.

0:10:22.340 --> 0:10:23.150
 Steve Tomkinson
 It can.

0:10:24.260 --> 0:10:39.950
 Steve Tomkinson
 You know, being misinterpreted if you're in very noisy environments, there's a lot of shouting going on and stuff like that, so you have to be a little bit careful with things like that, you know, because Google will translate everything you put in front of it. So somebody in the background talking really loudly, it will be.

0:10:42.630 --> 0:10:43.500
 David Antwis
 Yeah. Yes.

0:10:41.70 --> 0:10:46.120
 Steve Tomkinson
 Translating that stuff so some of the background detection and things we're again, we're getting better at that.

0:10:46.760 --> 0:10:59.60
 Steve Tomkinson
 And and and it responses again are good for that, but we can manage that side talk stuff which is really important you know so that that opening again.

0:11:0.180 --> 0:11:4.160
 Steve Tomkinson
 We can put that ourselves on hold. Just go. That's OK.

0:11:5.310 --> 0:11:6.520
 Steve Tomkinson
 Let me know when you're ready.

0:11:7.370 --> 0:11:8.90
 David Antwis
 I don't know less.

0:11:9.950 --> 0:11:10.300
 David Antwis
 Yeah.

0:11:7.260 --> 0:11:10.740
 Steve Tomkinson
 And it's just looking for a positive response, you know, so it's just going.

0:11:11.440 --> 0:11:14.670
 Steve Tomkinson
 So they're just waiting like you would as an agent.

0:11:14.530 --> 0:11:18.210
 David Antwis
 Let's see that it's it's great. It's yeah.

0:11:16.670 --> 0:11:18.890
 Steve Tomkinson
 Yeah, I know, I know. And then just go. Ohh, hi.

0:11:19.270 --> 0:11:26.180
 Steve Tomkinson
 I that. Yeah. Sorry. What was it in the bot starts introducing itself again, you know, and then you're off again.

0:11:24.780 --> 0:11:33.400
 David Antwis
 And you, you see that as an example. Anything. Well, what else can we do? The possibility is just start there, don't they? And they and they go on. Yeah, yeah.

0:11:35.90 --> 0:11:35.350
 David Antwis
 Yeah.

0:11:31.600 --> 0:11:36.740
 Steve Tomkinson
 That's right. No, no, it's it's spawn. I mean, IVR are a different thing.

0:11:37.0 --> 0:11:54.470
 Steve Tomkinson
 Umm, because you get a lot of that as well. If you're doing B2B stuff. So we've got some exciting projects coming up that are looking to be B2B around collections and stuff. So you have to get through to the right departments. And from that perspective again.

0:11:55.170 --> 0:12:10.440
 Steve Tomkinson
 Did I speaking to Tom earlier about how IVR he was saying? Well, look now I've been doing this for a couple of decades now. We've got Ivo nailed down. It's absolutely Bob on. We know exactly what we're doing with that. There is no limitation really with that.

0:12:15.490 --> 0:12:15.890
 David Antwis
 But.

0:12:11.490 --> 0:12:28.220
 Steve Tomkinson
 But you you have it on the context of what the call is, you know. So if if it's a an accounts debt collection stuff, then you're obviously looking for that particular option within the Ivars. If you're looking for customer service, if you're looking for general inquiries, whatever it may be.

0:12:31.630 --> 0:12:33.70
 David Antwis
 Nailed it, yeah.

0:12:29.70 --> 0:12:39.440
 Steve Tomkinson
 You deal with it. At that point, you know, and then you introduced definitely because you're gonna expectation, you know. So the opening conversation is completely different, you know, which is really strong.

0:12:40.780 --> 0:12:48.960
 Steve Tomkinson
 You know, so there are there are loads of these are opening challenges, but the voicemail is quite a particular one and we've been.

0:12:49.840 --> 0:13:0.70
 Steve Tomkinson
 Talking about it this week and I thought it would be valid for the podcast to kind of have some sort of explainer about how that works. You know, because it it is important.

0:13:0.860 --> 0:13:1.200
 Steve Tomkinson
 Yeah.

0:12:58.540 --> 0:13:1.430
 David Antwis
 I found it very interesting. Yeah, yeah.

0:13:2.790 --> 0:13:5.520
 Steve Tomkinson
 I mean, look, it's it's all about the overall.

0:13:6.800 --> 0:13:8.850
 Steve Tomkinson
 Natural conversation.

0:13:9.890 --> 0:13:21.720
 Steve Tomkinson
 That that our voice bots take, you know, again, we're never trying to fool anybody that they're not a robot. But people will forget. And that's the that's the absolute.

0:13:22.160 --> 0:13:22.980
 David Antwis
 That's the magic.

0:13:23.750 --> 0:13:24.120
 David Antwis
 Yeah.

0:13:22.630 --> 0:13:45.470
 Steve Tomkinson
 Jewel in the crown, you know, that is the point is that you just forget because it's kind of easy. You can say random stuff to it without too much difficulty. Still in context to a point you are funneling people to a certain extent, but you're only making the call because it's relevant. So you're not doing anything outside of those things and it works quite well.

0:13:47.450 --> 0:13:50.20
 David Antwis
 I'm picking up the you. You do love voice bots?

0:13:51.780 --> 0:13:56.580
 Steve Tomkinson
 Yeah, I like. I think the the problem with guys were against the background now where?

0:13:57.310 --> 0:14:0.210
 Steve Tomkinson
 It's really, really tough to get the.

0:14:1.90 --> 0:14:14.540
 Steve Tomkinson
 Volume of people you need in contact centres and you know and and have that make those calls and make those kind of outbound calls or even manage the inbound calls and those are.

0:14:15.220 --> 0:14:35.830
 Steve Tomkinson
 These are crucial parts of the business we still use calls and voice a lot and it's a very inclusive environment as well. So it doesn't matter what age you are, you're not technologically bound. So voice is gonna have its place for quite some time and it's important in this day and age when we've got all this.

0:14:37.660 --> 0:14:39.990
 Steve Tomkinson
 Absolute monkey business going on around us.

0:14:41.0 --> 0:14:48.200
 Steve Tomkinson
 The the businesses have an option to go right? We have we are behind by 15,000 calls. How do we do that?

0:14:50.860 --> 0:14:51.240
 David Antwis
 Yeah.

0:14:49.20 --> 0:14:51.800
 Steve Tomkinson
 How do we do that? You know, I was literally talking to.

0:14:52.210 --> 0:15:6.950
 Steve Tomkinson
 I'm a new client the other day and he's going. Look I'm. I've got 15,000 calls that we need to make. I can't make them. We can't make them. We have all this business that's to one side. I can't make them. They've asked us for.

0:15:7.640 --> 0:15:9.480
 Steve Tomkinson
 A communication and I can't do it.

0:15:10.610 --> 0:15:13.200
 David Antwis
 It's a common situation, isn't it?

0:15:13.490 --> 0:15:43.240
 Steve Tomkinson
 It's nuts. It's crazy, you know? But and this is part of the whole story, you know, making sure that we can do that, you know, the The thing is, these systems aren't just dial this, they're automation systems, you know. So even if we do get a voicemail or we get somebody, we're not quite sure it's a voicemail. We will phone them back. We'll reschedule the call. So it will hit it again. So all these things are part of a an overall automation. So it's huge. It's a huge space and.

0:15:49.440 --> 0:15:50.360
 David Antwis
 That's where we're at.

0:15:43.700 --> 0:15:50.510
 Steve Tomkinson
 Happily, we've got a great platform that manages it, so you know that's, that's where we're at and that's how it's gonna work.

0:25:32.470 --> 0:25:46.490
 David Antwis
 Thanks, Steve. I hope everyone found that useful. If there are any topics that you would like us to discuss, let us know via info, it disruptionworks.co.uk and we'll see what we can do until next time goodbye.

0:25:47.120 --> 0:25:47.510
 Steve Tomkinson
 Cheers.

People on this episode