1
00:00:00,200 –> 00:00:06,800
“AI shouldn’t improve CX. It should replace it.” What did you mean by that? Channels today
2
00:00:06,800 –> 00:00:12,000
have the tendency to be very fragmented. Is the risk of hallucination still a real risk, or is it
3
00:00:12,000 –> 00:00:18,240
largely a solved problem? Well, we have seen the last 6 months is that 20%, 30% has become now
4
00:00:18,360 –> 00:00:25,280
60% and 70% and we actually have some customers at 75% automation.
We all know it’s super easy
5
00:00:25,280 –> 00:00:31,240
to set up an LLM bot, so you will sometimes have people from the IT teams just build up something.
6
00:00:31,240 –> 00:00:36,280
Hey, we just scraped the web page, scraped the knowledge base, we built the RAG, we’ll launch it and
7
00:00:36,280 –> 00:00:40,520
we’re good to go. We don’t need to do anything more. But as we know, if you really want to kind of
8
00:00:40,560 –> 00:00:45,480
push them a lot, what’s possible in this technology, you really, really need to have the
9
00:00:45,480 –> 00:00:51,840
business, the CX, the contact centre people in. How can they be ready for the future when everyone
10
00:00:51,840 –> 00:00:57,520
has their own AI? And then it doesn’t really cost me any money to call, having my own voice bot to
11
00:00:57,520 –> 00:01:02,009
call your voice bot. So it was like 2 agents talking to each other and like, so it’s happening
12
00:01:02,009 –> 00:01:08,809
now, you know?
This conversation is with Henry Vaage Iverson, who is the CCO and Co-Founder at
13
00:01:08,809 –> 00:01:14,690
Boost.AI. Boost.AI is a Gartner Magic Quadrant leader. One of the players in the
14
00:01:14,690 –> 00:01:20,689
conversational AI for enterprise space. And this conversation is all about the rising
15
00:01:20,730 –> 00:01:27,050
trend around voice AI. Despite what you might think, after all the investment in digital
16
00:01:27,129 –> 00:01:34,010
native channels, in automation, in mobile, in web, the call volumes in contact
17
00:01:34,010 –> 00:01:39,409
centres are still rising and weirdly, I don’t know why, we speak about why actually. So I do know why
18
00:01:39,410 –> 00:01:46,329
now, but the voice AI adoption has really started to take off in businesses. I mean, the
19
00:01:46,330 –> 00:01:51,529
technology was kind of there 10 years ago, 5 years ago, but what large language models and
20
00:01:51,529 –> 00:01:56,649
generative AI has done is they’ve made these conversations really, really natural. So this
21
00:01:56,649 –> 00:02:02,099
conversation with Henry is all about how do you really maximise voice AI? How do you approach
22
00:02:02,099 –> 00:02:06,860
going from an AI agent that you have on a chat channel, putting it into the voice channel? What
23
00:02:06,900 –> 00:02:10,219
differences need to be considered? What kind of things do you need to change? What do you need? How
24
00:02:10,220 –> 00:02:15,500
do you need to test it? What does good look like and how do you then go ahead and scale it? So, if
25
00:02:15,500 –> 00:02:21,299
you’re interested in voice AI, if you’re interested in taking advantage of, finally, what
26
00:02:21,300 –> 00:02:26,739
I’ve been talking about since 2017, which is the voice AI revolution and the value of voice AI for
27
00:02:26,740 –> 00:02:31,619
your business, this conversation is one that you should absolutely pay attention to. Henry knows
28
00:02:31,619 –> 00:02:35,979
his stuff. He is incredibly experienced in this space. He’s got over a decade of experience working with
29
00:02:35,979 –> 00:02:40,819
this type of technology in the creation of Boost.AI. Boost.AI, I am a big fan of. It’s a great platform,
30
00:02:40,820 –> 00:02:45,619
great company, great people, great culture. And so this conversation I’m sure you’ll enjoy with
31
00:02:45,619 –> 00:02:48,139
Henry Vaage Iverson from Boost.AI.
32
00:03:01,350 –> 00:03:07,949
All right, Henry, welcome. Welcome back, I should say. Yeah, thank you Kane. Yeah, you’re welcome. Yeah, y ou joined us
33
00:03:07,950 –> 00:03:13,389
on, we had a chat a few months back when the whole Cognigy kind of acquisition happened. You
34
00:03:13,390 –> 00:03:17,310
joined us on a panel sharing some great insights. So, yeah, glad to speak to you
35
00:03:17,310 –> 00:03:23,750
properly again. Yeah, good to see you again. It feels like a lifetime, to be honest, but pretty
36
00:03:23,750 –> 00:03:29,070
fast in our industry. It is moving fast, it is, it is indeed. And, yeah, Boost is on the up and up.
37
00:03:29,070 –> 00:03:34,630
What we did, I saw something recently. Was it a Forrester wave or something? Something like that.
38
00:03:34,710 –> 00:03:38,949
What boost was featured in. I can’t remember exactly what it was. See, so I don’t know what it was. Yeah,
39
00:03:38,990 –> 00:03:42,590
we’re featuring a lot of stuff, though. Uh, but obviously, we are a leader in the Magic Quadrant,
40
00:03:42,590 –> 00:03:48,350
so. And, there are also a lot of other analysts, where we’ve been featured as a leader. So
41
00:03:48,389 –> 00:03:53,390
obviously, great to be recognised by the analysts as well. Yeah, 100%, I think I think kind of like
42
00:03:53,429 –> 00:03:58,759
Boost is definitely emerged and established itself as like a real sort of
43
00:03:58,800 –> 00:04:02,800
front runner in this space. So when it comes to conversational AI, you know, I think obviously
44
00:04:02,800 –> 00:04:07,160
you’ve got a good pedigree and, you know, building the company and really sort of like
45
00:04:07,160 –> 00:04:11,229
capitalising on the Nordics and stuff like that. But I would definitely say in the last sort of 2
46
00:04:11,230 –> 00:04:17,280
or 3 years, Boost is out here, up there with the establishment in terms of like a real serious
47
00:04:17,280 –> 00:04:22,720
player in the AI space. Yeah, yeah for sure and we are kind of born in the bank, actually, so I think
48
00:04:22,720 –> 00:04:27,359
it’s kind of our success has definitely been our focus on the regulated industries, which obviously
49
00:04:27,359 –> 00:04:33,358
include a lot within the financial services, but also in government, Telcos, but also in general,
50
00:04:33,359 –> 00:04:39,200
everyone, which has a complex need when it comes to customer service and CX. So I think that’s kind of
51
00:04:39,440 –> 00:04:44,359
being our sweet spot is shall we say, we love the complexity. And I think that’s something our
52
00:04:44,359 –> 00:04:50,119
customers really appreciate working with us. Yeah, yeah. What do regulated industries pose in
53
00:04:50,119 –> 00:04:56,219
terms of a challenge then in particular compared to non-regulated industries? Yeah. So I think
54
00:04:56,260 –> 00:05:02,019
obviously like nowadays they like building a LLM bot. It’s kind of a commodity. I think all
55
00:05:02,019 –> 00:05:08,380
of us can build something. You can, there’s a lot of tools you can use, external facing
56
00:05:08,380 –> 00:05:12,700
basically just kind of create something pretty quickly. But then obviously when you work with the
57
00:05:12,700 –> 00:05:17,660
bank, work with insurance company, they have certain kind of ways of working. There’s
58
00:05:17,700 –> 00:05:21,540
obviously one thing you need to think about, it’s just the whole aspect of like governance. How do
59
00:05:21,540 –> 00:05:28,220
you track? What you do in the solution? How do you make sure that you monitor changes, monitor
60
00:05:28,420 –> 00:05:35,379
maintenance, development and also improvements over time? Things like privacy,
61
00:05:35,579 –> 00:05:38,980
security is obviously also extremely important. And we’re kind of gone through all the
62
00:05:38,980 –> 00:05:44,939
certification, in terms of ISO. The latest one was HIPNO, we added. So those things are
63
00:05:44,940 –> 00:05:51,189
also super important. So we definitely make sure that if your organisation is in regulated industries,
64
00:05:51,230 –> 00:05:57,109
like with us, you can trust every conversation. So basically you are able to serve your customers
65
00:05:57,109 –> 00:06:03,150
at the highest level of quality. But then there’s also an opportunity for you as a customer to have
66
00:06:03,150 –> 00:06:07,629
an audit trail on everything which has happened in the conversation. So it’s not really a
67
00:06:07,630 –> 00:06:12,549
black box, because again, a lot of the stuff you can develop yourself if you build a LLM, that
68
00:06:12,549 –> 00:06:17,829
will very much be a black box. Yeah. Yeah. And that’s the challenge, isn’t it? With
69
00:06:17,870 –> 00:06:23,509
LLMS sometimes, I mean, they’ve gotten better in terms of kind of trying to explain their
70
00:06:23,510 –> 00:06:29,029
reasoning. Certainly, if you look at your likes of Claude and the one I say, Claude, I make them more
71
00:06:29,029 –> 00:06:34,429
consumer-facing applications. And I think that’s probably part of that kind of trust thing, isn’t
72
00:06:34,430 –> 00:06:38,109
it? You know, like helping the user understand what it’s done and how it’s gone through what it’s
73
00:06:38,109 –> 00:06:42,829
done type of thing. But ultimately, fundamentally, they are still a black box in terms of like, you
74
00:06:42,829 –> 00:06:48,268
can’t necessarily understand exactly why it’s predicted the tokens it’s predicted. But at least
75
00:06:48,269 –> 00:06:53,040
in a platform like Boost, presumably, you can get visibility into the pipeline and in terms of like
76
00:06:53,040 –> 00:06:57,040
what happened, where and when and ultimately be able to pinpoint if anything has gone wrong, where
77
00:06:57,040 –> 00:07:01,559
it has gone wrong and stuff like that, is it? Yeah, and it’s fundamental, very different use cases
78
00:07:01,559 –> 00:07:06,679
because obviously as a consumer I can live with it as a black box. As long as it’s useful, I can
79
00:07:06,679 –> 00:07:12,239
live. It’s not 100% very time I’ll ask a question, that’s good enough because obviously
80
00:07:12,239 –> 00:07:17,359
it’s creating a lot of value for me regardless. But I think if you are in, if you’re a bank and
81
00:07:17,359 –> 00:07:21,839
insurer, you definitely wanna make sure that every conversation you have is trusted in the
82
00:07:21,839 –> 00:07:26,719
sense that you actually deliver the level of quality you want. Because if you give a wrong
83
00:07:26,720 –> 00:07:31,960
quote, for example, maybe you hallucinate the product you don’t have, for example. Maybe you give
84
00:07:31,960 –> 00:07:37,799
people a financial advice. Those things obviously is not really something any one of our customers
85
00:07:37,799 –> 00:07:43,159
will accept, because that is things that will basically make the headline at some point, if
86
00:07:43,160 –> 00:07:49,809
it happens. Yeah. So I think hallucination is obviously still a key part of LLMS, I would say,
87
00:07:49,810 –> 00:07:55,729
but we need to be able to reduce that as much as possible because that’s definitely not a
88
00:07:55,730 –> 00:08:00,209
good customer experience for sure. Yeah. How much of a problem is hallucinations these days then?
89
00:08:00,209 –> 00:08:05,729
Because if you look at again, if you look at your likes of Claude and ChatGPT and those kind of
90
00:08:05,889 –> 00:08:10,650
again, it’s different use cases, but the consumer-facing stuff, they do tend to have hiccups every
91
00:08:10,650 –> 00:08:14,609
now and then, but they’re an awful lot better than they were. Whereas when you’re working with the
92
00:08:14,609 –> 00:08:18,849
API, you’re not always working with the same kind of layers of guardrails and all that kind of
93
00:08:18,849 –> 00:08:24,329
stuff that those applications have. So what’s your sort of perspective now on, like, is the risk of
94
00:08:24,329 –> 00:08:29,449
hallucination still a real risk, or is it largely a solved problem? I would definitely say it’s
95
00:08:29,450 –> 00:08:34,769
still a big risk. And I think for the consumer facing one, what I learned using the tool, I
96
00:08:34,810 –> 00:08:41,089
obviously use AI a lot, they are they tend to want to please you. Yeah. So for
97
00:08:41,090 –> 00:08:46,819
example, I’m very keen on running. So I’m coming back from an injury. Now and then,
98
00:08:46,819 –> 00:08:51,539
if I have any kind of good progress in the injury, then I will obviously ask them just to see, okay,
99
00:08:51,700 –> 00:08:58,539
how I am tracking on this injury right now? Can I maybe increase the volume of running now
100
00:08:58,539 –> 00:09:02,659
since I’m feeling better? And then, depending on how I ask the question, it will come back with
101
00:09:02,659 –> 00:09:06,699
very different response. Because obviously it wants to please me. If I’m very cautious when
102
00:09:06,739 –> 00:09:10,739
asking, maybe I need to be a bit more cautious. I would say yes, of course, you are totally right. You
103
00:09:10,739 –> 00:09:14,620
need to be cautious there on the other side saying, hey, I have a race next week. Is it a good,
104
00:09:14,900 –> 00:09:20,299
good idea and I have not practised it for. Yeah, it’s probably going to be a stretch, we probably can
105
00:09:20,299 –> 00:09:26,579
get you there. So I think it’s still a challenge. But I think when it comes to kind of
106
00:09:26,620 –> 00:09:32,859
businesses, when you serve on customer service, you don’t really have that kind of that type of
107
00:09:33,260 –> 00:09:38,659
conversation. So every conversation would be very much dedicated to one use case. You have
108
00:09:38,700 –> 00:09:42,979
kind of one problem you want to solve. So then it’s obviously much easier to put guardrails on
109
00:09:42,979 –> 00:09:48,469
it. And since we have basically, we’re done a bit unique in terms of guardrails. We’re not building
110
00:09:48,670 –> 00:09:54,549
in guardrails in instruction. We actually have separate LLMs monitoring the conversation. So as
111
00:09:54,549 –> 00:09:59,109
you go as part of the conversation, there’s actually going to be numerous different LLMs
112
00:09:59,109 –> 00:10:04,829
monitoring the conversation to come in and kind of guard that conversation and then have
113
00:10:04,830 –> 00:10:09,629
obviously an escalation on that if needed. So that means that I’m not going to say it’s going to be
114
00:10:09,630 –> 00:10:15,749
zero chance of hallucination, but we’re getting pretty close to 0% of hallucination, which is
115
00:10:15,749 –> 00:10:19,390
obviously a good thing. That’s brilliant. Yeah. Because that was the big prohibitor for
116
00:10:19,390 –> 00:10:25,869
a long time, I think, really. I saw on your LinkedIn a week or so ago that you were giving
117
00:10:25,869 –> 00:10:32,189
a talk somewhere and you had a slide behind you that said AI shouldn’t improve CX, it should replace
118
00:10:32,190 –> 00:10:38,949
it. What did you mean by that? Yeah. So I think as we work with a lot of organisations,
119
00:10:39,030 –> 00:10:42,718
we have also learned that a lot of organisations can of tend to going to want to
120
00:10:42,759 –> 00:10:48,239
improve the existing customer experience. So especially when it comes to banks, they had they
121
00:10:48,239 –> 00:10:54,319
had a lot of forms. They obviously have invested a lot of money in self-service, mobile banks and so
122
00:10:54,320 –> 00:11:00,959
on. And the problem with these also it becomes kind of a channel, which is obviously not
123
00:11:00,960 –> 00:11:05,159
easy to use because obviously, with all the stuff you want to put into it, there’s going to be a lot
124
00:11:05,159 –> 00:11:10,279
of complexity in terms of menus and how you want to navigate, and then people tend to like, how
125
00:11:10,280 –> 00:11:15,199
can we use AI to make those better? Because a lot of our, like, a lot of people have historically
126
00:11:15,200 –> 00:11:21,399
then used AI to steer people into a mobile bank, maybe steer them into a self-service tool. But I
127
00:11:21,400 –> 00:11:26,400
think where we come from is like, I think you basically need to redesign everything, because if
128
00:11:26,400 –> 00:11:31,999
you try to build AI on legacy, yes, sure, you’re going to get some short-term results of doing
129
00:11:32,000 –> 00:11:38,199
that for sure. But I think if you look at the longer picture and also looking at the
130
00:11:38,200 –> 00:11:43,650
potential of AI technology, I think you need to start rethinking a lot, but also start rethinking
131
00:11:43,650 –> 00:11:49,489
channels as a whole, because again, channels today has the tendency to be very fragmented. You can
132
00:11:49,489 –> 00:11:54,129
start the conversation somewhere, continue another one. It’s not really kind of collecting any
133
00:11:54,130 –> 00:11:59,489
context or any of that history. And again, with I think the promise for me, at least for AI, is
134
00:11:59,489 –> 00:12:06,288
basically it’s representing a completely new UI where you can then obviously talk to it or write
135
00:12:06,289 –> 00:12:11,489
to it, and it doesn’t really matter which channel you are in. I think that’s kind of the premise
136
00:12:11,489 –> 00:12:16,649
of, of the promise of the technology, which I think very few, to be honest, has really kind of gone into.
137
00:12:16,650 –> 00:12:23,049
But I think we are now getting much more customers kind of buying into that vision and
138
00:12:23,049 –> 00:12:27,649
start rethinking how they do things today. Yeah, which ultimately is what it’s all about, isn’t it?
139
00:12:27,689 –> 00:12:33,370
It’s a, you know, AI really is a transformative technology, and it kind of like pains me a little
140
00:12:33,409 –> 00:12:37,810
bit when I come across those use cases, which is, we’ve got all of these documents, let’s just, you
141
00:12:37,810 –> 00:12:43,340
know, at, the term that is starting to really bug me is just talk to our data. We want to talk to
142
00:12:43,340 –> 00:12:48,499
our data. It’s like, do you read? Do your customers just want to talk to your data? Really? You know,
143
00:12:48,539 –> 00:12:52,419
your customers want to get something done, and that data is probably actually going to prevent
144
00:12:52,419 –> 00:12:58,059
them from doing the thing they’re trying to do. And it kind of, yeah, I kind of, who was it? I
145
00:12:58,059 –> 00:13:04,499
remember who it was I spoke to recently that was talking about how sometimes AI taking
146
00:13:04,500 –> 00:13:11,379
the place of a phone call, if it does nothing else but just replace the phone call,
147
00:13:11,420 –> 00:13:16,938
then the business still hasn’t really gained because the phone call still happens. The person
148
00:13:16,939 –> 00:13:21,299
who used to answer that phone call is still there, still doing the job, and work takes up the time
149
00:13:21,299 –> 00:13:25,738
that they have. So unless it’s going to actually do as you say and actually transform something,
150
00:13:25,739 –> 00:13:30,899
get into the process, help you orchestrate the process, re-engineer how you deliver value,
151
00:13:30,900 –> 00:13:34,819
re-engineer how you serve customers, you know. So rather than it being one channel, it’s every
152
00:13:34,819 –> 00:13:39,189
channel. And rather than it being a front-end lipstick on a pig doing the UI. It’s actually
153
00:13:39,230 –> 00:13:44,109
orchestrating the actual process. That’s when it starts to become more valuable, isn’t it? Yeah, and
154
00:13:44,109 –> 00:13:49,390
I think also like to your point like when something goes kind of like the customer
155
00:13:49,390 –> 00:13:53,629
journey within an organisation is pretty much like, you know what they need. They will go into
156
00:13:53,669 –> 00:13:57,429
your web page, discover your product. They will always investigate in terms of your product,
157
00:13:57,430 –> 00:14:01,989
compare with others. There’s obviously kind of the whole purchasing, buying the product, but then also
158
00:14:02,030 –> 00:14:06,349
kind of when you kind of bought it, like the after service and so on, like as an organisation, you
159
00:14:06,349 –> 00:14:10,709
know, all these things. And you should also be able to know when Henry need support before actually
160
00:14:10,750 –> 00:14:15,869
Henry reaching out. I think that’s also some of the things people are also overlooking a bit
161
00:14:15,870 –> 00:14:20,869
because you are thinking about contact centres more like a as a traditional department. We’re, it’s
162
00:14:20,909 –> 00:14:25,069
kind of inbound. We’re waiting for people get a problem and then we solve it in the best possible
163
00:14:25,069 –> 00:14:31,269
way. But the challenge with that model is, to your point, that the volume is not going down,
164
00:14:31,510 –> 00:14:38,290
it’s actually going up. Yeah, the volume is going up and I think, like, if you look at
165
00:14:38,290 –> 00:14:42,729
phone calls, I think for 10 years ago we were like very much like, hey, phone. No one’s going to
166
00:14:42,770 –> 00:14:46,689
call anymore because obviously now we can do messaging. The phone calls are pretty stable, and
167
00:14:46,689 –> 00:14:50,969
that obviously depends a lot on the region, of course, but even in the Nordics, where things
168
00:14:50,970 –> 00:14:57,889
are very much digital and kind of messaging, phone calls are still the biggest channel.
169
00:14:57,889 –> 00:15:02,649
So I think that’s something people really overlook, like the ability to be more proactive in
170
00:15:02,650 –> 00:15:07,009
the customer journey and having a kind of 1 to 1 relationship with customers at scale, which is
171
00:15:07,010 –> 00:15:11,089
also I think is pretty interesting. Yeah. Yeah, exactly. It’s a rethink, I think, from the whole
172
00:15:11,090 –> 00:15:14,570
organisation to go in that direction. Because if you think about contact centre and reducing costs
173
00:15:14,570 –> 00:15:19,929
there, this is obviously a completely different kind of use case in that sense. Yeah. But
174
00:15:19,929 –> 00:15:26,129
ultimately that’s where, that’s how a business actually becomes better though isn’t it. So the
175
00:15:26,129 –> 00:15:31,009
the kind of, you know, the whole like 5 whys process isn’t it. If you follow that down to try
176
00:15:31,010 –> 00:15:36,778
and get to the root cause of why someone’s contacting and then address the root cause, and if
177
00:15:36,779 –> 00:15:41,499
you can’t address the root cause, pre-empt the root cause. It’s all that kind of stuff. Like, you
178
00:15:41,500 –> 00:15:44,339
know, people are not calling because they want to. They’re calling because of a reason that
179
00:15:44,340 –> 00:15:48,020
something’s happened, and so why has that happened or something failed? Well, why is that a failure or
180
00:15:48,020 –> 00:15:52,939
because of something? Or why’s that happened? If you keep kind of peeling back the layers, that’s how
181
00:15:52,940 –> 00:15:57,699
you end up actually solving the problem rather than just trying to replace the interruption. Yeah.
182
00:15:57,780 –> 00:16:02,059
And I think I completely agree. And this is also kind of where the whole trust part
183
00:16:02,100 –> 00:16:08,379
also plays a very significant role because both on the end customer side, they need to have trust
184
00:16:08,380 –> 00:16:12,900
in these systems. Like if you talk to an AI, you need that trust. Does it take care of my data in
185
00:16:12,900 –> 00:16:19,500
the proper, proper way? That’s the kind of abuse any of my data? Does it really understand me
186
00:16:19,500 –> 00:16:23,939
in the end? Like, because obviously things also very contextual. Things could also change. I can
187
00:16:23,939 –> 00:16:29,379
have data about Henry, or I can have data about you, Kane, but obviously, we also change as people. So
188
00:16:29,579 –> 00:16:33,869
how do you make sure we update that? So that’s one part of it. But also organisations really kind of
189
00:16:33,870 –> 00:16:39,229
start utilising this, really need to be able to understand and trust these systems. So this
190
00:16:39,229 –> 00:16:43,549
again kind of being able to have that transparency and the solution is just extremely
191
00:16:43,550 –> 00:16:47,949
important because you need to figure out how the AI comes to a conclusion, how the AI actually
192
00:16:47,950 –> 00:16:52,189
responded, where it fetched information, and how can I build up the logic. If you’re not able to
193
00:16:52,190 –> 00:16:56,669
give that out for every conversation, then it’s going to be very difficult to do any of the
194
00:16:56,670 –> 00:17:03,669
things we talk about now. Yeah, 100%. Where do you stand on stuff like computer use and those type
195
00:17:03,669 –> 00:17:10,468
of technologies? So again, on the consumer side, those tend to make sense because you know, having
196
00:17:10,469 –> 00:17:17,469
an AI like Claude or Open Claw or whatever, use your computer in your absence to do stuff makes
197
00:17:17,510 –> 00:17:23,269
kind of sense. But, or use the browser. You know, navigate to a website in your browser. That all
198
00:17:23,270 –> 00:17:29,310
makes sense for me, at least, because you’re not in control of any of that. I can’t go and just make
199
00:17:29,310 –> 00:17:33,959
Booking.com give me an API that I can authenticate with my user and then pass that to
200
00:17:34,000 –> 00:17:39,919
to Claude. I can’t do that. But a business that at least has at least on the face of it, more
201
00:17:39,920 –> 00:17:44,680
control. So I’m kind of caught in 2 worlds. I’d be interested in getting your perspective on
202
00:17:44,680 –> 00:17:49,279
stuff like computer use and stuff like that. Is it like, is that something that can really unlock
203
00:17:49,280 –> 00:17:55,319
value for a business on the back end, or is that just papering over a crack in lieu of an API or a
204
00:17:55,319 –> 00:18:00,919
more efficient way of building systems? Yeah. So you mean kind of having your own kind of personal
205
00:18:00,960 –> 00:18:05,838
kind of AI then talking to an organisation? No, no. Sorry. No, I meant like, for example, let’s say that
206
00:18:05,839 –> 00:18:10,959
there’s a business that has been blocked on some kind of initiative because they don’t have an API
207
00:18:10,959 –> 00:18:16,079
access to this system and instead what they’ll have is they’ll have, like a boost agent on the
208
00:18:16,079 –> 00:18:20,959
front end that then triggers a process that because they don’t have APIs, they’ll use this
209
00:18:20,959 –> 00:18:25,199
other technology that’s emerging like computer use in order to navigate some of their back
210
00:18:25,239 –> 00:18:29,890
office systems. Yeah, because I’m hearing a lot of that lately and businesses like that. That would
211
00:18:29,890 –> 00:18:34,169
be really useful for us because it could just do this type of thing. So like, I don’t know if
212
00:18:34,209 –> 00:18:37,369
there’s a lot of businesses now who are wondering about our technology, whether they could be using
213
00:18:37,369 –> 00:18:41,409
it internally. But for me, I’m not sure whether that’s the right thing to do, because surely it
214
00:18:41,410 –> 00:18:46,410
should fix the problem, which is your data and your connectivity to that data, you know? Yeah, it
215
00:18:46,410 –> 00:18:51,289
definitely seems that this organisation has some kind of, say, legacy systems, but also not
216
00:18:51,290 –> 00:18:55,249
being able to kind of pull out the data when you need it. And I think that’s probably also the
217
00:18:55,250 –> 00:18:59,049
reality of a lot of organisations. That’s just obviously also the thing though, like if you have
218
00:18:59,050 –> 00:19:03,809
that challenge and I think we’re also seeing something similar, I think you have a lot of banks,
219
00:19:03,849 –> 00:19:09,689
a lot of insurance, also have a lot of like legacy systems, which is not really easy to connect to. So
220
00:19:09,689 –> 00:19:13,689
then you just need to figure out how we can use AI to extract information from that and kind of
221
00:19:13,689 –> 00:19:18,689
use in the conversation. But, but I think in general, if you are able to rethink it, that’s
222
00:19:18,729 –> 00:19:23,409
obviously the best way to go, but I can definitely see the case of kind of doing
223
00:19:23,410 –> 00:19:29,380
something in the in the meantime, if you want to get things up and running. But let’s make sure
224
00:19:29,380 –> 00:19:32,660
we don’t focus too much on that solution and think that’s the end product, because I think that would
225
00:19:32,660 –> 00:19:38,379
definitely be a trap. Mhm. Yeah, definitely. You mentioned there that you know voice, the
226
00:19:38,379 –> 00:19:42,420
voice channel is kind of remained stable. I actually came across some research that I
227
00:19:42,420 –> 00:19:49,420
couldn’t properly get into the source of it, but it was a company, I don’t think it was a tel/,
228
00:19:49,620 –> 00:19:56,459
a C-Cass company, but it was something similar to that. Who did a study of 600
229
00:19:56,459 –> 00:20:02,900
businesses and profiled their voice kind of demand. And I would say that it’s anecdotal
230
00:20:02,900 –> 00:20:06,420
because, again, I couldn’t find the source of the data. I just saw it in the,
231
00:20:06,420 –> 00:20:10,140
article from the company that did the study. I’ll try and dig it out and put in the show
232
00:20:10,140 –> 00:20:13,740
notes. But it was suggesting, actually, that the opposite, not the opposite, but it was
233
00:20:13,740 –> 00:20:18,299
suggesting that voice is actually growing as a channel in terms of demand, which is totally
234
00:20:18,300 –> 00:20:23,339
counter to what you would expect, given the investment in digital and mobile and all that
235
00:20:23,339 –> 00:20:29,589
kind of stuff. And so at the same time, though, it’s taken, it feels like for me, and I don’t know if
236
00:20:29,590 –> 00:20:33,910
it’s the same for you, but it seems like it’s taken a while for businesses to really kind of
237
00:20:33,949 –> 00:20:40,069
realise that if the voice channel is growing, AI is a perfect fit for that channel, especially now
238
00:20:40,069 –> 00:20:44,030
with LLMs. So I don’t know what you’re seeing now. And I’ve done this a few times. We did a webinar
239
00:20:44,030 –> 00:20:48,510
with Boost with Sam Rosendorff last year about voice AI, and we spoke a little bit about the
240
00:20:48,510 –> 00:20:54,229
trends. It feels as though everyone’s got really interested in voice AI in 2026, but it’s
241
00:20:54,230 –> 00:20:58,269
hard to tell whether what is that just noise and hype or is that actually translating to
242
00:20:58,310 –> 00:21:03,790
deployment? So what are you seeing from your customers in terms of voice AI adoption? I would
243
00:21:03,790 –> 00:21:08,109
definitely say that most of our kind of major customers are definitely going into voice now.
244
00:21:08,110 –> 00:21:11,948
Some of them obviously have launched already and some of them are in pilot, proof of concept, but I
245
00:21:11,949 –> 00:21:16,790
think they all really much think about it. I think again, the US market, which is a bit different,
246
00:21:16,790 –> 00:21:22,949
they’ve definitely been on that train for like maybe 3 or 4 years, I think I’m definitely
247
00:21:22,949 –> 00:21:26,680
seeing Europe now starting to catch up. And I think there has also been some limitation in
248
00:21:26,680 –> 00:21:32,400
terms of English compared to some of the other languages, including the Nordics and Slovenian and
249
00:21:32,400 –> 00:21:38,119
stuff like that. But I also saw some benchmark now in terms of like how well these models have
250
00:21:38,119 –> 00:21:42,519
improved over the last year, because I think I don’t think there’s been a lack of interest
251
00:21:42,520 –> 00:21:47,519
around voice, especially in some of the countries in Europe. But it’s just been honestly a
252
00:21:47,520 –> 00:21:53,399
bit lack of kind of the, the capabilities of the speech text. And now we obviously have a lot of
253
00:21:53,400 –> 00:21:59,239
new voice models coming out, speech-to-speech and so on, which is then creating a much better
254
00:21:59,239 –> 00:22:05,999
natural conversation. But I would say most of the customers we get in
255
00:22:06,000 –> 00:22:11,799
now, they will have voice as part of that project, most of them, which is obviously a significant
256
00:22:11,840 –> 00:22:17,359
uptick for just a couple years ago, when we talked about maybe 20% of customers were doing
257
00:22:17,359 –> 00:22:23,129
voice, now most of them are doing it. That’s brilliant. So I think we are going to continue
258
00:22:23,170 –> 00:22:28,529
seeing a significant investment in voice, and people have a bit different perspective where
259
00:22:28,530 –> 00:22:35,129
they want to start. Again, it’s very easy to take a traditional IVR and just rip it out and then
260
00:22:35,129 –> 00:22:40,889
do more kind of intelligent routeing. So we basically have a more natural conversation, things
261
00:22:40,890 –> 00:22:46,169
like authentication of the user. It’s also very useful using AI. Like we have customers like
262
00:22:46,170 –> 00:22:51,969
spending maybe up to one minute per conversation just to attempt to get the user. And we start adding
263
00:22:52,129 –> 00:22:58,369
up some of these things, it obviously creates a very healthy business case. But again, the cool,
264
00:22:58,409 –> 00:23:03,409
the cool stuff you can do is obviously, then, when you can do end-to-end automation, to start a
265
00:23:03,410 –> 00:23:07,689
conversation, you attempt to get the user and then you kind of follow out the whole conversation on
266
00:23:07,689 –> 00:23:14,009
voice. So, yeah, I think we probably predicted that it’s going to be
267
00:23:14,569 –> 00:23:20,489
2025, 2024 is going to be kind of the year of voice. But I think 2026 is definitely where
268
00:23:20,489 –> 00:23:27,269
I’ve really seen a significant uptick for sure. Interesting. What do you think is
269
00:23:27,269 –> 00:23:31,789
behind that, then? Do you think it’s kind of like the models of all of a sudden got better? Is it
270
00:23:31,790 –> 00:23:36,109
that awareness is now got to such a degree that people are now kind of getting comfortable
271
00:23:36,109 –> 00:23:42,789
with it? Like what? What do you think is the catalyst behind the growth in voice AI? So I
272
00:23:42,790 –> 00:23:47,390
think there’s a few things I think. First of all, there is for sure some new vendors coming into
273
00:23:47,390 –> 00:23:53,549
the market. We have a lot of local vendors. We have Speechmatics, we have 11 Labs. We have a
274
00:23:53,550 –> 00:23:59,629
lot of companies outside the kind of the usual Microsoft and Google, which has invested a lot in
275
00:23:59,630 –> 00:24:04,829
this technology. And they have all their kind of like pros and cons in terms of strengths, in terms
276
00:24:04,830 –> 00:24:10,589
of industries and language and so on. They have definitely created a lot of innovation in this
277
00:24:10,590 –> 00:24:17,269
space. That’s for sure. But then also I also think it’s about kind
278
00:24:17,270 –> 00:24:24,240
of the maturity around how people involved AI within the organisation
279
00:24:24,240 –> 00:24:30,040
has also changed a lot. Remember when we started like the Boost? Like ten years ago? Like we sold
280
00:24:30,040 –> 00:24:35,239
the chatbot that was ported on to your contact centre. We didn’t kind of really get into kind of
281
00:24:35,280 –> 00:24:39,319
the wider ecosystem at the start because again, we are kind of a new channel. People were like super excited
282
00:24:39,319 –> 00:24:44,799
about it. But then like as we have gone, gradually, we’ve been more and more a key part of the whole
283
00:24:44,800 –> 00:24:51,679
contact centre CX. But now people again have done chat. They have done messaging. They have really
284
00:24:51,680 –> 00:24:56,920
been able to utilise AI in customer service, especially, but also internally. So now they’re
285
00:24:56,920 –> 00:25:02,920
obviously looking for something more to automate and also make better. And what is interesting
286
00:25:02,920 –> 00:25:07,280
is also to see like it’s not all about automation, but also being able to create a better experience
287
00:25:07,280 –> 00:25:13,959
because mostly in the in most contact centre across the globe there is the skew. Um,
288
00:25:14,240 –> 00:25:19,208
so just being able to have something which is more scalable, it’s obviously super useful. So
289
00:25:19,209 –> 00:25:25,049
that’s definitely some of the things we have seen, being a key part of kind of the wave
290
00:25:25,050 –> 00:25:29,969
of new voice projects. So you will see this now, but also in the coming months and year that
291
00:25:29,969 –> 00:25:35,410
there’s going to be a lot of new deployments coming out. Good. That’s really good. I
292
00:25:35,410 –> 00:25:40,489
remember a while ago there was a couple of vendors. I won’t name names, but they were kind of
293
00:25:40,530 –> 00:25:47,449
like um, in order to sort of like try and really sort of get voice AI off the ground their kind
294
00:25:47,449 –> 00:25:54,369
of position was voice enable your chatbot, just stick, kind of like speech to text and text to
295
00:25:54,369 –> 00:25:59,528
speech on top of your chatbot. Stick it on a phone line and you’ve got yourself a voice bot.
296
00:25:59,529 –> 00:26:05,569
And at the time, you know, if you’re working with NLU, it wasn’t as simple as that because people
297
00:26:05,569 –> 00:26:10,810
spoke differently to how they typed. And so the model wouldn’t always be the same model. And then
298
00:26:10,810 –> 00:26:16,500
also the responses needed to be different because people’s attention span are a lot shorter on
299
00:26:16,500 –> 00:26:20,579
voice than they are over chat. On chat, you can have UI affordances like tables and images and
300
00:26:20,579 –> 00:26:25,379
stuff like that. In voice you can’t. So it was always quite a challenge to just take an existing
301
00:26:25,380 –> 00:26:31,660
chatbot and simply kind of voiceify it, so to speak, with large language models now, and the way
302
00:26:31,660 –> 00:26:37,179
that kind of like these applications are built, is that becoming more feasible in your estimation, or
303
00:26:37,180 –> 00:26:41,339
is it still best practice to create a separate instance and a separate agent for the voice
304
00:26:41,340 –> 00:26:46,739
channel? I would definitely say that it’s become much easier. I think just in general, building this
305
00:26:46,739 –> 00:26:52,019
agent, even on messaging, it’s so much easier now when you have the help of the different LLM
306
00:26:52,020 –> 00:26:56,939
engines. I think when it comes to voice and chat, as you said, like 2 very different channels.
307
00:26:56,939 –> 00:27:03,338
So they have significant differences. I also think when it comes to voice,
308
00:27:03,939 –> 00:27:10,699
it’s like you, people get very hung up on details, on voice. I think if you get in
309
00:27:10,699 –> 00:27:15,869
response on chat, it’s kind of very much like either you got what you wanted. It’s binary. You’re
310
00:27:15,870 –> 00:27:20,550
happy or not. I think in voice, you kind of look. You can look for the latency. How quick it is. You
311
00:27:20,550 –> 00:27:25,670
look at the voice, how you pronounce certain things. And if something is off, you will
312
00:27:25,670 –> 00:27:30,550
immediately notice it. So people, I think in general, are very picky when it comes to voice,
313
00:27:30,550 –> 00:27:37,389
which is which is obviously, which also makes it a good reason to
314
00:27:37,430 –> 00:27:42,189
do some more work on voice, for sure. So I think even if you can reuse a lot of the stuff you do
315
00:27:42,189 –> 00:27:46,669
on kind of messaging and chat into voice more than before. I would definitely say it’s
316
00:27:46,670 –> 00:27:51,629
definitely beneficial to think about how you actually create that experience. Maybe you need to
317
00:27:51,630 –> 00:27:57,349
add some certain words, maybe how it pronounces things that could obviously be a ton of
318
00:27:57,349 –> 00:28:01,429
different dialects when you remember when you write. Especially in most countries, you kind of
319
00:28:01,430 –> 00:28:07,670
write in a similar way. But when we talk, we talk very differently. So that also kind of comes with
320
00:28:07,670 –> 00:28:11,349
some complication that you actually need to be able to kind of then fine-tune it and train it
321
00:28:11,350 –> 00:28:18,319
for that certain language, and that dialect and so on. So yes, in short, it’s become definitely
322
00:28:18,319 –> 00:28:24,319
much easier. But we definitely recommend spending more time building out the voice
323
00:28:24,319 –> 00:28:29,119
journeys for sure. Yeah, I think you’re right. That’s a good observation. That which is that
324
00:28:29,120 –> 00:28:34,759
people are a lot more picky with voice, because I am very pedantic when it comes to
325
00:28:34,800 –> 00:28:40,759
voices like, you know. And those 2 things don’t really go together in the LLM world. Like, you can’t be
326
00:28:40,759 –> 00:28:47,239
so hung up over detail because you ultimately don’t have total control over it. So it’s like.
327
00:28:47,439 –> 00:28:52,999
getting the voice right is 1 thing. So that’s a really important step. Getting the tonality in
328
00:28:52,999 –> 00:28:59,799
terms of how it pronounces things is also a step, but also the specific words it says. And
329
00:28:59,799 –> 00:29:05,199
that’s the challenge. It’s that sometimes when it’s just generating text, you might have a call and
330
00:29:05,199 –> 00:29:09,719
it’s perfect. It’s just got everything right. But then the next call, it might not say the same
331
00:29:09,720 –> 00:29:13,689
sentence in the same way. It might not feel, it’s a feeling, isn’t it? It doesn’t quite feel the same.
332
00:29:13,730 –> 00:29:17,930
So I don’t know if you’ve got any sort of like, I know, observations or recommendations in terms of
333
00:29:17,970 –> 00:29:24,489
like how do you make the voice UI a more consistent in a probabilistic kind of
334
00:29:24,529 –> 00:29:31,010
landscape? Yeah. And I think the observation, like we also see the same thing, that like, depending on
335
00:29:31,010 –> 00:29:36,448
what type of model you use? Again, with voice, there is so much more components involved in that
336
00:29:36,449 –> 00:29:40,929
conversation. Like there’s a lot of steps you go through. So you can also imagine when you
337
00:29:41,010 –> 00:29:46,929
actually do end-to-end, like you need to go into the CRM system, you need to authenticate the user.
338
00:29:46,929 –> 00:29:50,929
You then need to update some information. You need to pull out some information present like
339
00:29:51,250 –> 00:29:56,409
things which obviously in a chat, the messaging can also be complicated. But again, when it comes
340
00:29:56,409 –> 00:30:00,688
to voice, it becomes even more complicated because then you don’t really have time to pull out the
341
00:30:00,689 –> 00:30:04,729
information. You don’t really have time to say, hey, hold on, we are fetching some information, even if
342
00:30:04,729 –> 00:30:11,659
the human also takes some time, though, to fetch the information. But I think one way
343
00:30:11,659 –> 00:30:16,579
you can definitely tackle a lot of that. Those challenges is basically to be able to test things.
344
00:30:16,620 –> 00:30:21,699
So in Boost, we have kind of very extensive testing tools, meaning that before we go into
345
00:30:21,699 –> 00:30:25,859
production, you’re actually able to test out all the different scenarios, looking at, okay, if the
346
00:30:25,859 –> 00:30:32,699
user was angry, if different sentiments, maybe there’s some dialects, so you can really
347
00:30:32,740 –> 00:30:37,300
kind of get a feel of the whole experience, but also do that multiple times because as you said,
348
00:30:37,340 –> 00:30:43,019
like maybe it works perfectly one time, but then obviously the second time is a bit different. So
349
00:30:43,019 –> 00:30:47,619
being able to kind of run ideally hundreds or thousands of different simulations, and then you
350
00:30:47,619 –> 00:30:51,979
can see if there are things you need to do changes on. I think it’s going to be difficult to
351
00:30:51,979 –> 00:30:58,099
have like the exact same result every time. So I think I also hope the customer also can be a bit
352
00:30:58,100 –> 00:31:03,259
less picky sometimes. And also think about like, okay, if I get the support I need, this is a good
353
00:31:03,259 –> 00:31:08,829
experience overall. I’m okay with that. And obviously, things will also
354
00:31:08,829 –> 00:31:14,749
become better over time, for sure. There’s no doubt about that. Yeah. I mean, compared to NLU, you know,
355
00:31:14,789 –> 00:31:18,708
what would you rather have something that can understand everything and respond to everything,
356
00:31:18,709 –> 00:31:23,149
even if it can’t satisfy the need? At least it can respond to you and have an actual conversation.
357
00:31:23,149 –> 00:31:27,709
But it might not get things exactly right in terms of how it pronounces everything, or
358
00:31:27,710 –> 00:31:33,389
something that doesn’t understand 9 out of 10, or like or 3 out of 10 things that you say,
359
00:31:33,430 –> 00:31:37,790
you know, you would you would rather sacrifice a bit of control for the for the whole conversation
360
00:31:37,989 –> 00:31:43,469
rather than the rigidity and not understanding everything from the NLU side, you know? Yeah, yeah.
361
00:31:43,509 –> 00:31:46,869
But the beauty in the platform, you can actually go in and find, you know, a lot of these things. And
362
00:31:46,870 –> 00:31:50,509
that’s also the kind of thing where you will spend a bit more extra time than you would doing
363
00:31:50,550 –> 00:31:54,790
kind of more chat messaging that you can really fine-tune those things. And I think it’s going to
364
00:31:54,790 –> 00:31:59,150
be a journey for most companies. I think for me, like the recommendation is always like, how can we
365
00:31:59,150 –> 00:32:05,709
get started? You sure you can get started? Kind of late in one of the
366
00:32:05,770 –> 00:32:10,249
menus. Find one specific use case and do that really well. Or you can take the route where you
367
00:32:10,250 –> 00:32:14,688
can basically do routeing and start with kind of that the topics. And but then you also always
368
00:32:14,689 –> 00:32:18,969
route to human in the first instance. That’s also a good way of kind of building up the model you
369
00:32:18,970 –> 00:32:24,609
need. And then it’s going to be kind of like make sure that you do that necessary kind of
370
00:32:24,609 –> 00:32:29,248
evolvement of the solution, looking at conversations, having the AI to test it out,
371
00:32:29,290 –> 00:32:34,809
improve it, and also kind of all. So the sum of all these kind of details will definitely make sure
372
00:32:34,810 –> 00:32:39,929
that you have a good experience for sure. And how do you approach the testing of the voice channel?
373
00:32:40,050 –> 00:32:45,969
Do you care about testing the speech recognition, the speech generation, or do you kind of bypass
374
00:32:45,970 –> 00:32:50,129
the telephony and the voice stuff and you just test the model text in, text out? How do you
375
00:32:50,129 –> 00:32:55,089
approach the testing? So we do a lot of testing on our own just to make sure that we have the best
376
00:32:55,089 –> 00:32:59,608
practices and also the best models we can recommend for customers. So typically in a project,
377
00:32:59,609 –> 00:33:06,379
you will. especially when you go into some of the with a non-English countries. You definitely need
378
00:33:06,379 –> 00:33:10,938
to see who is the best vendor. Sometimes there’s some local vendor, sometimes there’s some of the
379
00:33:10,939 –> 00:33:14,939
bigger ones. And we also need to see how much you can actually customise that model for the
380
00:33:14,940 –> 00:33:20,260
customer. So sometimes Microsoft can offer something which is not like amazing out of the
381
00:33:20,260 –> 00:33:24,819
gate, but then you also have the possibility to fine-tune it, and then you can obviously get it to
382
00:33:24,819 –> 00:33:31,780
where you need to get it. But we also do, we actually create voice agents. So the
383
00:33:31,780 –> 00:33:37,300
voice agent basically then calls the voice agent you are going to have for your customers.
384
00:33:37,460 –> 00:33:43,939
Right. So it’s basically simulating kind of a real conversation. And that’s kind of gives you a very
385
00:33:43,939 –> 00:33:48,979
good idea on like different profile of the customer. You can look at different age, how they
386
00:33:48,980 –> 00:33:53,579
talk and so on. And that would be basically something we just call in directly to the
387
00:33:53,580 –> 00:33:59,619
to the voice agent. And then that will definitely generate results we can have a look at
388
00:33:59,619 –> 00:34:06,348
afterwards. Nice. And what is it that you’re looking for when you
389
00:34:06,349 –> 00:34:12,589
make the voice call? So like when you spin up a voice agent, and you have it make a call to the,
390
00:34:12,629 –> 00:34:17,189
to the one that you’ve just built and you have that simulated call. What’s the difference between
391
00:34:17,189 –> 00:34:22,748
doing that versus kind of skipping the voice channel entirely and just having a text agent
392
00:34:22,790 –> 00:34:28,228
have a text conversation with the back end of your voice agent, if that makes sense. Yeah. So I
393
00:34:28,229 –> 00:34:34,309
think it’s definitely things like latency. It’s very important in that sense. That
394
00:34:34,310 –> 00:34:39,909
could also be just in terms of like there’s also differences in terms of how we actually
395
00:34:39,909 –> 00:34:44,789
understand kind of the intent of the user. Obviously, in voice, the conversation will also
396
00:34:44,790 –> 00:34:49,550
have more turns, I think on messaging is very much you say something and then you predict something,
397
00:34:49,550 –> 00:34:54,509
get something back obviously involves the conversation. It’s more fluid in the sense that it
398
00:34:54,510 –> 00:34:59,269
continues. And then you have some, um, yes, please repeat it and so on. So I think just the
399
00:34:59,270 –> 00:35:04,319
conversation flow in itself is too small to analyse in that sense than on chatter messaging,
400
00:35:04,320 –> 00:35:09,719
which I feel is very often very much more direct, which obviously voice where you actually engage
401
00:35:09,719 –> 00:35:14,638
the customer are a bit different. And then you also need to be able to tackle all the, let’s
402
00:35:14,639 –> 00:35:20,159
say, the small talks and the things you also do outside, just asking for that specific question.
403
00:35:20,199 –> 00:35:25,760
That’s also key part of the experience. Yeah. What are some of the, so for someone who is you
404
00:35:25,760 –> 00:35:30,079
know, they’ve got an AI agent, it’s a chat-based kind of thing. They’re looking to kind of move
405
00:35:30,080 –> 00:35:35,680
into voice, maybe, you know, it’s becoming a bit clearer, they might not be just lift and shift
406
00:35:35,840 –> 00:35:41,360
things. What are some of the key things that need to change? You mentioned latency on voice. That’s a
407
00:35:41,360 –> 00:35:45,958
big consideration. What are the other sort of like real key considerations that you need to take
408
00:35:45,959 –> 00:35:52,959
into account to get a voice AI right? I think testing is very important because
409
00:35:53,280 –> 00:35:59,040
what we learned is also like a conversation which work perfectly on chatter messaging doesn’t
410
00:35:59,050 –> 00:36:05,968
necessarily work perfectly on voice because, again, there could be things like if you want
411
00:36:05,969 –> 00:36:10,648
to extract information from the user and then that information needs to be like, you need to
412
00:36:10,649 –> 00:36:17,648
write in, for example, numbers or maybe an address, for example. On some cases, I say the
413
00:36:17,649 –> 00:36:22,249
voice is very good at extracting a lot of that information, but in some cases it’s also hard for
414
00:36:22,250 –> 00:36:26,249
the customer to pronounce some of these, like some of these, like if you go to Finland, for example,
415
00:36:26,250 –> 00:36:32,649
some of these kind of like, these sweet names of like 30 characters. So it’s not that easy for the
416
00:36:32,649 –> 00:36:35,929
voice to understand. So in those cases, you actually need to think about the whole experience
417
00:36:35,929 –> 00:36:41,089
though. Like, does it make sense to have it on voice, everything, or can we think a bit more
418
00:36:41,090 –> 00:36:45,370
multimodality. So maybe we can actually kind of combine those so we can actually have the
419
00:36:45,370 –> 00:36:50,928
conversation on, on, on a voice. But then we can also use a messaging or chat interface to kind of
420
00:36:50,970 –> 00:36:56,128
put in things like address things we know is going to be super hard for the voice to
421
00:36:56,129 –> 00:37:00,459
understand. So I still think there are certain things where I don’t think it makes sense
422
00:37:00,460 –> 00:37:05,259
necessary to do it in voice. We can do most of the things, but I think there is some things which is
423
00:37:05,259 –> 00:37:09,819
a bit harder to do. So I think just looking at the customer experience, how that would look like
424
00:37:09,940 –> 00:37:14,699
because as you said, like you’re not able to give links, you’re not able to give forms. You’re not
425
00:37:14,699 –> 00:37:20,099
able to give images and things like that, which is still very much used in the chat messaging world.
426
00:37:20,100 –> 00:37:25,299
And then you need to translate that to voice. So basically, the content will be quite different
427
00:37:25,300 –> 00:37:30,299
than in some cases. But again, if it’s just an answer, then you can obviously use the text
428
00:37:30,340 –> 00:37:37,019
pretty much. That’s interesting. So, you know, like how, you know, the
429
00:37:37,020 –> 00:37:43,059
routeing example is a great example because that’s a kind of fairly, shallow use case in
430
00:37:43,060 –> 00:37:47,459
terms of like, you know, that conversation to get to the right agent might be like a number of
431
00:37:47,459 –> 00:37:53,579
turns, but it’s not going to go on forever. There might be there might be cases where, yeah,
432
00:37:53,620 –> 00:38:00,069
maybe the UI is important. So you might want to kind of devolve to a different channel if you
433
00:38:00,070 –> 00:38:04,189
need to submit documents and stuff like that. That seemingly sounds like it might be a problem on
434
00:38:04,189 –> 00:38:11,070
voice. I’m wondering whether kind of like given the power of LLMS now, is it feasible
435
00:38:11,070 –> 00:38:17,909
to, you know, what am I trying to say? I suppose I’m trying to get to like, how complicated a use
436
00:38:17,909 –> 00:38:24,269
case can we really tackle on voice? In terms, you know, I was looking recently at like government
437
00:38:24,269 –> 00:38:29,869
based use cases and there’s like, you know, blue badge applications and bus pass application and
438
00:38:29,870 –> 00:38:34,589
driving license renewals and this and yeah, fair enough. Those things might require you to submit
439
00:38:34,590 –> 00:38:39,029
documents, but really they’re kind of just forms. And there’s loads of things like income and
440
00:38:39,030 –> 00:38:45,550
expenditure forms in like debt collection and like insurance quotation forms or claim forms, all
441
00:38:45,550 –> 00:38:50,789
those things that are just forms that were a nightmare to do with NLU because it’s just too
442
00:38:50,790 –> 00:38:53,789
hard to class it, even though you’re just essentially slot-filling in some cases, it’s very
443
00:38:53,790 –> 00:38:58,919
difficult to sort of do LLMs with the power of language capabilities that they have now. Where do
444
00:38:58,919 –> 00:39:04,600
you stand in terms of like, how complicated and sophisticated can we get in terms of end-to-end
445
00:39:04,639 –> 00:39:10,719
automation on the voice channel, do you think? Yeah. So I think, I think there’s, there’s a number we
446
00:39:10,759 –> 00:39:17,039
kind of monitor with our customers, which I again, I think is very which give a good answer on that
447
00:39:17,040 –> 00:39:22,879
because again, when we started a lot of the voice project, if you look away from kind of, I think
448
00:39:22,919 –> 00:39:29,079
routeing, but also just looking at the whole use case, like if we do tackle everything a
449
00:39:29,080 –> 00:39:35,240
customer can do on voice, we typically saw maybe 20% to 30% automation. So when we talk about
450
00:39:35,240 –> 00:39:40,079
automation, it’s basically containment, or you’re actually able to resolve that conversation or
451
00:39:40,080 –> 00:39:45,520
that question for the customer. So that’s kind of how we define automation here. So that’s of course
452
00:39:45,520 –> 00:39:51,519
we’re not going into the contact centre. What we have seen in the last 6 months is that
453
00:39:51,520 –> 00:39:58,089
20% or 30% has become now 60% and 70%. And we actually have some customers being 75%
454
00:39:58,129 –> 00:40:04,089
automation. So you can imagine 75% automation of all your incoming
455
00:40:04,330 –> 00:40:11,329
phones. That’s a pretty big number. And yes, there’s for
456
00:40:11,330 –> 00:40:15,849
sure that’s going to be some more calls coming in. So sometimes it’s difficult to have a 75%
457
00:40:15,850 –> 00:40:21,249
reduction in incoming because some of the more complex questions will get through.
458
00:40:21,250 –> 00:40:25,610
So those will take more time. I think also when you have a channel which is more available and
459
00:40:25,610 –> 00:40:30,889
driven by AI, you will generally get more volume. So the volume will naturally just grow a bit,
460
00:40:30,889 –> 00:40:35,689
which is not necessarily a bad thing. But when you’re looking at 60 to 70% automation in voice, I
461
00:40:35,689 –> 00:40:39,810
think it definitely tells a pretty interesting picture where you can
462
00:40:39,810 –> 00:40:46,489
actually go, and we have some customers where we basically then decide, okay, where do we want to be
463
00:40:46,490 –> 00:40:51,139
in 3 years time? So in 3 years time, we want to make sure that we can reduce cost of
464
00:40:51,139 –> 00:40:55,899
customer service by 50%. We also want to increase the NPS score by 10 points, for example, at the
465
00:40:55,899 –> 00:41:02,779
same time. So basically better experience lower, waiting time, creating more revenue. Um,
466
00:41:02,979 –> 00:41:09,540
but then also obviously reducing the cost. So that’s possible today. So I think. Yeah, where
467
00:41:09,540 –> 00:41:14,579
are we going to end up with that number? I don’t think we’re going to get to 100% any day soon, but
468
00:41:14,580 –> 00:41:19,419
I definitely expect that number to continue to climb as we go forward. That’s wicked. I mean,
469
00:41:19,459 –> 00:41:26,459
75% is huge. You know, even if it’s 75% of one use case within a suite of use cases,
470
00:41:26,460 –> 00:41:31,419
that at least that problem becomes very much a solved problem, which is brilliant.
471
00:41:31,459 –> 00:41:38,300
Yeah. What’s your observations now on terms of, like, in the NLU days? You know, it was
472
00:41:38,300 –> 00:41:44,378
quite challenging to build these systems because first of all, you had to have someone who is more
473
00:41:44,379 –> 00:41:48,060
akin to a machine learning engineer, not necessarily to that level of depth, but someone
474
00:41:48,060 –> 00:41:52,919
who knows how to train a model, getting a good type of training data. Making sure that it’s all
475
00:41:52,959 –> 00:41:57,559
kind of like decent data. Training the model. Testing the model. All that kind of stuff. Then you
476
00:41:57,560 –> 00:42:02,039
have to have a conversation designer who could plan out and map out the conversation and all
477
00:42:02,040 –> 00:42:08,199
that kind of stuff. And so you had to have, you know, skill sets that were very hard to come by
478
00:42:08,600 –> 00:42:14,039
because conversation design is a very unique practice. NLU modelling is a very unique practice.
479
00:42:14,039 –> 00:42:18,839
And putting it all together with your kind of customer experience lens and all that kind of
480
00:42:18,879 –> 00:42:24,239
stuff. It’s all quite a kind of unique sort of skill set. LLMs have opened up access to this
481
00:42:24,240 –> 00:42:29,919
technology far beyond what NLU ever did. Anyone now can access an API, and even if you don’t work
482
00:42:29,919 –> 00:42:34,999
with APIs, you can work with Claude Code and it can build your stuff these days. So the access to the
483
00:42:35,000 –> 00:42:41,279
technology is far greater. But it seems to me at least that still, it’s not just a case of grabbing
484
00:42:41,280 –> 00:42:44,679
an LLM and building the chat bot to build something that’s actually robust. And as you said
485
00:42:44,679 –> 00:42:50,769
at the beginning, something that’s like regulatory sound, and provides a good
486
00:42:50,770 –> 00:42:56,329
experience and has low latency and is consistent over time. That’s still not something that is a
487
00:42:56,330 –> 00:43:02,330
trivial kind of activity. So I wonder what your kind of thoughts are and what you’ve observed in
488
00:43:02,330 –> 00:43:08,929
your clients in terms of, like, what is the maturity level of organisations right now? Is it
489
00:43:08,969 –> 00:43:13,049
heading in the right direction in terms of businesses are beginning to build the right
490
00:43:13,050 –> 00:43:16,490
capabilities to be able to do this stuff themselves? Are you still having to do a lot of
491
00:43:16,490 –> 00:43:22,490
this work yourself? Like, how is the maturity of organisations now in terms of their AI competency?
492
00:43:23,370 –> 00:43:27,888
So it’s definitely increasing a lot. So I think in general we’re seeing much more educated people,
493
00:43:27,929 –> 00:43:32,168
educated buyers, they know they definitely know the stuff. And people are setting up the AI
494
00:43:32,209 –> 00:43:38,089
hubs. They’re putting up a lot of initiatives. AI is obviously a board discussion. CEOs want to do
495
00:43:38,129 –> 00:43:44,888
AI. So that’s definitely a very important. In the sense of kind of the capabilities you
496
00:43:44,889 –> 00:43:49,259
need to have to actually kind of build out a fantastic experience. I think you’re definitely
497
00:43:49,260 –> 00:43:53,939
spot on that you still need, you need someone really understanding the CX role of it. And I think
498
00:43:53,939 –> 00:43:57,739
this is also maybe where I see a lot of organisation, some organisation at least go into
499
00:43:57,740 –> 00:44:04,019
kind of be a trap that they it’s that we all know it’s super easy to set up an LLM bot. So you will
500
00:44:04,020 –> 00:44:08,539
sometimes have people from the IT teams just build up something. Hey we just scrape the web
501
00:44:08,539 –> 00:44:13,459
page, scrape a knowledge base, we build a RAG, we launch it and we’re good to go. We don’t need to
502
00:44:13,459 –> 00:44:17,739
do anything more. But as we know, if you really want to kind of push the envelope to what’s
503
00:44:17,780 –> 00:44:22,820
what’s possible in this technology, you really, really need to have the business the CX, the
504
00:44:22,820 –> 00:44:29,418
contact centre people in. And when we started a company, we focussed, we said to
505
00:44:29,419 –> 00:44:33,339
ourselves, we don’t want our platform to be used by technical people. We want people to be
506
00:44:33,340 –> 00:44:37,939
retrained from customer service. So that was a bit of a unique take on that. And that’s something we also
507
00:44:37,979 –> 00:44:44,909
kind of continued to do to this day. But it also adds
508
00:44:44,909 –> 00:44:50,749
to your point, like evolved a bit in terms of what is the skill set of this of this person?
509
00:44:50,750 –> 00:44:55,350
Because before you need to build training data, a lot of tedious, boring work you don’t need to do
510
00:44:55,350 –> 00:45:01,350
anymore, like the generative AI really helps you. So I think now it’s more how I look at our
511
00:45:01,350 –> 00:45:07,149
solution is you basically you kind of log in, the AI will proactively kind of nudge you. Okay, there
512
00:45:07,149 –> 00:45:11,749
are certain things you need to look at here. Here is things I tested. Okay. How can we make sure we
513
00:45:11,750 –> 00:45:17,469
can improve that for next time. So you as a human, need to understand the conversation. You need
514
00:45:17,469 –> 00:45:23,429
to understand the business, because you also need to be able to kind of figure out how do I pull
515
00:45:23,429 –> 00:45:28,149
that information? What is the architecture? Because you can imagine how we are using information from
516
00:45:28,149 –> 00:45:32,669
the whole organisation. So we are pulling information from the CRM system. We’re pulling
517
00:45:32,669 –> 00:45:39,030
from a SharePoint or other knowledge sources. So there’s also some architecture in terms of how
518
00:45:39,030 –> 00:45:45,719
you build that out. How do you fetch the information. But then again, how do I create
519
00:45:45,720 –> 00:45:49,680
that amazing experience? How do we make sure they like the customer? Don’t ask them too many
520
00:45:49,680 –> 00:45:55,398
questions. How can I make sure that I populate the conversation with as much information as possible
521
00:45:55,399 –> 00:46:01,280
to use about the user as possible? So yeah, I think in general the role has changed a lot from just
522
00:46:01,280 –> 00:46:06,719
doing all the boring, tedious things to build out, to know, , to now having less people on the
523
00:46:06,720 –> 00:46:12,359
team for sure. I think you can do more now with less team, but again, focus on the experience part.
524
00:46:12,399 –> 00:46:16,878
I think that’s, that’s where you’re going to win customers. Hmm. Interesting. Where do you
525
00:46:16,879 –> 00:46:23,399
stand in terms of kind of like, in terms of where this is going and where this is heading?
526
00:46:24,760 –> 00:46:30,919
For me, it’s always seemed as though you’re going to need those skills and those activities that
527
00:46:30,919 –> 00:46:35,759
you just highlighted. It seems to me at least, that businesses are going to always need that, because
528
00:46:35,760 –> 00:46:39,360
they’re going to be building new use cases that need to be optimised in existing use cases as a
529
00:46:39,360 –> 00:46:44,369
whole kind of raft of things you need to do to make really great experiences at scale. But at the
530
00:46:44,370 –> 00:46:51,289
same time, you’ve got other companies that their kind of market position is,
531
00:46:51,649 –> 00:46:57,449
you know, our goal really is to basically take as much off your plate as possible. And our
532
00:46:57,449 –> 00:47:02,169
technology is going to be very opinionated in terms of what it does. And all you need to do. Mr
533
00:47:02,170 –> 00:47:06,409
and Mrs customer is just give us your APIs and give us your data. And it’s as good as a
534
00:47:06,409 –> 00:47:10,649
solved problem. So you kind of have, like, what I would say a Boost would be, would be kind of like
535
00:47:10,649 –> 00:47:16,049
that platform company orchestration company, where essentially, you can build your stuff and you can
536
00:47:16,050 –> 00:47:20,330
have total control and you can build what you want, whereas you have other companies that have
537
00:47:20,330 –> 00:47:24,129
more kind of, you know, they’ve specifically taken up that position in the market, which is this
538
00:47:24,129 –> 00:47:27,929
stuff is a lot easier and more trivial than you might think, just gives you a data and
539
00:47:27,929 –> 00:47:34,409
gives you APIs. Do you see that? I don’t know. Does that have merit at all? For me, this is a bit of a
540
00:47:34,409 –> 00:47:38,489
loaded question, but for me that seems a bit shortsighted and a little bit like you’re only
541
00:47:38,490 –> 00:47:42,259
going to get so far with it, but obviously you’re a lot closer to the technology than I am. So I
542
00:47:42,260 –> 00:47:48,779
don’t know if what your what your position is on the future of building agents. I think if your
543
00:47:48,780 –> 00:47:53,060
focus is only the bottom line, you just want to do things as cheap as possible. Sure, that’s
544
00:47:53,060 –> 00:47:57,179
definitely a viable strategy, I don’t think, but I don’t think that’s kind of where you’re going to
545
00:47:57,179 –> 00:48:00,979
win customers. That’s not where you’re going to win market share. But I think like the customer
546
00:48:00,979 –> 00:48:04,939
experience, it’s extremely important. Like every brand we interact with, there is an element of
547
00:48:04,939 –> 00:48:08,819
customer experience through it. That is how we feel and touch that brand is extremely
548
00:48:08,820 –> 00:48:14,299
important. And I think especially now in the days we are right now, it’s going to be even more
549
00:48:14,299 –> 00:48:19,459
important. And I think we’re already seeing now a lot of like AI slop going out there, like people
550
00:48:19,459 –> 00:48:23,499
producing slides or whatever, like writing stuff on LinkedIn. Like there’s a lot of things which
551
00:48:23,540 –> 00:48:29,820
you can definitely see is sensing coming from an AI, which I don’t think like, I can almost see it
552
00:48:29,860 –> 00:48:34,178
immediately when I see a presentation. Okay. This is created by a Claude. Okay, fine. It looks great.
553
00:48:34,220 –> 00:48:39,429
The content is pretty good, but it doesn’t really stand out. So I think for me, there’s going to be a
554
00:48:39,429 –> 00:48:46,229
merit that it’s going to be companies which use humans to create, think creatively about, like
555
00:48:46,270 –> 00:48:50,869
how we can build a brand and the feel of it and that it’s still going to have value. I don’t think
556
00:48:50,870 –> 00:48:56,909
AI is going to be able to think out of the box in that same way for a certain brand. So I
557
00:48:56,909 –> 00:49:02,908
think that touch to it, it’s going to be crucial. Yeah, I agree definitely. I think it’s
558
00:49:02,909 –> 00:49:09,549
interesting how like nostalgia works. I think the timeline of nostalgia is getting shorter and
559
00:49:09,550 –> 00:49:15,550
shorter and shorter, which is that like, you know, when Covid came around, I know this is a common
560
00:49:15,550 –> 00:49:18,509
reference. When Covid came around, everyone was locked down. Everyone’s in the houses, all that
561
00:49:18,510 –> 00:49:23,949
kind of stuff. And then, as soon as it kind of finished, it was almost like nostalgic to kind of
562
00:49:23,989 –> 00:49:27,830
create some sort of event where everyone gets back together again and is in person type of
563
00:49:27,830 –> 00:49:32,949
thing, and you’re already seeing a little bit of that in the market, which is that some companies,
564
00:49:32,949 –> 00:49:36,839
like I think it was Clan of fairly recently was like, we’re not going to stop you speaking to our
565
00:49:36,840 –> 00:49:39,999
people and trying to use like, the fact that you’re not always going to be at a talk to a
566
00:49:39,999 –> 00:49:44,639
person as like the differentiator. For me, I think it’s a little bit too early for that to be a
567
00:49:44,639 –> 00:49:49,079
differentiator. I think that we’re still well and truly in an automation world and an automation
568
00:49:49,080 –> 00:49:54,399
phase where more and more things are going to become automated. But I think I do think that in
569
00:49:54,399 –> 00:49:59,120
the same way as you’re saying that, I noticed I stop online all the time and so do I. I notice it
570
00:49:59,120 –> 00:50:02,759
a million miles. That’s why I’ve stopped using AI to write. I don’t write with it anymore because it
571
00:50:02,759 –> 00:50:09,159
just does my head in. But like, in the same way I can see a world where the
572
00:50:09,159 –> 00:50:16,120
quest to automate ends up with actually very average experiences, average to poor
573
00:50:16,120 –> 00:50:21,398
experiences. And therefore the differentiator is not necessarily we’re going to put you through to
574
00:50:21,440 –> 00:50:24,878
people, which I think some people think that that might be the thing. I don’t think it is. I think
575
00:50:24,879 –> 00:50:31,079
the differentiator will actually be that this conversation was just like talking to a person, you
576
00:50:31,080 –> 00:50:35,259
know, so above average from the point of view of it doesn’t sound like a bot. It didn’t just kind
577
00:50:35,260 –> 00:50:39,699
of give me what I wanted and got on with the day. It was actually a pleasure to interact with, and I
578
00:50:39,699 –> 00:50:44,379
think that’s kind of the sweet spot to try and get to, isn’t it? Yeah, and I think it’s on some
579
00:50:44,379 –> 00:50:48,859
certain use cases at some times in life. You also want to talk to a human. You want to kind of get
580
00:50:48,860 –> 00:50:53,220
that personal experience. If you’re a bank customer and you want to get a mortgage, for
581
00:50:53,220 –> 00:50:58,099
example, like you’re buying a house, that’s a that’s a pretty big decision to make. I
582
00:50:58,220 –> 00:51:02,378
would definitely envision that a lot of people still want to go, maybe in physical offices and
583
00:51:02,379 –> 00:51:07,579
talk to the advisor, kind of get some advice and so on. So I think that’s not going to
584
00:51:07,580 –> 00:51:12,699
go away. And I think to your point, probably we want more of that in the future. But I
585
00:51:12,700 –> 00:51:17,379
still think there is going to be a race now in the market just to be able to be more having a
586
00:51:17,379 –> 00:51:22,100
bit more of a, say, AI native infrastructure to do those things. I think that’s obviously with the
587
00:51:22,100 –> 00:51:25,820
translation we’re going to see in the next couple of years is that people need to transition to a
588
00:51:25,820 –> 00:51:32,668
bit more efficient way of working, because remember when the volumes just continue just
589
00:51:32,669 –> 00:51:37,109
increasing you, kind of increasing the cost. And another thing we’re also seeing as a pretty big
590
00:51:37,110 –> 00:51:44,070
trend is now like, you know, with Open Claw, for like how the personal AI agent is going to
591
00:51:44,070 –> 00:51:48,029
be in the future. We don’t really know, but we’re already seeing that people are starting to use
592
00:51:48,030 –> 00:51:53,989
those agents to ask the bank for questions like maybe put in like a weekly schedule, where you
593
00:51:53,989 –> 00:52:00,189
going in? Ask 10 insurance company to get a better price, for example. So you will also the
594
00:52:00,189 –> 00:52:03,869
consumer will also have AI. So that also means that the contact centres then really need to invest
595
00:52:03,870 –> 00:52:09,349
in this because how can they be ready for the future when everyone has their own AI. And then it
596
00:52:09,350 –> 00:52:16,110
doesn’t really cost me any money to call my own voice bot to call your voice bot. So then you
597
00:52:16,110 –> 00:52:20,749
can imagine the volume is going to be massively increased in the future. So you also need to be
598
00:52:20,749 –> 00:52:26,949
able to be ready for that. Yeah, I think that might happen sooner than we think as well. Yeah. We
599
00:52:26,950 –> 00:52:33,239
did an event last, what was it a couple of weeks ago? And we had Nick and Sherry on your side
600
00:52:33,240 –> 00:52:39,080
were there actually. And what we did is before the event, we called everybody with a voice agent
601
00:52:39,080 –> 00:52:43,279
trained on my voice. And so it was just sounded like me making them a phone call saying, hey, I
602
00:52:43,320 –> 00:52:47,559
still coming. Have you got any dietary requirements? Stuff like that, which was brilliant. It was
603
00:52:47,560 –> 00:52:53,839
a great kind of, a great touch. But what was really interesting was that it
604
00:52:53,840 –> 00:52:59,359
was hitting call screening, you know, on people’s phones where it just doesn’t answer. And basically,
605
00:52:59,399 –> 00:53:03,839
like an AI will just answer it for you and it’ll just screen the call, who are you, what you’re
606
00:53:03,840 –> 00:53:09,999
calling for? And so our voice AI was hitting people’s call screening, telling them who it is
607
00:53:10,000 –> 00:53:13,959
and what I’m calling for. The call screening was going back to the person saying, hey, it’s Kane
608
00:53:13,960 –> 00:53:17,439
calling about the event. He wants to know if you’re going to turn up. That person then said,
609
00:53:17,440 –> 00:53:21,519
yeah, yeah, I’ll be there. Their kind of call screening agent came back into the call and says,
610
00:53:21,559 –> 00:53:24,919
yeah, yeah, Jim will be there. And I was like, okay, brilliant. I’ll mark him down. It’s coming. And that
611
00:53:24,919 –> 00:53:29,769
was it. So it was like two agents talking to each other and like so it’s happening now, you know?
612
00:53:29,850 –> 00:53:35,208
Yeah. That’s amazing. So there we go. Brilliant. Well, Henry, this has been an absolute pleasure. Really,
613
00:53:35,209 –> 00:53:41,128
really great talking to you. Where can people go and find out more? It’s Boost.AI. It’s
614
00:53:41,129 –> 00:53:45,528
pretty simple. Yep. I’ll also stick your LinkedIn down here if you’re happy for people. To connect
615
00:53:45,529 –> 00:53:50,209
to me on LinkedIn if they’re anything. Obviously, we’re always happy to support again. There’s a
616
00:53:50,210 –> 00:53:54,610
lot of, I think a lot of questions around, like, how do we do this? How do we get started? Like
617
00:53:54,649 –> 00:53:58,649
there’s millions of different types of questions you can ask. And we also know that a lot of
618
00:53:58,649 –> 00:54:04,249
organisations also have like different forces internally fighting for kind of like what AI
619
00:54:04,249 –> 00:54:08,929
initiative you want to do and so on. So we definitely spend a lot of time to kind of guide
620
00:54:08,929 –> 00:54:12,729
people also through that process. So you can make sure that you can show up to your boss,
621
00:54:12,730 –> 00:54:17,050
you can show up to your board and really have a successful project because not as we know, not
622
00:54:17,050 –> 00:54:22,009
every project is successful. Brilliant. Fantastic. Henry. Absolute pleasure. Thank you so
623
00:54:22,009 –> 00:54:25,889
much. Thank you again. Thank you. Very much. Thank you for having me. And thank you all for tuning in.
624
00:54:25,929 –> 00:54:28,329
We’ll see you again on the next one. Cheers. Bye.
