Latest / The Adults In The AI Room / Paul Lasserre on AI & Customer Experience - Episode #7

Transcript

Paul Lasserre on AI & Customer Experience - Episode #7 · The Adults In The AI Room

• 0:36
  SPEAKER_03: Good morning, good afternoon, good evening, depending upon your time zone.
• 0:40
  SPEAKER_03: Welcome to today's podcast.
• 0:42
  SPEAKER_03: My name is Seth Urb.
• 0:43
  SPEAKER_01: And I'm Chris Featherstone.
• 0:45
  SPEAKER_01: It's good to be with you again, Seth, as always.
• 0:48
  SPEAKER_03: Great to uh great to see you.
• 0:49
  SPEAKER_03: And uh I I understand you had a short week.
• 0:52
  SPEAKER_01: Yep, we got uh we got well-being days um at Salesforce where I work.
• 0:57
  SPEAKER_01: And so, you know, today's one of my mental health days that I get, you know, once a month.
• 1:01
  SPEAKER_01: So this is and you decided to decide to spend it with us.
• 1:05
  SPEAKER_01: Well, I mean, I it's it's it's you and it's also the guest, right?
• 1:10
  SPEAKER_01: It's it's this is the this is the perfect, yeah, the perfect segue into um, you know, I don't know if you want to say Serenity Now or Goose Fraba or whatever the the thing is that gets you, you know, nice calm and centered into, you know, right?
• 1:26
  SPEAKER_03: Yeah, I try to get my I try to get a little bit of my uh my mindfulness and meditation every day.
• 1:30
  SPEAKER_03: And I'll tell you, it does make a huge difference.
• 1:32
  SPEAKER_03: I'm a really big fan of that.
• 1:35
  SPEAKER_03: Um so before we start, I do want to thank our sponsors.
• 1:38
  SPEAKER_03: We have the Marketing AI Institute.
• 1:40
  SPEAKER_03: You got to check them out.
• 1:41
  SPEAKER_03: They have some wonderful courses and and a conference and uh simpler media.
• 1:45
  SPEAKER_03: You might be familiar with them from uh their website, CMS Wire, uh, and of course, uh early information science.
• 1:51
  SPEAKER_03: Uh today's guest started his career as a French naval officer, specializing in counter-piracy work, which has nothing to do with AI, or maybe it will.
• 2:01
  SPEAKER_03: Maybe uh you'll you'll tell us what that has to do with AI.
• 2:04
  SPEAKER_03: Uh yet today he's the global segment lead for applied AI solutions at Amazon Web Services.
• 2:09
  SPEAKER_03: I certainly want to hear more about that journey.
• 2:11
  SPEAKER_03: He's a lover of good wine.
• 2:12
  SPEAKER_03: And joining us from France, please welcome Paul Laserre.
• 2:16
  SPEAKER_03: Thank you.
• 2:17
  SPEAKER_03: Hi, everyone.
• 2:18
  SPEAKER_04: Thanks for having me today.
• 2:20
  SPEAKER_01: So yeah, Paul Paul is one of these rare, rare, you know, one talents, but he's also a good friend.
• 2:27
  SPEAKER_01: And I'm I'm excited to have him on.
• 2:30
  SPEAKER_01: And I'm glad that you know, Paul, that you would uh agree to uh come on to our podcast.
• 2:35
  SPEAKER_01: We have, I think we've grown, Seth.
• 2:38
  SPEAKER_01: We're past the two people that listen to our than two now.
• 2:43
  SPEAKER_03: Yeah, yeah.
• 2:44
  SPEAKER_03: The other day you asked me about both of them.
• 2:46
  SPEAKER_03: You said two.
• 2:47
  SPEAKER_03: I said, well, both of them.
• 2:49
  SPEAKER_01: Yeah.
• 2:50
  SPEAKER_01: I think yeah, we we we've grown past that.
• 2:52
  SPEAKER_01: And and actually I'm getting you know some interesting questions from folks and stuff that we interact with.
• 2:57
  SPEAKER_01: And you know, so Paul, thanks for doing this.
• 2:59
  SPEAKER_01: I think that the cool, the cool part as we get into um this podcast is really understanding, you know, how you got into it.
• 3:08
  SPEAKER_01: I because I there's this, there's this notion, as you know, of of that AI has some mystical elements and properties to it, and it really doesn't, right?
• 3:17
  SPEAKER_01: It's it's it's pretty simple and straightforward.
• 3:19
  SPEAKER_01: I think if you just take that simple, straightforward approach, then you'll you'll find that outcomes that you're really driving to can can really help answer a lot of those questions, but also open the thousand more.
• 3:30
  SPEAKER_01: So, you know, with that, you know, my good friend uh Paul Lessera, he and I were partners in crime at AWS.
• 3:38
  SPEAKER_01: And you know, I can't, I like I said, I'm so pleased to have mom.
• 3:41
  SPEAKER_03: Yeah, like was so so maybe um Paul, you can tell us a little bit about your background.
• 3:47
  SPEAKER_03: And uh, you know, uh I I understand you were on a uh is it a nuclear submarine?
• 3:52
  SPEAKER_04: What was your uh yeah, in the French Navy and thanks, and and once again, thanks for having me super excited and so cool to uh have to spend some more time with uh with Chris again.
• 4:02
  SPEAKER_04: I miss Chris since he went to sales for.
• 4:06
  SPEAKER_04: But to um to to come back on your initial question set on like what's the link between like AI and chasing uh pirates of the Horn of Africa.
• 4:16
  SPEAKER_04: Look, as much as I like to connect the dots backward on this one, I can't really say there was any uh relationship.
• 4:22
  SPEAKER_04: The closest I can relate my experience in the Navy to uh what I uh what I've been doing for the past uh like eight years, nine years now in AI, it's probably uh submarines, uh as you said, where I got the intuition uh uh during my time uh on nuclear submarines that you so if you if you're not familiar with submarines, when you're underwater, there are basically two things uh that keep you busy.
• 4:50
  SPEAKER_04: The first is to figure out what's going on around you, like you're blind, you're underwater, you're just listening, it's all passive.
• 4:58
  SPEAKER_04: So try to just mask your eyes and go the middle of uh crossroads and try to infer the position, speed, course of others around you.
• 5:08
  SPEAKER_04: It's a very difficult mathematical problem.
• 5:11
  SPEAKER_04: The second one is classifying hey, is it like a Russian submarine, a whale, a French submarine, uh what type of vessel?
• 5:20
  SPEAKER_04: And same thing.
• 5:21
  SPEAKER_04: Uh it's a problem which mathematically is pretty difficult if you try to program it, you have explicit rules to find the solution to the problem.
• 5:28
  SPEAKER_04: But if you approach this problem in a machine learning way and uh really apply some basic supervised learning methods, then you tremendously simplify these equations.
• 5:40
  SPEAKER_04: And so that that was my first kind of aha moment, like, wow, this new thing is really powerful for these types of problems.
• 5:46
  SPEAKER_04: And I used what I understood then to pitch in my like business school application uh that I wanted to combine that with my, as you mentioned earlier, with my passion for wine as a good Frenchman.
• 5:58
  SPEAKER_04: And um, and so that's that's what I did um uh during school, like started a little company with a good friend of mine who was uh finishing up his PhD in in deep learning in 2012-2013, that was pretty early on, still, uh not as hot as today.
• 6:16
  SPEAKER_04: And we went and built a system that could help uh online retailers, uh wine retailers to have some type of Netflix of wine online and to better recommend wine based on people's preferences, and that was the name of the company preferences.
• 6:31
  SPEAKER_04: So that's how I made my transition.
• 6:36
  SPEAKER_04: Like it's like you need to like you don't need to be impressed by people who make a career in AI often that has to do with like just like and just like how things come together.
• 6:46
  SPEAKER_04: In my case, that was like totally by accident.
• 6:49
  SPEAKER_01: So, you know, Paul, I'd love to know how you went how you got over to um to because you you started part of the the the D probably AI approaches to all of this, especially in customer service, and you know, really looking at it from the customer experience lens over at Genesis.
• 7:05
  SPEAKER_01: How'd you get over to Genesis?
• 7:07
  SPEAKER_04: Yeah, and at probably a little bit less glorious, where the thing in wine was super like exciting, we were real, but we could not we could not make enough money to sustain ourselves in the Bay Area.
• 7:19
  SPEAKER_04: Uh so after after after school, we decided not to pursue this this uh little project of ours.
• 7:26
  SPEAKER_04: And so I was in the Bay Area with like married and uh a kid a second on the way.
• 7:31
  SPEAKER_04: I was like, what what will I do to have a regular paycheck at the end of the at the end of the months?
• 7:36
  SPEAKER_04: And I figured, hey, you know what?
• 7:38
  SPEAKER_04: Uh going into your patient, like wine for a living, might not be the best idea.
• 7:43
  SPEAKER_04: You have a lot of people who want to do the same thing.
• 7:45
  SPEAKER_04: A patient is meant to spend money, not to make money at the end of the day.
• 7:49
  SPEAKER_04: And so I tried uh to to to identify the industries where I could apply uh what I learned.
• 7:57
  SPEAKER_04: And and talking with like the the person who would become my boss at Genesis, I've I I thought that hey, customers, customer support, customer experience sounds like a fantastic industry.
• 8:08
  SPEAKER_04: It's huge, it's not as shiny as wine, so you might not have the same competition there.
• 8:14
  SPEAKER_04: And it's at the end of the day, if you think about like what customer experience, customer support is all about, it's answering questions um and and fixing issues that tend to be the same, come back very often given a given organization.
• 8:30
  SPEAKER_04: So it's it's a great machine learning problem.
• 8:33
  SPEAKER_04: Uh, it's a great pattern uh recognition pattern matching problem.
• 8:37
  SPEAKER_04: And so that's how I got into customer experience.
• 8:42
  SPEAKER_04: Once again, 2014-15, it was early.
• 8:45
  SPEAKER_04: Like uh AIML, I remember very well this deck from our CMO was one of the 20 futuristic trends.
• 8:52
  SPEAKER_04: Uh it was it was early for many of our customers and investors.
• 8:57
  SPEAKER_04: Uh, so I did a few things before really starting to uh apply AIML at Genesis.
• 9:03
  SPEAKER_04: Uh, when I think maybe one or two years later it became obvious that Contact Center would be definitely changed by this technology.
• 9:11
  SPEAKER_04: That's when I got the opportunity uh with my uh with my good friends uh uh Kono in RD to start an AI group that's like from scratch grew to more than 100 people in a couple of years, acquired a few companies, started a few very exciting partnerships, including AWS, Google, and others.
• 9:31
  SPEAKER_01: That's awesome.
• 9:32
  SPEAKER_01: So your time at Genesis, I know that was it was uh probably built a lot of foundations that we you know took, you know, later because I was certainly grateful to have that, you know, somebody that was helping me and accelerate what I was trying to do, but at the same time, and we'll go into the history of you and I in a minute, but just in terms of those baseline foundations that were critical, right, for us to be successful.
• 9:54
  SPEAKER_01: So, in terms of you know, like the pieces at Genesis, you built up a team, and then um, so what was it what was it like to to um to take because Genesis has been around a while, especially with the Prem products?
• 10:10
  SPEAKER_01: What was it like to take these new concepts and ideas and have to shop them around, right, to the different product owners and things and try to get a pigeonhole?
• 10:19
  SPEAKER_01: And what was the value prop that you would go and you know, because internal selling is a little bit different from let's say the proposition to the customers, right?
• 10:27
  SPEAKER_01: What was that like?
• 10:28
  SPEAKER_04: It's a good question.
• 10:29
  SPEAKER_04: So you you happy dig into my memories, but um I'd say um double sale, like we need to sell to our owners and the leadership of the company to like bet with us, and as you said, you need to go and convince all of the product owners.
• 10:47
  SPEAKER_04: Um on the on the ladder, uh, we've not been successful.
• 10:53
  SPEAKER_04: Uh meaning that and you I'm sure you've seen that at Cisco and other places, Chris, where you come, you see you see like the revolution that is about to happen, and you go and you pitch that to people, and some people tell you, we are we're already doing this, like there is nothing new.
• 11:10
  SPEAKER_04: And in well, in fact, not really.
• 11:12
  SPEAKER_02: Yeah, right.
• 11:13
  SPEAKER_04: Right.
• 11:15
  SPEAKER_04: Isn't it?
• 11:16
  SPEAKER_04: And and and so we we could not really um uh make it work with a decentralized approach, and we ended up, and then I'll answer the first uh question, but we ended up building something quite centralized where we put all of the people in the same team, uh product managers, engineers, and and uh pre-sales, if you will, and under the same roof, and and build critical mass and AI platform, launch the first product before it became uh ubiquitous and and decentralized and went to all of the products because it was hard to start uh in a decentralized way.
• 11:53
  SPEAKER_04: Uh, before getting there, to convince the owners and the leadership of the company, um especially when it was still early and not as obvious as maybe today or two or three years ago, uh we we got like some like uh discrete under the radar POC uh POC's proof of concept going with large customers, starting with uh something we we called at the time predictive matching that became predictive routing, which means that you use machine learning to better uh route the calls and the chats that go to uh customer service representatives based on the outcome you try to optimize instead of rules like hey, you want your NPS to go higher and control the average time the conversation lasts.
• 12:42
  SPEAKER_04: These are things machine learning systems can do very well.
• 12:45
  SPEAKER_04: So we did some POCs and uh some great successes, and that that was kind of the hard evidence uh that uh really was enough to give us like uh get us started and then acquire Alto Cloud in Ireland and and then go and from success to success growth to a decent team.
• 13:06
  SPEAKER_03: I I want to pick up on that thread of you went to people uh with this vision and they said, Yeah, we're already doing that, right?
• 13:14
  SPEAKER_03: And and yet they weren't.
• 13:16
  SPEAKER_03: And and that's and I've seen that a lot, right?
• 13:18
  SPEAKER_03: I've seen when people say, Oh yeah, yeah, we got that, we got that.
• 13:21
  SPEAKER_03: Um, and then and it so tell me a little bit more about the delta between what people thought they were doing or what they what they said they were doing versus what your vision was.
• 13:33
  SPEAKER_03: And how did you kind of get to that level of fidelity where you can say, no, no, no, no, you're not actually doing this thing that we're talking about.
• 13:41
  SPEAKER_03: You do you want to talk a little bit more about that?
• 13:44
  SPEAKER_03: Because I've seen it quite a bit.
• 13:45
  SPEAKER_04: Yeah, and look, uh, it really comes down to um people, relationship, and human at the end of the day.
• 13:53
  SPEAKER_04: Uh, I don't think there is one valid answer for all scenarios.
• 13:56
  SPEAKER_04: I saw people like quickly just with like uh concrete evidence and just by uh logical uh demonstration, like come come along and others uh take a uh an irrational path, protecting some like baby, or you know, sometimes it's just emotional.
• 14:14
  SPEAKER_04: And I've been like that in some cases too, so I totally get it.
• 14:17
  SPEAKER_04: And in this case, when you can't get people along with just like hard evidence, you uh, as Jeff Bezos said at Amazon once for two people competing on the same project, said like two is better than zero.
• 14:30
  SPEAKER_04: And so you basically uh you let these things run in parallel, and you you accept to have a little bit like uh waste of resources in a way.
• 14:39
  SPEAKER_04: Sure.
• 14:39
  SPEAKER_04: And at one point it becomes so obvious that one is winning that they just like help everybody uh rally around the winner.
• 14:47
  SPEAKER_03: So it's it's doing it's it's doing those experiment, those experiments, it's funding those POCs, it's trying these different approaches in parallel.
• 14:56
  SPEAKER_03: As you say, it's there's some experimentation that may be a little bit of inefficiency, but that way you can actually see how these things play out, um, which I think is interesting.
• 15:05
  SPEAKER_03: You know, again, I I I remember uh uh seeing this in one organization and they said, Oh, yeah, we're doing all that stuff.
• 15:11
  SPEAKER_03: And they said, But if you're doing all that stuff and you're not getting the results, then you're not really doing that stuff.
• 15:17
  SPEAKER_03: You know, it's like, what are your results?
• 15:19
  SPEAKER_03: What can you measure?
• 15:20
  SPEAKER_03: Um, there's an interesting thing when it comes to measurement that you you talked about uh improving NPS, for example, net promoter score, right?
• 15:28
  SPEAKER_03: Well, you you know, what's the likelihood of of um of recommending this company and controlling the length of a conversation?
• 15:35
  SPEAKER_03: So time per incident or uh or uh uh whatever that whatever that metric is.
• 15:40
  SPEAKER_03: And many times those are in opposition.
• 15:42
  SPEAKER_03: If you want to talk a little bit more about that, especially in the contact center, you know, there there are many times conflicting objectives and conflicting measures.
• 15:51
  SPEAKER_03: How do you kind of approach that?
• 15:53
  SPEAKER_04: I know it's and and yeah, no, that that's a good question.
• 15:56
  SPEAKER_04: And on this particular example in routing, uh, some people who manage like routing strategies into like thousands of employees, contact center will tell you it's sometimes more uh an art than a science, uh, because there are so many parameters and you need to tweak that as uh as an instrument, a musical instrument.
• 16:15
  SPEAKER_04: And um the it and that's where by the way, you see the power of of the human intelligence, because if you let if you let do a black box and optimize for some outcome, uh like every now and then, like uh a Spanish speaker will talk to a Japanese agent, and you know what I mean?
• 16:35
  SPEAKER_04: There are some there are there are some aspects where there is really no point to have uh a deep learning system like where explicit rules can do the job.
• 16:45
  SPEAKER_04: Uh like if language is a good example, like uh some keywords like like trigger an escalation, like there are there are plenty of examples where you really want to control with rules.
• 16:56
  SPEAKER_04: That's one.
• 16:57
  SPEAKER_04: Uh number two is to your question how do you control the parameter?
• 17:01
  SPEAKER_04: Like it's it's awesome if you have uh uh one point, one extra NPS net promoter score uh points on average, but if all of your conversation lasts 10 times uh like your your CFO might not like it.
• 17:16
  SPEAKER_04: So we you basically um optimize your reward function to uh define the weight and the limit you want to put on each of these uh outcomes, like the outcome could could go on the classic contact center metrics, like uh uh resolution.
• 17:37
  SPEAKER_04: First call resolution, uh the the the time of the of the average time of the conversation, uh the number of like transfers, this type of thing, and to the more like and more and more to the more uh sales and marketing outcome.
• 17:54
  SPEAKER_04: Uh so NPS is is obviously one, but like retention of customers, uh upsell, cross-sell, and you can have a nice mix of hey, depending on the queue, you you might want to have this trade-off or this outcome.
• 18:09
  SPEAKER_04: And so that really is where the ah of a good like uh running.
• 18:16
  SPEAKER_04: Yes, comes into play.
• 18:17
  SPEAKER_04: Absolutely.
• 18:18
  SPEAKER_03: Yes, yes.
• 18:18
  SPEAKER_03: And I like what you said about the fact that sometimes a rules-based system is okay, right?
• 18:23
  SPEAKER_03: There was a there's an article, there's a uh special topic issue in Harvard, Harvard Business Review on on uh workplace AI.
• 18:30
  SPEAKER_03: Actually, um, my article was also featured in that on uh data infrastructure for uh AI.
• 18:36
  SPEAKER_03: Uh, but in one of the articles, it talked about how less data is actually going to be better in some situations because you'll put it in the context of a rules-based or expert-based system, because there's still value in expert-based systems, there's still value in rule-based systems, there's still value in explicit knowledge capture, right?
• 18:59
  SPEAKER_03: You can't go all the way at the end of the spectrum and say, oh, it's machine learning and data, data, and data, and more data, more data, more data, right?
• 19:06
  SPEAKER_04: Yeah, at the end of the day, yeah, I totally agree.
• 19:09
  SPEAKER_04: And there is a bit of hyper on AI, and people want to have as many AI systems in production as possible.
• 19:16
  SPEAKER_04: But at the end of the day, you want to apply these technologies for amount of data where uh uh a human mind cannot make sense of the patterns, cannot really understand like the logic and connect the dots.
• 19:27
  SPEAKER_04: And that's where the power of machine learning system that scale almost like infinitely uh is making a huge difference.
• 19:35
  SPEAKER_04: But with small data sets, like what why would you want to do that if uh you make a L B.
• 19:43
  SPEAKER_03: And in cognitive systems, we talk about cognitive AI.
• 19:47
  SPEAKER_03: I always talk about that is a bit of a misnomer because the systems don't think, but what they do is they reduce the cognitive load on the human.
• 19:55
  SPEAKER_03: Do you want to talk a little bit about that?
• 19:57
  SPEAKER_03: Because I think there's a lot of um organizations that are still kind of wrestling with well, how I call it the last mile of cognitive AI, right?
• 20:07
  SPEAKER_03: The last mile is the knowledge and terminology and expertise of the organization, right?
• 20:14
  SPEAKER_03: Because you're gonna have say a general learning, uh, general language model that will interpret, you know, common terminology.
• 20:21
  SPEAKER_03: But when it comes to you know, a high, highly specialized, like uh, you know, pharmaceutical uh like uh life sciences uh technology company, you know, there's a lot of very specialized nomenclature and a lot of very specialized knowledge that still needs to be captured and codified, and that you can't have a general uh language model apply to.
• 20:41
  SPEAKER_03: Do you want to talk a little bit more about that and uh how that works in terms of you know uh AI solutions for the call center, that that explicit knowledge uh capture and codification?
• 20:52
  SPEAKER_04: Yeah, and and that's I I don't think that's specific to contact center, but that's where once again you see like that the like the the winning association is often machine plus uh human.
• 21:05
  SPEAKER_04: And right it's true in in games like chess, and that's true in like technology and industry in many industries as well.
• 21:13
  SPEAKER_04: Um, if you like I was uh we we had a conference uh this uh this year some great customer success stories um uh based on uh what we call intelligent document processing, the ability to process a lot of of these documents uh uh semi-automatically to do most of the job uh with a robot, and have like a suggestion given to an expert who is able to, uh, for instance, for COVID in airports to understand like what the machine assessment was and to quickly have a look and to make a call based on their like medical knowledge, but accelerate the process tremendously.
• 21:52
  SPEAKER_04: In contact center, it starts with uh tagging the data if you have specialized corpus like.
• 22:00
  SPEAKER_04: That's what like Chris and I were like doing for this solution we launched at AWS called uh contact center intelligence, where you use primitives like transcription, later language processing, translation, they are general like uh capabilities that anybody has developed.
• 22:18
  SPEAKER_02: Like a platform service.
• 22:21
  SPEAKER_04: Exactly.
• 22:22
  SPEAKER_04: But if like just like proper names and just your medicine names and and um the vocabulary you might use in a given industry in a given company cannot be trained by default by a general uh system like this, and that's where uh uh having like some custom um uh language models, some custom vocabulary, some ability to really add the specifics of your business with this like general model is quite powerful.
• 22:49
  SPEAKER_04: And for that, you need people, you need people who come who have the expertise, who tag uh these uh different uh semantic elements, uh tweak the system, and then at the end of the day, often you have human in the loop, you have like a confidence level, and you need someone with the expertise to make a call.
• 23:10
  SPEAKER_01: Right.
• 23:11
  SPEAKER_01: There was there was a there was a video that I saw not too long ago where this guy was comparing um AI and deep learning to um what you need to do in terms of uh, you know, it was like a um like a domino.
• 23:25
  SPEAKER_01: You know, you flip the dominoes over and they just it it starts to go down and do a lot of things automated.
• 23:31
  SPEAKER_01: However, in this scenario, you know, it was like I push the domino and then it pushes a ball, and that ball pushes a stick, and that stick pushes a string and that goldberg apparatus.
• 23:43
  SPEAKER_01: Exactly.
• 23:44
  SPEAKER_01: However, he goes, Now, this is this is the difference between what um what does you know you do in a normal automated way, but in ANML, there's somebody there that's every step that there has to be that there would be a new transition into something different.
• 23:59
  SPEAKER_01: There was somebody there that was, and he was like shepherding it the whole way.
• 24:02
  SPEAKER_01: It was super funny because it looked like he was doing it all himself, right?
• 24:08
  SPEAKER_01: But the idea was that uh you almost have to act like a shepherd through a lot of these cases to make sure that you know, like all these new questions start to come up.
• 24:19
  SPEAKER_01: Am I getting the right outcome?
• 24:21
  SPEAKER_01: Is it driving the right decision?
• 24:24
  SPEAKER_01: Am I introducing bias?
• 24:26
  SPEAKER_01: Am I, you know, doing the reinforcement learning in the right way?
• 24:30
  SPEAKER_01: Am I, you know, have fully supervised learning?
• 24:33
  SPEAKER_01: You know, I mean, there's all of these things now to your point, um, you know, Paul, of bringing in the right people to actually help shepherd it.
• 24:41
  SPEAKER_01: It really does, I think a lot of folks have this misnomer that it's it's just a submit and forget hands off, and I get this wow dazzle amazing thing in the outcome, as opposed to it's a hundred percent now more hands-on than it ever has been in the past, because the data is is now the right data, it's the wrong data, it's the wrong data shape, it's the wrong data type, it's the wrong format, or it's all the right ones now.
• 25:08
  SPEAKER_01: Pulling all those together to make some normalized sense of it is now you know super, super hard or important, right?
• 25:17
  SPEAKER_04: And it and if I can, it shows true with the people you see, like with classic software, you had a bunch of like software developers like together in a room, like writing great codes, like the best uh in the era of machine learning, the best teams are made of like data scientists and subject matter experts, sometimes they don't even code.
• 25:35
  SPEAKER_04: And this collaboration uh between people who understand the technology and people who understand the specifics, said to your previous point, I believe, of are things you can't you can't learn quickly.
• 25:45
  SPEAKER_04: You need to have spent like years in a university seen patients, seen the industry work, since like edge cases come and fix the right boundaries and reward the right way.
• 25:55
  SPEAKER_04: A model is usually the winning theme.
• 25:58
  SPEAKER_01: Well, I mean, we know that that's you know, that's it's always in in my mind, there's there's very few like true, true, true AI experts, and they're the ones that are actually writing the algorithms, right?
• 26:11
  SPEAKER_01: Now, some of that's even questionable in terms of because they'll they'll even say themselves, this is my first scientific you know theory and notation on what I wanted to do, and this is what it should provide.
• 26:22
  SPEAKER_01: Now, results are always going to vary, right?
• 26:24
  SPEAKER_01: It's like, you know, my hat's off to you, Paul, because you know, you were doing early, early AI stuff for speech and rec without things like open source BERT models and the like, right?
• 26:36
  SPEAKER_01: And you know, we were dealing with a lot of these pieces, you know, to try to help create a better customer experience.
• 26:43
  SPEAKER_01: And yet, you know, speech recognition, just one of the foundational primitives here, has been around for a long time, right?
• 26:54
  SPEAKER_01: And so to your point of, well, I already do that, yeah, but it's like driving a car.
• 26:59
  SPEAKER_01: I may have a Model T Ford out there and I drive a car, but it also, you know, gets top speed of 25 miles an hour, right?
• 27:10
  SPEAKER_01: So I'm great, great in, or I have a golf cart, I'm really phenomenal in the neighborhood.
• 27:14
  SPEAKER_01: But if I want to drive to visit my friend in California, it's gonna be a rough go.
• 27:19
  SPEAKER_01: So let's it's interesting.
• 27:21
  SPEAKER_02: Go ahead.
• 27:22
  SPEAKER_01: I was just gonna say, so you know, going back to you know, your time at Genesis, um, you know, one, what was what was the you know, pro, you know, you always I I think your your art form, Paul, is to is to do exactly that, like go in and internally sell the idea and get these things adopted that then become products.
• 27:45
  SPEAKER_01: What do you what's the most you know, the your proudest moment at at uh Genesis?
• 27:50
  SPEAKER_01: And then, you know, why did you leave?
• 27:53
  SPEAKER_01: You know, what was the opportunity at AWS?
• 27:55
  SPEAKER_01: Because that's you know where you are today.
• 27:56
  SPEAKER_04: Yeah, that's that's a good question.
• 27:58
  SPEAKER_04: So the the the product, I'm I'm frankly, I'm very proud about like when I see, especially now looking from AWS uh uh perspective and and seeing the success Gen Z still has in machine learning, one of our largest uh users of some of our AIML products.
• 28:18
  SPEAKER_04: Um very proud of the like being gone a long way, and it was very fun to start it from scratch and to like design the first PowerPoint and go and do some random convincing and start and fail and start differently.
• 28:32
  SPEAKER_04: And so that that that that made me very proud.
• 28:35
  SPEAKER_04: I would say the thing we did uh which I think is still like a work in progress, and it's still a work in progress in many organizations, is not only to come up with new AI products, but then have the ability once you start to have two, three, four, five different uh applications or products, how can you now create a new platform, a new ecosystem of AI ML products that's all of them reinforce, uh reinforce one another?
• 29:02
  SPEAKER_04: For instance, we talked about this predictive routing system where you want to optimize like who talks to whom based on the outcome you you seek.
• 29:11
  SPEAKER_04: Like imagine now in a contact center if you can uh optimize the scheduling of your agents, knowing the average like uh uh characteristic of like a thousand colours this day at this time, and and you know like the forecast of calls and you know the attributes of your agent, and now you you map to uh machine learning systems that each pursue the same outcomes uh with different boundaries, and and you got you do that with the next system and the next system, and your entire enterprise now works together with loosely correlating uh applications, correlated applications all working against the same goal.
• 29:50
  SPEAKER_04: So that I think that's something we started to do at Genesis.
• 29:53
  SPEAKER_04: I see as a trend in the markets, and uh that I spareheaded in a way and makes makes me very proud.
• 30:01
  SPEAKER_04: Um you wanted to do that.
• 30:02
  SPEAKER_03: Yeah, no, I think that I think you're making some really great points is that the the synergies between these different and these different components and the different primitives that you're putting together and assembling these in these, you know, uh Lego building block uh types of paradigms.
• 30:20
  SPEAKER_03: And then you're seeing lots of incremental improvements throughout the enterprise because the same things that can answer questions for the call center agent can answer questions for the HR representative or for the engineer or for the salesperson or for the marketing person, right?
• 30:36
  SPEAKER_03: So the same ideas can be applied everywhere in the organization.
• 30:39
  SPEAKER_03: That's when you start seeing those really interesting.
• 30:42
  SPEAKER_03: I like to say we speed up the information metabolism of the enterprise, which is getting everybody to get faster uh solution answers to their questions.
• 30:51
  SPEAKER_03: By the way, before we continue, I just want to uh uh introduce you again uh with uh Paul Lester, who is the global segment lead for applied AI solutions and Amazon Web Services.
• 31:03
  SPEAKER_03: Uh, and I also want to uh once again thank our sponsors, uh the Marketing AI Institute, uh, early information science, and uh simpler media.
• 31:12
  SPEAKER_03: So just wanted to uh uh get that in uh before we carry on.
• 31:16
  SPEAKER_03: Um, one question I have is proactive versus reactive customer service.
• 31:21
  SPEAKER_03: That was one of the things we talked about before.
• 31:23
  SPEAKER_03: Do you want to talk about how the evolution of AI machine learning has kind of assisted, you know, usually a call center is like, oh, you have a problem, you know, now call me and now I'll react to that problem.
• 31:34
  SPEAKER_03: You want to talk a little bit more about that kind of evolution to proactive uh versus reactive.
• 31:39
  SPEAKER_04: Yeah, and and uh I'll do and I'll just quickly address the the question Chris asked about like moving from Genesis to AWS.
• 31:48
  SPEAKER_04: One of the reasons, and I'm and I'm happy to talk about contact center all day long, but one of the reasons was that I became very, very specialized in contact center.
• 31:57
  SPEAKER_04: And as much as I love contact center and I loved my time at Genesis, I did not see myself necessarily do a career in contact center.
• 32:04
  SPEAKER_03: And and when I got the opportunity at AWS, not that there's anything wrong with the career in contact center.
• 32:11
  SPEAKER_04: Like, believe me, I try to spend less time in contact center that always come back to me and I and I never say no because I enjoy it.
• 32:18
  SPEAKER_04: But I had the opportunity to basically apply what I learned at Genesis and to do it uh across a bunch of different like industries and and and technology domains, like from media to business intelligence to many enterprise like systems, uh families.
• 32:34
  SPEAKER_04: So that was the the the the main rationale, but to to come back to contact center sets on the uh proactive versus reactive, which is true in many domains, but you're right, customer experience is one of probably one of the best examples of these shifts, where if you think about it for years, for decades, enterprises, organizations have waited for people to have a problem to reach out and want to talk to their customers, usually unhappy customers at this time.
• 33:06
  SPEAKER_04: And and and you can't blame uh this organization because how would you do it differently?
• 33:11
  SPEAKER_04: Like, do you want to pay people to just monitor every single one of your customers and call them up and say, Do you have any problems now?
• 33:18
  SPEAKER_04: Do you have any problems now?
• 33:19
  SPEAKER_04: Is everything okay?
• 33:20
  SPEAKER_04: Everything okay?
• 33:22
  SPEAKER_04: It's just it's just not it's just not it was just not possible.
• 33:26
  SPEAKER_04: And and now in this world of like this treasure trove of data and the ability to have systems, understand the boundary of what look uh good looks, and and and when someone is going out of what a good customer journey is supposed to be, right?
• 33:42
  SPEAKER_04: Then proactively reach out, and that's what's happening in so many organizations today who understand that the technology allows these shifts, this totally uh different way to talk to your customers.
• 33:54
  SPEAKER_04: Uh, for instance, in Contact Center, I talked about Alto Cloud, which was an acquisition Genesis made in 2017, 18, uh 18, I want to say, which was specialized in predictive engagement, which is exactly what you what you talk about here, which is hey, just have a system to monitor what's going on online.
• 34:15
  SPEAKER_04: Uh, if you start uh an application for a loan online and you feel a questioner and you stop feeling the questioner, it's likely that something wrong is happening.
• 34:25
  SPEAKER_04: So don't wait for this guy to look for the number and call you, just like assess what's the so first assess when you should do something.
• 34:33
  SPEAKER_04: Like, is it just like a temporary, like someone like gets coffee, or is there a real issue?
• 34:39
  SPEAKER_04: Assess that uh probably probabilistically, and two use a second model to assess what's the best uh intervention, proactive intervention with likely to bring this person back on the what the what right should should be, whether that's an offer, whether that's assistance, whether that's calling you, whether that's chat, whether that's just uh an article, and and you can apply this logic in the entire like world of customer support.
• 35:06
  SPEAKER_03: Right, right.
• 35:07
  SPEAKER_03: It's so interesting.
• 35:08
  SPEAKER_03: And you know, you you talked about just now you mentioned what is the right in prevention, right?
• 35:14
  SPEAKER_03: Because what's the thing that people really hate when you walk into a store, someone walking up and saying, Can I help you?
• 35:20
  SPEAKER_03: Can I help you?
• 35:20
  SPEAKER_03: Can I no, no, no, just looking, right?
• 35:22
  SPEAKER_03: Just looking.
• 35:23
  SPEAKER_03: It's it feels like an intrusion, right?
• 35:25
  SPEAKER_03: So, how do you do that without being intrusive?
• 35:27
  SPEAKER_03: When I go to websites, the first thing that pops up is a bot saying, How can I help you?
• 35:32
  SPEAKER_03: It's like, you know, I want to exit out immediately, right?
• 35:36
  SPEAKER_03: Yeah, yeah.
• 35:37
  SPEAKER_03: So that proactivity can get kind of step over that line and be intrusive, right?
• 35:43
  SPEAKER_04: Yeah.
• 35:43
  SPEAKER_04: And and look, you then you have different ways to to act on it.
• 35:49
  SPEAKER_04: Like, I don't I don't want to sound too salesy with like Amazon, uh but like wherever you are my my Amazon ad.
• 35:58
  SPEAKER_04: I I and as like frankly, as a just as a customer of Amazon.com, I'm always uh impressed by the proactive nature of customer support on Amazon.
• 36:09
  SPEAKER_04: Not that someone is calling you or putting a uh a chatbot uh in your face, but we are super data-driven, so we understand what's going on.
• 36:20
  SPEAKER_04: We can analyze where these journeys sometimes go uh left or right and and and and and act accordingly with the right um uh help at the at the at the time and the place you're most likely to need them, right?
• 36:35
  SPEAKER_04: Uh fixing issues before they happen based on the experience you've seen before.
• 36:38
  SPEAKER_04: So proactive can mean just changing like the roles customers are taking, they are less likely to call you.
• 36:46
  SPEAKER_04: Yeah, and then and that's what we see.
• 36:48
  SPEAKER_04: Like, usually what works the best is like like diminishing the the reasons to to call.
• 36:55
  SPEAKER_04: Now you will always have people who want to call you, want to chat, want to reach on the channel of their choice, the the device of their choice when they want, uh, with their rules, and you want to be there for that for them, and and there is no shortcut to and that's the beauty of automating what you can.
• 37:12
  SPEAKER_04: You free up your your people for the human touch in what really matters.
• 37:16
  SPEAKER_03: And the critical piece of this is understanding that journey, and understanding it can go wrong, right?
• 37:23
  SPEAKER_03: And anticipating that, and then looking for the anomalous behaviors, the ones that says, wait a minute, they should have done this, they didn't do that.
• 37:31
  SPEAKER_03: Why did that happen?
• 37:33
  SPEAKER_03: And that's and you don't understand that, you can't anticipate that's the key.
• 37:37
  SPEAKER_04: You can you yeah, you can't manage it if you if you can't measure it.
• 37:40
  SPEAKER_04: And with Chris, we always uh told customer uh and and our partners uh on this topic in just call something's voice.
• 37:47
  SPEAKER_04: Like, start by opening the black box, like like transcribe the entire like months of like communications with your customer, transcribe it, analyze it, put it on the dashboard, and look at the top three issues people are calling you for before you you take any action.
• 38:05
  SPEAKER_04: Just go in a room, analyze that, look at it, understand it, and then you can think about the actions that will uh diminish these level of issues.
• 38:15
  SPEAKER_01: Define irony because how many opportunities, and this is this is for all of those out there who are trying to gather their customer sentiment off of surveys, right?
• 38:26
  SPEAKER_01: Which is like which is one lot, and or to your point, measure MPS or CSAT or whatever, which is interesting because you know, to your point on the on the proactive, it it can be overbearing and cause a negative experience.
• 38:39
  SPEAKER_01: And on the other side, if they're not reactive at the right timing, that causes a negative experience.
• 38:45
  SPEAKER_01: So now I'm gonna I'm gonna have a I'm gonna have a you're you're you know, just to set the scenario, you're calling in, you're mad about something or concerned about something.
• 38:54
  SPEAKER_01: I can make the conversation worse or better.
• 38:58
  SPEAKER_01: And then I'm gonna ask you, hey, can you please take a survey at the end of this just to let me know how I did?
• 39:04
  SPEAKER_01: Okay, well, we know that less than one percent of people even want to take a damn survey anyway.
• 39:09
  SPEAKER_01: And two, the only ones that are actually really emotionally escalated or upset are gonna be the ones to take that survey because people that are happy won't take it, right?
• 39:19
  SPEAKER_01: So now we have all these false, you know, these false negatives in terms of these result sets that are going to skew and formulate our MPS scores and our CSAT, it makes no sense at all, as opposed to trying to get markers and scores in there of the conversation at the beginning, you know, into the middle and into the end, and actually deriving out like we're talking about, did they take the optimal journey?
• 39:42
  SPEAKER_01: Did that optimal journey yield the results that they were looking for, which was solving their problem or not?
• 39:47
  SPEAKER_01: And in terms of the context, and are they calling back in?
• 39:51
  SPEAKER_01: I mean, Paul, we ran into that how many times with different customers that were like, well, no, no, no, you know, talk about I already do this.
• 39:58
  SPEAKER_01: No, uh today I'm I I've got surveys that are collecting all of this information about you know, about how I'm doing.
• 40:06
  SPEAKER_01: And you wonder why you have a loyalty or brand issue.
• 40:10
  SPEAKER_01: You know what I mean?
• 40:11
  SPEAKER_01: Yeah, and then there's I mean companies today do it all the time.
• 40:15
  SPEAKER_01: And you know, it's helping convince them past that to say, listen, guys, let's let's give you some primitive building blocks.
• 40:22
  SPEAKER_01: Let's just, like you said, Paul, let's take your voice conversations and just analyze crawl walk run, just the recorded calls that you have first and understand what's going on.
• 40:32
  SPEAKER_01: And then let's maybe get into you know a real-time NES best action where instead of, like you said, the proactiveness of being always on top of trying to make sure you're okay, creating automated connection points that are dictated by the customer.
• 40:48
  SPEAKER_01: Yeah, I want you to reach out to me when there's a problem, but let me know across what channel first, right?
• 40:55
  SPEAKER_01: Because of how we we've interacted.
• 40:57
  SPEAKER_03: Well, and and the other is uh to reach out to me when you say you will.
• 41:02
  SPEAKER_03: I mean, I think there's a lot of process issues and there's a lot of basic stuff blocking and tackling.
• 41:07
  SPEAKER_03: And I'm I'm still looking forward to the opportunity to tell my story about my horrendous Verizon wireless customer service experience.
• 41:18
  SPEAKER_03: But I'm gonna I'm gonna like drip that out over time, I think, you know, get another mileage out of it.
• 41:24
  SPEAKER_03: We should have uh suffice it to say, they were gonna call me back three times, three separate engagement uh uh uh situations and never did.
• 41:32
  SPEAKER_03: But uh but it's like doing what you say you're gonna do, right?
• 41:35
  SPEAKER_03: Understand the process, understand what those signals are, but then even doing the basic blocking and tackling, I think is so important, right?
• 41:43
  SPEAKER_03: You might be looking for these insights or these hidden motivators or these hidden signals, but you know what?
• 41:49
  SPEAKER_01: There's a lot of explicit stuff that you just to field service type of uh activities like you're talking about.
• 41:57
  SPEAKER_01: What's more frustrating than I've got this issue and they give me a window of time that they're gonna call back on, like between 8 p.m.
• 42:04
  SPEAKER_01: and 8 a.m.
• 42:05
  SPEAKER_01: I'll get back to you, right?
• 42:07
  SPEAKER_01: Anyway, let's get back to it.
• 42:08
  SPEAKER_01: So, Paul, you made you you made a comment, and this is one of the things that I was actually I wanted to um to double click on because that's why I asked you that notion of of what was intriguing about moving to AWS, right?
• 42:21
  SPEAKER_01: And you mentioned that it was really the opportunity to look at so many different AI and ml problems and and uh opportunities, right, with customers and stuff.
• 42:33
  SPEAKER_01: Give give me a give me a uh, you know, the the folks that are listening to this, give us an idea of what that entails.
• 42:39
  SPEAKER_01: We always see we know customer service.
• 42:41
  SPEAKER_01: What else do you get to work on in AWS and what does that that uh scope look like?
• 42:46
  SPEAKER_04: Yeah, no, that's that's that's a good question.
• 42:49
  SPEAKER_04: And and and on what you said just before, Seth, which is a transition to extend what we did in in Contact Center uh with fulfillment, like uh doing uh some great work with uh companies like Pega, WorkAdO, more on the business automation side of things, who can uh talk to the systems who like listen to you and get insights and can automate some of the fulfillment.
• 43:12
  SPEAKER_04: So you don't stay waiting for a company uh was promised to call you three times, it's all part of the same process and the same journey, and and you take action.
• 43:20
  SPEAKER_04: So that's that's one thing uh also uh that that I find super excited outside of customer experience.
• 43:27
  SPEAKER_04: Um, but a few areas where we've seen like a lot of success with customers and and and and analysts and also partners of AWS in the past like year or two.
• 43:38
  SPEAKER_04: I mentioned before like Community Dairlines who uh came talk on on stage in last December in Vegas, talking about how they developed in no time uh an application using your primitives for uh text uh recognition and NLP to very quickly be able to understand if you were compliant.
• 44:00
  SPEAKER_04: With your like coming from different countries, with your uh vaccination form, with your COVID test, which on average took like seven minutes uh for like uh controllers at the gate, which was a nightmare for airlines.
• 44:15
  SPEAKER_04: So uh that was uh one of the solutions, and we we apply the same logic to insurance claim uh processing to bank loans and automate this process where it usually takes a lot of people and training to understand different forms possible.
• 44:32
  SPEAKER_04: Uh we uh also and and Chris was involved.
• 44:37
  SPEAKER_04: We and I saw that one of your guests was Adam uh Sutherland, one of uh our an colleague at AWS as well, where we developed a solution uh with like close to 20 companies well known in media and entertainment, in education, in tech, to same concept as in contact center, open the black box.
• 45:00
  SPEAKER_04: Like you have archive, you have data, uh like just analyze it, understand it.
• 45:06
  SPEAKER_04: So the next time you search for something, you want to do a special edition of your last like 50 years uh of like magazines, as uh was the case for one of our customers, instead of hiring like three interns and go and listen and watch uh uh like uh hours of content, just like uh find it very quick, uh, create thumbnails, uh uh semi-automatically, um translate and localize content in games in no time, uh, monetize your uh your your content and your videos, especially in an era where cookies are in danger, uh like optimizing advertising based on content and and not based on who is watching it, uh, is also a theme that that uh is super exciting and we we do work on.
• 45:54
  SPEAKER_04: Uh lastly, uh I should say, Chris, uh I don't know if you followed reInvent or conference in December, we announced something called AI for data analytics, which basically means that we worked with the leaders in business intelligence, in data warehouses, such as the tableau, the snowflake of the world.
• 46:14
  SPEAKER_04: And and basically uh like the thinking is one of the reasons why many organizations today struggle with machine learning is because they don't find the talent.
• 46:26
  SPEAKER_04: Data scientists are a scarce resource today, they tend to hang out with large like technology companies, hot startups, and many organizations would like to do more with machine learning, they just don't find the uh the people.
• 46:38
  SPEAKER_04: And so, what if you were to empower good data people?
• 46:44
  SPEAKER_04: People who are subject matter experts, know how to manipulate data, build a dashboard, uh, find insights, but don't code, like are afraid of a common line interface.
• 46:55
  SPEAKER_04: And the idea was let's let's meet these people where they are and go work with the tools they are using today so they can move from descriptive analytics to predictive and prescriptive forms of analytics without having to learn new skills and no need to code, no need to create a model.
• 47:15
  SPEAKER_04: So that's some of the uh the the concrete application of machine learning where we see a lot of interest from our customers who look for turnkey solutions with vendor that can make it easy for them to install and and and and put in place for the business.
• 47:33
  SPEAKER_03: But they still need to understand their process, right?
• 47:36
  SPEAKER_03: You can't always start with you don't understand you can't automate a mess.
• 47:42
  SPEAKER_04: Exactly, exactly.
• 47:44
  SPEAKER_03: That comes back to those fundamentals, right?
• 47:47
  SPEAKER_03: You can use those platforms and applications and those building blocks to know what your outcome is going to be, what you're trying to accomplish, what your business problem is, how you're gonna measure it, all of those sort of sorts of things.
• 48:01
  SPEAKER_01: So what's next?
• 48:02
  SPEAKER_01: What's next for for you, Paul?
• 48:03
  SPEAKER_01: You know, we did.
• 48:04
  SPEAKER_01: I mean, I'll I'll tell you what uh um I was so grateful to have you as uh somebody helped me drive a lot of the contacts and intelligence.
• 48:11
  SPEAKER_01: And congratulations, by the way, for getting that on the main stage at at uh the big customer conference for uh for AWS.
• 48:18
  SPEAKER_01: That uh yeah, there's a lot of um well, and and the reason why you know we did that was, you know, as as like you know was mentioned, um, you know, if you have two solutions is better than zero solutions, even if they're cooperative, right?
• 48:32
  SPEAKER_01: And competitive.
• 48:33
  SPEAKER_01: And so Paul and I created a very competitive and cooperative type of an approach to all the customer service stuff with contact centers.
• 48:41
  SPEAKER_01: And uh, and it was no, you know, at first we had uh it was just you know, he and I driving this whole uh program.
• 48:48
  SPEAKER_01: And then it did balloon and and got bigger and perpetuated itself past us, which was the goal, which there was a lot of org parts of the organization that didn't didn't like, but we're proud of what we did because that was one of the, you know, that's how you know you're creating creating something that's uh disruptive and or good because there's you're annoying people.
• 49:10
  SPEAKER_01: You're exactly well, it's it's just you know, it's disrupting what they're doing, right?
• 49:15
  SPEAKER_01: Right.
• 49:15
  SPEAKER_01: And so, you know, so that's how we were able to do some some phenomenal stuff.
• 49:20
  SPEAKER_01: And then we started in the media side of things, like we talked about with Adam, you know, on the previous podcast.
• 49:25
  SPEAKER_01: But we started with media and did some phenomenal things in media, but taking these building block primitives and actually applying, there's really no one size fits all.
• 49:32
  SPEAKER_01: However, there are building block services that you can put together to actually just create an accelerator into something that can be tailored and you know customized.
• 49:42
  SPEAKER_01: So now with with all these uh data insights.
• 49:45
  SPEAKER_01: So what's next, for you, my friend?
• 49:46
  SPEAKER_01: What do you what are you uh what are you working on now that you can share with us without you know without getting you know going and blowing up some of the uh some of the internal compliance names of AWS?
• 49:57
  SPEAKER_04: Yeah, so we don't we don't communicate on roadmaps uh as you know, Chris.
• 50:01
  SPEAKER_04: So I can't I can't share what's not been announced, but I can tell you that think about all of the hot use cases in machine learning in at least in the world of enterprise, from industrial AI to uh enterprise search, uh content moderation, like these are all they are all hot topics where we have the capabilities at AWS, and and and really what's uh what excites me today and what I'm very proud to see coming back said to our discussion about I'm I'm already doing it, I've seen and I've seen that genesis, I've seen that in the world of startup, I'm seeing it sometimes at AWS.
• 50:35
  SPEAKER_04: There is some proud around developing your own machine learning systems, like in-house.
• 50:40
  SPEAKER_04: And the the truth is that like a small company that is very specialized in a given industry will likely not be known for transcription.
• 50:49
  SPEAKER_04: And so uh my message often to customers and our partners is really like pick your bottle.
• 50:55
  SPEAKER_04: Like right, uh, if if if a machine learning system is not going to differentiate your business, like do something else, like just focus your engineers, your talents in data science to the the the path that will be proper to your business, what you want to be known for.
• 51:09
  SPEAKER_04: And I've seen that really go ahead, that's a differentiator, right?
• 51:13
  SPEAKER_03: It's it's exactly you can't outsource your competitive advantage, but you shouldn't be not part of your competitive difference.
• 51:21
  SPEAKER_04: And when you say it, it sounds so natural and obvious that you will not believe the number of companies that kind of outsourced initially their machine learning to a group of data centers were excited by like doing cool shit, excuse my French.
• 51:35
  SPEAKER_04: And and and and they could be they could beat Google, they could beat AWS on this like uh like transcription, translation, uh, OCR system, you name it, and but there is really no point in my opinion.
• 51:47
  SPEAKER_04: So that's that's one of my messages.
• 51:49
  SPEAKER_04: I was uh I was fortunate enough uh to have like a discussion with Anju Wang a couple of years ago where I was telling uh some of uh some of these challenges with engineers, and he told me, Look, Paul, that's exactly the same here.
• 52:02
  SPEAKER_04: I work with some of the most talented students and and they could do it all themselves, and I constantly have to find so they focus on the thing that is going to uh they want to be known for.
• 52:12
  SPEAKER_04: And and the way he phrased it is don't sprint in front of the train, and it sticks with me, and that's what I repeat to my customers and partners.
• 52:20
  SPEAKER_04: Like, of course, you can beat like a large company, like in the short term, like on this particular problem, but you will probably not win in the long run.
• 52:29
  SPEAKER_04: And right this battle to start with is not meaningful, it's not relevant for you.
• 52:33
  SPEAKER_04: So pick your battle, uh, decide what you you want to be known for, and just like leverage the work done by the entire industry and build on top of it.
• 52:42
  SPEAKER_04: And I think like most people get it today and apply that logic, and that's why we see so much fast innovation in AIML.
• 52:48
  SPEAKER_03: That's great.
• 52:49
  SPEAKER_03: Thank you so much.
• 52:50
  SPEAKER_03: I love that.
• 52:50
  SPEAKER_03: Uh I I didn't catch how you phrased it, but it's don't don't sprint in front of a train.
• 52:56
  SPEAKER_03: Like, don't you can you'll you'll win in the short term, but you won't live in the long term.
• 53:02
  SPEAKER_03: That's exactly the idea.
• 53:03
  SPEAKER_03: Yeah.
• 53:03
  SPEAKER_03: Right.
• 53:04
  SPEAKER_03: You don't want to be hit by that freight train.
• 53:06
  SPEAKER_03: Yeah.
• 53:06
  SPEAKER_01: Well, you know, a phrase, a phrase we used at Amazon a lot was let's not break into jail.
• 53:12
  SPEAKER_01: Let's not break in jail.
• 53:13
  SPEAKER_01: Yeah.
• 53:13
  SPEAKER_03: Well, listen, we come to the end of our time today.
• 53:16
  SPEAKER_03: It really has been a tremendous pleasure uh speaking with you, Paul.
• 53:19
  SPEAKER_03: This has really been a lot of fun.
• 53:21
  SPEAKER_03: And uh I do want to again remind people our sponsors Tippler Media, which is uh CMS Wire, Early Information Science, and the Marketing AI Institute.
• 53:30
  SPEAKER_03: And again, uh Chris, wonderful uh uh having you to uh help out with this and be our partner at Prime.
• 53:36
  SPEAKER_03: And Paul, fantastic having you on.
• 53:38
  SPEAKER_03: We'd love to have you on again.
• 53:40
  SPEAKER_03: And and uh just some tremendous insights and all right, guys, thank you so much.
• 53:44
  SPEAKER_03: Have a wonderful uh week and weekend, and we'll talk to you soon.
• 53:49
  SPEAKER_04: Thanks for having me.
• 53:50
  SPEAKER_04: Take care, folks.
• 53:55
  SPEAKER_00: Thanks for listening to the early AI podcast.