Charlotte Ward: 0:13
Hello and welcome to episode 317 of the Customer Support Leaders Podcast. I'm Charlotte Ward. Today Andrei Kamarouski talks about how AI is reshaping customer support tools. Today I'd like to welcome Andrei Kamarouski. Andre, it's lovely to have you back after a few years' break. Lovely to see you again. Welcome back.
Andrei Kamarouski: 0:41
Yeah, thank you, Charlotte. Happy to have you again in this talk. It's incredibly that so many years already. And yeah, my life changed a lot. So we talked last time it was about relancing. Uh that time I was doing more, and now it's completely different life.
Charlotte Ward: 1:02
Yeah, yeah. I mean, everyone's lives has changed
Charlotte Ward: 1:05
in the last I mean three days, it feels like. Never mind, six months to a year or a few years. Um, do you want to tell us what you're up to now then?
Andrei Kamarouski: 1:14
Yeah, sure. You know, like six years ago, I was doing a lot of consulting and doing many projects, uh, playing with just my two hands with uh 20-30 projects at the time. And I was to operate the guy on Upwork community about Zandask, customer support, research. And then I started doing more and more activity in uh my startup, yeah, uh, where I was co-founder and and still co-founder. And we were focusing exactly on AI in customer support domain with focus on Zendesk apps and tools. You know, I was not the guy who just managed this, but the guy who was building some AI-based of myself and still doing this today. Yeah, so my my current focus is actually building the product like kind of solo founder uh track now, mostly where I'm doing uh everything for true with AI and where AI works. And you know, that's uh like a biggest evolution that I personally was feeling over these years, not just as a guy who sells AI, yeah, but as a guy who works with AI, yeah, and this takes a lot of significance now in everything.
Charlotte Ward: 2:34
Yeah, it really does, doesn't it? It touches every part of how you build a product, how you sell a product, how you even think about the scope of a product, what a product is, right? I um and I guess as a founder, um, particularly having I I suppose kind of worked this several times now, but first as a freelancer and that and now as a founder. Um do you is your experience one of like reshaping the whole process?
Andrei Kamarouski: 3:06
Yeah, absolutely. So let's let's say
Andrei Kamarouski: 3:10
just to uh periods before and after generative AI. Yeah, before generative AI, you just use AI as a very short tool that should be specifically customized, like trained, fine-tuned, whatever you name it, for your needs. Yeah. And once generative AI uh was like happened like three, four years ago, and yeah, with some time, not initially, but then you can do just a lot of things out of the box with these tools, right? Like if you want classify tickets or you want to analyze them, whatever. And like, you know, I remember when I was building my first uh ticket classification model, I spent months just for collecting data, cleaning this, etc. etc. And now it just works with a correct drone just up to the box, and many other things just up to the box, yeah. So another big shift is that you can actually build products with AI. I mean, identicoding everything around this. Like before, you need um weeks and months for building one feature, and today you can do this in uh in a few days, right? It's uh incredible speed of uh building that is possible. It doesn't mean that everything you build is meaningful, yeah. You you might test, etc., but the speed of changes, the speed of how the market generally is running uh up today is is also uh incredible. And I think that we're still in the phase where AI is mostly augmenting us. It's a good teammate, yeah. It shouldn't be kind of full automation, just in some simplest operations. And that's uh but we are heading to the uh situation where AI will be more and more autonomous and will be able to deliver not just execution layer or simple management like observation layer, but also some more sophisticated decision making in the in the stack of all the operations and tools that we have.
Charlotte Ward: 5:19
So it's so we're entering a world where it's starting to to exhibit, if not actually have,
Charlotte Ward: 5:25
it's starting to do something that looks like judgment, right?
Andrei Kamarouski: 5:29
Exactly, exactly. Judgment. Um, you know, human thinking it's it's a pretty complex process, but still it is uh kind of um yeah, it can be reproduced in some way, right? And more challenging things like imagination, creativity, empathy, all this stuff is kind of more challenging yet. Yeah. But uh rational, like data-based, uh like algorithmic or workflows-driven thinking is quite uh possible to automate. The challenges for complex thinking is to provide complex context, like different different data, it should be provided uh in a comprehensive way from multiple finals, in a correct uh format, etc. Right? Plus, the rules, how decisions are being made, should be also clear and transparent, prompted properly, and so on and so on. You know, um like the the one tool that uh we like have now um uh real-time analytics that uh today it just provides the signals, like the data at a glance, what happens, right? The next step here should be in like that. I will working is introducing kind of copilot that will consume a lot of signals and make suggestions early. Let's say you need more work, uh workforce here or here, plus making suggestions, or if you allow them, making some automated decisions, changing where agents, what groups they are working, what skills they have, uh, etc. So depending on different metrics like the volume, the SLAs, the CSAT, whatever, yeah, that you have. That's uh kind of from copilot to more uh yeah, more autonomous agent that could work on your behalf with you in the loop where you want it.
Charlotte Ward: 7:33
Yeah, yeah, yeah. I mean, it's fascinating to me how the dynamics are shifting. Um that that's I can feel we're on the brink of we've gone from helper to co-pilot, as you said. Like we're already there, really. Particularly for simple, not tasks even, but ecosystems where um the thinking that you're talking, talking about is driven by experience, by human experience, but that human experience is in a quite um well-defined domain or a well-defined data set, or you know, and has usually a relatively finite set of directions in which it might take a person historically. So so as a customer support agent or a customer support leader, um we apply a certain set of experiences around how a customer might behave, or how what things might typically keep a customer happy in this circumstance, or um workforce management. I I you know, I definitely I know the experience levels of my agents in this region. I need to do something slightly different in terms of staffing. Those are kind of the the domain in which each of those is operating does require human experience. Those are skills that you build by being in the job for long enough, but they're codifiable and and with enough, to your point, enough capturing of the context and the data and the and and I guess almost downloading of the the possible decisions or likely decisions born from that experience. That's that's what we're really able to do now, isn't it? As you said.
Andrei Kamarouski: 9:26
Absolutely. So you know that the the more uh of different layers of operations and decision making can we kind of absorb with these AI tools, the more the higher the complexity of the tool themselves might be. Yeah. What I see and what I'm
Andrei Kamarouski: 9:45
building myself, uh and I think that many brands already like leaders did it already, many still doing, some are thinking. Like, okay, if we have great agenda coding tools, why couldn't we just build everything in-house? Yeah. Uh like during the last few uh months I was speaking with different uh people uh from past consulting projects and etc. And you know, two guys told me we just rebuilt kind of mini Zendesk in our own company. This was impressive. You know, Zandesk is not a small uh they didn't uh add at all uh yeah, belts and whistles there, but the the core that does the main job was there for sure. And that was uh much cheaper for some for some brands building this was just as paying one month fee, yeah, or tools like Zendesk. Yeah, uh so that's it a big movement of replacing software purchased with software build, yeah. This buyer buyer-build dilemma is one of the I think this the next year will be very uh vibrant and hot for many, many teams if we should build or still proceed in buying this, yeah. And this uh is the main like point here for customer support. Specifically, we are the domain where most of the things that happen in a com as a communication, it's a language, yeah. And language is something that AI is very good at right now. Yeah, really, yeah. Like visual AI or like audio and yeah, images, all these things also good, but language what what it it's everything initially kind of uh started being uh very good, and which means that ability to rebuild this is even more likely, yeah, that you you you can kind of uh do it, do it. Yeah. So the the question I was asking myself in this situation, okay, if I'm still building some software in this domain, what should be this software then should be interesting to companies so that they're still purchasing and buying, yeah, from you.
Charlotte Ward: 12:08
Yeah, I think that's really interesting.
Charlotte Ward: 12:10
Yeah, I think that's really interesting. And to your point about Zendesk and how big of a product footprint Zendesk has, um your um your friend you were talking to about, you know, they built something in-house that effectively does the job for them. You use the word without all the the phrase without all the bells and whistles. And that really stuck with me because um most people, when they buy a product the size of Zen desk, don't use all the bells and whistles. And and while Zendesk might continue to market, Salesforce the same, right? Or, you know, all of these big software um platforms the same. Um while the Zen Desks and the Salesforces of this world might have tons of bells and whistles that they continue to market and say, this is going to be amazing. We've solved this problem for you now, now we've solved this problem and this problem and this problem. Most people most people don't have all of those problems. So you end up using 50% of the product that you bought, or 60%, or 30% just because you buy it for the core thing plus one other feature or something and don't need a great deal else. And then in fact, if anything, it's all even though you only use 50% of it, it still only solves 50% of your problems. So you have to find another product, or you're still left doing a bit of build anyway. So it's kind of the shape of products is changing, I think, is what you're saying, or is going to have to.
Andrei Kamarouski: 13:39
Exactly, exactly. Great point. So yeah, you know, I did hundreds of Xandesk implementations and optimization products, and I rarely saw brands that utilize Zandesk to its maximum, what was available on their plan. There was always, I don't know, like 30, 20, 30 percentage approximately of futures that were not utilized at all or not utilized properly in the best way, let's say this way. And this is exactly the thing. So you probably uh don't need like many companies might be still beneficiating from getting this uh utilized, but in fact, over years they might be working fine without this, right? I mean, the future utilization follows the practices that the company uh can absorb, the team can kind of uh yeah, carry on on a daily basis. Because if you start utilizing a lot of things, this means that you need a lot of small and big disciplines over the day to follow in in order to do this very often. Yeah, yeah. Which means historically software was you know kind of average product for um like you collected a lot of things that were developed for some specific fraction of customers and not for others, yeah. And this means even with plans ladder, it's still like this. And yeah, that's one point. At the same time, you know, again, if we head into the situation where AI is uh yeah uh copilot or autonomous like high autonomy uh helper,
Andrei Kamarouski: 15:20
it means that this guy can leverage the functions and the features, yeah, and this can kind of be uh yeah, utilized and software again uh is getting more complex. I mean the brands when they can build or their again, a uh from building or managing perspective, you can do more, right, with the eye, or you build with the eye or you work with better with the eye, yeah. Which means the comp the the the software idea that should be uh yeah working here should be pretty much complex. And I think it should be kind of horizontally very much uh integrated and well integrated to each other, yeah, because like by Zendesk or other big platforms, historically they acquire different startups and they're trying to integrate them, and usually it's uh never never is uh ideally integrated and they're still kind of disconnected, fragmented, where I don't know, WFM tool doesn't talk well to K tool and to the core uh support. Yeah, support tool. This is a very common thing, which means that there will be for me more space for complex horizontally integrated softwares on top of these core platforms that could do these, you know, operations later, uh integrity better and deliver better and more complex value as a result here.
Charlotte Ward: 16:48
And and more personalized, I suppose, as well. Because what what you find is that at least in my experience is you buy a product like Zendesk and you use 50% of it, but even then the 50%, you end up kind of having to shape your processes to and shape your humans too. So your humans have to work with the product rather than the other way around. And so that I mean, that's part of the attraction of build, isn't it, for anyone? It's like we'll just do the thing we need to do, and we don't need to worry about uh, you know, have we got are we capturing the data in a certain way, or do we do we know that about our customers, or do we care that it, you know, joins up with another tool in a certain way? We can just make it happen the way we want to make it happen. Um, yeah, and and and I can I completely uh I completely understand your point about like when you look at something like Zendesk, not to lay into Zendesk too much here, but they are a good example, Salesforce as well. They keep acquiring these products, and even though they have the Zend Desk Badgeon or the Salesforce badge on, they don't talk well to each other. And that's also part of what Build versus Buy can give you. It's the ability to connect. And that's where the software layer that you're talking about sits, I think.
Andrei Kamarouski: 18:04
Yeah,
Andrei Kamarouski: 18:05
yeah, I agree. Two points here, just what I want to add to your uh so first, like software again, historically it was pretty static. I think the more the more uh robust and and uh fast and capable identity coding will be with time, the more we'll be nearing to more dynamic software that even on a front-end layer on visible uh UIs can be uh like adapting dynamically to what you are doing and like even observing some habits and and things like this, yeah. Uh from simple things like making some simple changes, this button bigger or another color, or moving to another place, or add something a little bit here. You know, it should be kind of generative software. I think like in the next three, five years we will be having uh more and more things like this, so that software is kind of um building itself while you're using it and you're talking to the software, right? Via AI agents who is kind of manager of this software, right? Now, I mean this is pretty much already happening, but with uh a lot of like oversight uh from the humans, like developers, builders who are doing this, but I think this might be pretty much uh quite automated, and this way software might be highly personalized for the companies that using it.
Charlotte Ward: 19:44
Yeah, yeah.
Andrei Kamarouski: 19:45
Yeah, there will be there might be some core because everyone just needs some core things, but then you can uh dynamically uh yeah, expand and change the software per specific uh customers so that this is uh still maintainable at the same time.
Charlotte Ward: 20:02
Yes, yeah. I I think it's I I think it's super interesting the evolution of the button. I've just had this, this has just occurred to me and it's just blown my mind a little bit. Like 10 years ago, um, we were talking about the product feedback loop as like, you know, for example, you you make a product decision, you put a blue button there on the left hand side, and you have a customer come in with an opinion about what they want the button to look like, they want a red button there on the right hand side instead. And so you have product managers weighing up the value of the button and where it should be placed. And and and because it's a one-shot deal, you've got one button, you want to please most of the people most of the time. Changing that button is an engineering effort and a UI design effort, probably as well, and a product manager effort. Um, and the and the blue, you know, the red button might get shipped a month from now. Um and suddenly we're in a world right now where customer A wants blue button, customer B wants red button. They can both have what they want because realistically, making that request goes into the factory and you can serve the product differently tomorrow or this afternoon for two different customers. And I'm just it's just occurred to me like there's a there's a world not in the too distant future where we won't have buttons at all. Like customers won't care about buttons because either they've made their own buttons or they don't need them at all because we're so plugged in, you know, the communications are just driving everything.
Charlotte Ward: 21:35
Humans are actually almost reverting to just talking to each other and saying what they want to happen to each other, and everything is happening a layer below.
Andrei Kamarouski: 21:44
Yeah, yeah, exactly. Another point I just remembered that I wanted to uh also discuss quickly, which is more even more interesting, you know, starting from let's say personalization for the customer perspective, yeah, in order to really personalized like service support, you need to know a lot of things about the customer. You need to remember a lot of things, right? You need to store this correctly and like uh present this accurately to the customer as well. Yeah, because historically very often you are being served as an average customer of an average company, right? The personalization which really makes like I don't know like I don't know like some your your family member helps you right with something. This level of personalization and care and empathy is still really rarely to see across yeah okay companies they have say rules etc but at least from human to human interaction perspective it might be uh like this and should be like this ideally but yeah that's still kind of challenging and in order to be really uh like personalized the support should be extremely uh knowledgeable about the customer which means you need to collect a lot of signals store them properly leverage them properly and apply correctly while you're making customer support and this is again still uh still challenging another point is maybe more uh yeah interesting um there will be more and more AI representatives from real customers contacting customer support and talking to uh uh AI representatives for the customer support teams yeah that means AI will talk to AI yeah like I will on on my behalf I will um ask my AI helper to get I don't know refund or whatever uh sort out with this company when I have the issue which means again that uh in some non-distant future there will be at least on this initial level AIs will be dealing out somehow and imagine that a super smart ai from one side and from another side yeah it's it will be completely also like the demand of being um personalized and able to provide all the data is is is incredible. Let's say these my customer me as a customer my AI will will say hey guy you don't remember I asked you like half a year ago this this thing yeah and you told me this thing and now you tell me completely different they might be just yeah they don't store this they or then don't leverage this information or whatever they process they lose this piece of context or yeah so that might be funny that I don't know uh AI is punishing another AIs for serving bad but yeah that's completely uh yeah to me it will be super big challenge for companies when not humans who can forget who can forgive make mistakes etc but real AIs who kind of if you ask them they don't forget they don't forgive they are very very smart and they will be talking to your customer support real agents or or AI agents imagine how high the requirements of being like capable talking to them correctly will be yeah that might be crazy time um I think I think that's a topic for another podcast Andre should should we should we do that one?
Charlotte Ward: 25:30
Come back in a you know a few months and let's just see how close we are to that because it feels like we are on the brink of that I I think I truly think like it's um it's very very it seems like a minority report or something you know it seems like such a like such a futuristic uh vision but it's we're on the brink of it aren't we and I think there's a whole discussion we can have there around um how close AI needs to get to being human without being human because to your point an AI will never be fallible it will remember it's not going to be easily influenced it can't be blackmailed it can't be you know it can't it won't respond to it won't be empathic in times of not truly empathic in times of need and things like that. We have to talk about that another time will you come back and have that conversation with me thank you so much. It's it's super interesting to discuss the possible shape of tooling going forward and um I I think I it's it's really something to watch.
Charlotte Ward: 26:35
I'm super interested to see what you do with that. And so I think what I'm also going to do is have you back to talk about that.
Andrei Kamarouski: 26:42
So I'm what I'm doing now is banking two more conversations with you Andre thank you yeah it was a great conversation I enjoyed it and thank you for having me today.
Charlotte Ward: 26:51
Yeah thank you so much for for coming I'll talk to you soon that's it for today go to customersportleaders.com forward slash three one seven for the show notes and I'll see you next time
