Show all episodes

There’s Only One Thing Wrong with this Car featuring Alyona Medelyan

Released on MARCH 22, 2024

Before Baywatch, David Hasselhoff starred in Knight Rider, a television series that featured KITT, an AI-enabled 1982 Pontiac Firebird Trans Am that featured the iconic pulsing red scanner bar. But while Hasselhoff was the star of the show, it was Bonnie Barstow, played by Patricia McPherson, who was the brains behind KITT. As the lead design engineer for KITT’s systems, Bonnie’s character was a pioneering representation of women in STEM fields.

While the names grabbing many of the headlines related to AI belong to men, plenty of incredibly talented women are making their mark in the field as well. While Bonnie was using AI to help Michael Knight battle criminals, Alyona Medelyan has been using AI to understand customer feedback. Her love of linguistics combined with her early interest in computers led her to a PhD in Artificial Intelligence and the founding of Thematic.

We discuss:

  • Understanding AI and Machine Learning
  • Women in Tech and AI
  • Starting Thematic
  • Gathering and Analyzing Customer Feedback
  • Benefits of Y Combinator
  • Maintaining Focus and Achieving Growth

Connect with Alyona on LinkedIn


Music courtesy of Big Red Horse


Rob Dwyer (00:03.15)
Alyona Medelyan, thank you for being Next in Queue. Welcome to the show. How are you?

Alyona (00:10.67)
Thank you so much. I'm excited to be here. I'm doing well, thanks.

Rob Dwyer (00:15.598)
We've got to talk about my pronunciation. How did I do there?

Alyona (00:21.496)
Fairly good, I'd say. Acceptable and yeah.

Rob Dwyer (00:23.534)
Fairly good. Okay. I got a strong C plus maybe a B

Alyona (00:30.176)
Hang on, let me do the American way. You did awesome.

Rob Dwyer (00:36.686)
I love it. I love it. So you are joining today to talk about a wide range of things. AI and customer experience related. You have been in the AI field for I feel like longer than most people. You actually have a PhD in.

Correct me if I'm wrong, computer science with a focus in AI or something along those lines.

Alyona (01:08.14)
Yeah, yeah.

Rob Dwyer (01:11.342)
What drew you to that field?

Alyona (01:18.158)
Um, well, initially I was very interested in languages. I grew up in Ukraine and very early on, um, loved German and English and was studying those languages. But also when I was 15 or so, my dad brought a computer and, and I was one of the first people on my street to have one. And I felt like I didn't have this barrier or fear of computers and.

I was studying, I decided to become a translator and enrolled at a university in Ukraine. And then in the applied mathematics, I've heard there's this thing where they use computers to translate text automatically. And I thought, huh, I should join those people because this seems way more future -proof, way more exciting.

And I felt like I could combine two of my passions in one. So this is how in, this was the year 1998, I decided to find a way of studying a field that's called computational linguistics or natural language processing, where you use computers to decipher what people are saying.

Rob Dwyer (02:41.87)
You've got to tell me what that first machine was. What was that first computer?

Alyona (02:49.582)
Something 365 or something. I don't even remember. But it ran Windows.

Rob Dwyer (02:57.038)
You don't even remember. That's OK. OK, all right. It was a Windows machine. All right. That tells us an awful lot. We can make some assumptions on that. So.

Alyona (03:01.87)
Yeah, it ran Windows. I remember that.

Rob Dwyer (03:14.862)
You mentioned a couple of terms just now. Computational, linguistics, and natural language processing. I'd actually like to spend a little bit of time having you maybe break down some of these terms that we hear all the time in the media these days. Now the media is actually in on it, right? AI is everywhere, but I feel like there is this...

lack of transparency or lack of understanding about some of the specific terms and what they really mean. Can you maybe just start with something that I think we don't hear a lot, but what's the difference between like AI and machine learning? Or are they the same?

Alyona (04:08.174)
Great question. I think I would say machine learning is one of the methods to.

implement an AI algorithm.

It's where you basically learning something from the data. Most recently, all of the methods that people use before to make sense of text that were more algorithm based and less machine learning, they kind of, um, they're basically going away and nobody's not that many people are using them. And pretty much everybody's using machine learning today to make sense of.

numbers and data and language. So this is why they kind of becoming more synonymous, but originally they weren't.

Rob Dwyer (05:05.518)
Can you talk about the difference in something like, I don't know, text analytics and semantic analysis? What are those? Why should I care about them?

Alyona (05:21.23)
Well, text analytics is more like, okay, what are we doing here? We're analyzing text and it's very much a term used in business settings and not in academic settings. In academic settings, or in research, you use natural language processing and this encompasses, again, various methods to make sense of text. And, um,

machine learning, a lot of them are using machine learning. And I think what's critical today is this idea of a language model. And this is something that has been quite a niche term that we used to use in, in when, when I studied and, and then ultimately when we were, have been building these algorithms and that Chan GPT came along and.

Suddenly everybody knows what a language model is. And chat GPT is basically using GPT 3 .5 or GPT 4. A language model that is a very large one, trained on huge volumes of text. And this model can not only make sense of text, it can also generate text. So it can do way more.

than original language models that were created. For example, one of the first ones that's been extremely successful was a language model that can detect what language the text is written in.

So by creating, creating, like inferring those rules from the text. Yeah, sorry.

Rob Dwyer (07:01.87)
I imagine that is.

Rob Dwyer (07:07.214)
Yeah, and that would be very useful. I wonder, have you been surprised by the velocity of change, particularly as it relates to large language models, having been in this field for so long and write your love of language and AI and computers working with and decoding?

Have you been surprised by the velocity of change?

Alyona (07:41.774)
Definitely, I think most of us have been very surprised. Even the people who created Chagy Pity, I bet they were surprised as well. And I read the article about Geoff Hinton, the creator of deep learning algorithms that drive a lot of these large language models.

And he was surprised. And at first I felt bad. How could I not have predicted this? And then when I read that he was surprised, I like, you know, it's okay.

Rob Dwyer (08:20.27)
makes you feel better when you realize some other people that might have predicted it also didn't.

Rob Dwyer (08:31.342)

There are a lot of people that we see on TV, on the internet that are involved in AI. And I've got to tell you, most of them that I see are men.

Is it a struggle for you at times to be a woman in this field? Have you seen progression? Like what's your take on women in tech and in particular in AI?

Alyona (09:09.87)
There's some incredibly amazing women in AI and I always seek them out, follow them. They're my role models. Same with women in business. There are very few women and successful women in business as well. They're not as well supported. And it is hard. It is hard because a lot of those things is like pattern matching. Who do you pick to join your team? Who do you give your money to as investments? Who do you go with?

We all seek familiarity. So I don't necessarily blame other people for the situation. I want to be, I don't want to talk about the fact that it's hard. I want to put myself out there just like I'm doing today and tell my story and hope that it will inspire other women to take a chance and follow their dreams and be an example to other women, to their daughters.

I'm trying to kind of look at this in a positive way.

Rob Dwyer (10:14.414)
Yeah, I love that. I mean, speaking of your story, can you tell us how thematic came about, how you got started?

Alyona (10:27.79)
Yeah, absolutely. So I knew that I wanted to start my own company and I was listening to all of these podcasts, how to start a startup, entrepreneurship course on Stanford. And they all said, you need to start with a problem worth solving. And I just didn't have that problem. There was nothing in my life where I was like, oh, I need to.

I need to fix this and I think I know how. So I decided to become a consultant in like a freelance machine learning AI expert. And I've always been writing blog posts about things I know. And this is also putting my code as open source available to others. And this is how people were finding me and or through my network. And.

giving you a variety of tasks where they thought AI could help. And some of them are really, really interesting. For example, there's the biggest liquor search engine in the world is ran out of New Zealand. And as you can imagine, all of the wine names, it's language, right? And they needed to map them. And they were like, what if...

Rob Dwyer (11:47.566)

Alyona (11:54.894)
We could hold our phone to a wine bottle and immediately it would bring up all of the information about that wine and the rating. What if we could say the name of wine into the telephone? And so these are the kinds of solution that I was working a variety. And then gradually several companies reached out to me with a problem. It was very, very similar. In fact, it was very niche as well, which is really great for starting startups.

and it was analyzing net promoter score feedback.

So two of them were actual end users. So large corporates where NPS was a KPI for senior leadership. And the insights managers were, they knew that the insights is all inside this open ended text. Why did you give us a score? How could they convert this into the drivers of NPS? So this was about nine years ago. And...

I happen to know NPS because my husband worked at a company called Serato that did music software for DJs, really famous. And he was bragging to me how their NPS is higher than Apple's. You know, he was very proud, very proud. And so I knew what NPS was. And so I started to think like, how can we solve this problem using

Rob Dwyer (13:13.934)
Ha ha.

Alyona (13:28.862)
AI and there were language models at the time. Some of the first one were emerging and I tried the model and I could see that it can actually make sense and string together things that when somebody says, um,

a solution was expensive or it costs too much. It could make this connection. I think these two things are actually the same. And this is what we're trying to do when analyzing customer feedback. We're trying to turn all of this messy text into and structure it and count it and quantify it. So this is how thematic started. I asked those companies for some data, wrote some software.

created a PowerPoint, a few PowerPoint slides that show the drivers and ask them, is this kind of what you're, what you have in mind? And they said, yes, yes, it looks really good. And, and then I asked them, well, would you like to see a proposal? And we basically signed our first customer before we even had a proper solution.

Rob Dwyer (14:42.83)
exciting to be able to really kind of dig into a problem and solve it and go, oh, OK, I've got something here. Let's go.

Alyona (14:56.27)
Yeah, it's, uh, and it's very difficult for people who work in technology to, to not be too interested in the technology itself and more to be more interested in the solution to the problem that people actually have. And I found, I really struggled with this as well, because you can, you can build so much and now you can build even more.

with large language models, you can go in and it will start writing code for you. You don't even know how to program. It can help you. And I feel like this actually creates more problems. Like you have to be more careful about what you're actually building and will people want it.

Rob Dwyer (15:48.494)
Yeah, it certainly is understanding is there a market demand or what you're developing before you put a lot of effort into it makes a makes a big difference.

I am curious how.

Alyona (16:05.806)
but it's also links to customer experience as well. Sorry.

Rob Dwyer (16:10.51)
Yeah, it absolutely does. I wanted to before we get there, I want to know how you ended up in New Zealand.

Alyona (16:22.958)
Well, I was studying in Germany and I knew that the language of research is English and I really needed to improve my English. And somehow there was an exchange between the two universities, between machine learning lab in Germany and machine learning lab in New Zealand. And I first came on an exchange and I fell in love with the country and people here. And it didn't bother me that it's so far away.

And I ended up moving here to do my PhD and stayed here. But I did also move to US and I lived in US for several years. And I also love the culture and people in the United States as well.

Rob Dwyer (17:07.982)
I've never heard anyone say anything bad about New Zealand and I'm waiting for it, but so far it's not happened. So.

Alyona (17:19.758)
Oh, there's plenty of problems locally. Yeah.

Rob Dwyer (17:20.718)
Well, I think we're all missing out. And personally, I want to check it out. But let's talk about customer experience and AI. This is actually a huge talking point. There are all kinds of different technology solutions out there. There are people making

their own GPTs, there are people creating solutions, some of which it's questionable whether there's a problem involved or not. And you mentioned that impacts customer experience. So can we explore that a little bit?

Alyona (18:11.95)
Yeah, absolutely. I think they're massive.

advantages and benefits that AI can bring and already has brought to customer experience.

Um, the user, a lot of user interfaces that people interact with are driven by AI and it provides a great customer experience. Like if you think about it, like, even if you write a text message to your friend and it order suggests the next word, what it helps you order complete, come order complete your email. That's natural language processing and it creates a more seamless and positive experience because you don't have to do like text speak or.

And it's just easy. Or when you're talking to your phone and it creates, it writes the text. And I think this is how companies should think about AI. What are some of the unique challenges that their customers have? What are the difficult, tricky parts for them? And then how can technology help me solve those problems? And.

Try to think small and find little nuggets and things that are contained so that you're not end up with trying to solve too much at once.

Rob Dwyer (19:40.91)
Yeah, I think that is a really great advice for anyone involved in trying to improve customer experiences. Really just narrow your focus on what can potentially be some really big wins in a narrow focus, because it's easier to solve a singular problem than it is to just try and fix everything all at once. And...

There are some really interesting and powerful tools today that allow you to do those things if you apply them correctly.

Alyona (20:25.134)
Yeah, exactly. So what we help companies do is analyze customer feedback and find some of those gaps. So for example, we've analyzed thousands of app store reviews for different bank apps. And I know you guys still live in Stone Age in the US. Do you still use checks?

Rob Dwyer (20:47.47)
We do still up some people occasionally use checks. Yes, they are still a thing here.

Alyona (20:57.358)
And it continues to be a massive problem for users. Huge pain point. A friend of mine just moved to New Zealand. He had to deal with check, receiving a check, and the app wouldn't recognize his check. And this is a solution that a lot of technology is out there to solve this, right? And it would improve.

the experience for so many people if it just worked. The way that we unlock our phone and it's just with your face, it works, right? Why can I not recognize a check if you must use them?

Rob Dwyer (21:42.446)
I can tell you what that experience is like because I did just recently deposit a check. Now, I will say I have been using Capital One since it was actually ING before Capital One bought ING's banking division. And they've had mobile check deposit for a long time. But that process is a lot of...

holding your phone and aligning it just right and hoping that your camera focus works really well. And then sometimes it'll take the picture and then it'll be like, I decided I didn't like that very much. Can you do that again for me? It's certainly easier than going down to the branch to deposit the check.

But you're right, it can be a frustrating experience if it doesn't work very well. And I feel like my app probably works better than most just because they've been doing it so long. I feel like they've got it down pretty good. I know some other banks are just now introducing that and I can only imagine the pain that is in that when it doesn't work.

Alyona (23:03.246)
Yeah. And there's so many similar examples, right? When people are sharing like, this doesn't work for me. I can't log in. This is like across all of our customers. This is a big, big thing. Cannot access my account. Forgot password. Has to have to contact support. Imagine how much time support time can be saved if that part also just worked.

Rob Dwyer (23:29.262)
I'm curious, for you to go out and analyze feedback, how hard is it to keep up on all of the avenues where customers provide feedback and ensure that you're gathering feedback from all of those sources? How do you approach that?

Alyona (23:53.262)
So at Thematic, we have built integrations. And when we onboard a new customer, we try to make sure that there is no manual uploads and try to streamline the process of gathering feedback. But another way that companies solve this is by using a central database where they pull data from various tools that they use into that database, central database, BigQuery or Redshift.

and then push via our API, this data to us. So these are two main reasons, but it is, it is a tricky, a tricky problem because then you have to decide, okay, if you were to prioritize one issue versus another, how do you handle a, let's say a quarterly survey versus a firehose of call center interactions?

This is where at this stage you need the human leading those insights, making those decisions and gathering evidence. Yeah.

Rob Dwyer (25:06.83)
Yeah, there are so many different avenues and I feel like it is a challenge for leaders to understand where they should prioritize what they need to put the most weight on. I don't know. Do you offer any suggestions on that or are you more in the business of just trying to deliver the insights and let them figure out where?

they should put that priority. What experience do you have?

Alyona (25:44.206)
I feel that delivering insights is more important. We are all about democratizing insights and making sure that anybody in the company who is making decisions can get access to the voice of customer, can interpret the data that's relevant to what they're doing. And I think there are so many variables that are used when you're trying to prioritize.

And a lot of this we don't even have access to. It could be strategic initiatives that are coming from the C level based on outlook, what's in the market, who is building what. And this is the role of product manager or a strategist, you know, anybody in strategy. Like we're just there to make sure that they have the access to the data that they need in a way that's.

really easy for them to make sense of it.

Rob Dwyer (26:45.71)
Absolutely. I'd like to talk a little bit. You had an experience with Y Combinator, which is a startup accelerator for those that don't know about that.

For people who have never had that kind of experience, can you just give us a little bit of insight into...

What benefits do you get out of that and maybe what the hardest part of being involved in that was?

Alyona (27:23.342)
Yeah, We're big fans of Y Combinator, but before I joined them, I didn't know much about them or the program and who it's for. And it's like, if you can think about it like Harvard for startups, it's very difficult to get in. They only pick the best companies. And...

the likelihood of success is much higher. So companies like Airbnb, Dropbox, Stripe came out of this program. And so they use interesting criteria, very different to VCs that kind of try to pattern match and often follow each other. Whereas Y Combinator,

It's all about some sort of a unique insight into an industry that other people didn't have. So for example, when I started thematic, I had some of the largest customer feedback management solution tell me what's wrong with word clouds. All of them had word clouds. And I'm like, no, this is not what C -level is using to make decisions on.

And so we were very quick to...

Rob Dwyer (28:50.446)
It makes great wall art, though. Great wall art.

Alyona (28:54.894)
Yeah, some sort of a placeholder. People call them the mullets of the internet.

Rob Dwyer (29:04.334)
Ha ha!

Alyona (29:08.878)
And they look at the team as well, because the number one reason why startups fail is when the founders disagree on something. So this is a big criteria and then traction and unusual traction. So how come a company from New Zealand is signing Stripe, which was one of our first customers and they were already a...

extremely successful company. And how come did we sign this Fortune 500 on a quite a significant deal? Um, it was like two people. How did they do this? So they clearly, they saw something in our company and they almost kind of head hunted us. They had this program where they did online mentoring of startups and they encouraged us to apply. And the program is you basically get there and you're.

one of 100 startups and you just meet twice a week once with your, with your kind of designated partner one -on -one and one as a, as a big group and a small group. And, but in the rest of the time, you're just building your business, but you're surrounded by other people on the same journey, trying to work towards this event called demo day where you basically.

present your progress over three months and that is how much you've achieved and there are investors in the room and you we ended up very quickly raising around straight after that, but we really loved kind of the energy and and focus on one metric, which is your revenue in majority of cases. Can you increase your revenue? And they have a, a model that, um,

make something people want, which I feel is quite related to the whole idea of customer experience. What is it that our customers want? It's not about what they need. It's not about what they'd like to or unhappy with. It's like, what do they want? And the want is measured in revenue.

Alyona (31:36.27)
And so how do you do that is again, related to customer experience. You go and talk to your users, you go and listen to your customers. And whenever you're not listening to your customers, you build the solution based on what you've heard and you repeat again and again. And if you're a small startup, this is all you're supposed to do. And they keep repeating this, this Vantra.

And this is why a lot of the startups succeed because they live very focused. Whereas, um, if you don't have, it's so easy to get distracted and start doing things that are irrelevant or, um, I don't know, go to meetups or some events or spend time online, um, learning about things you, you just, you should just be listening to your customers. And I think.

the most successful CX programs are all about that as well.

Rob Dwyer (32:41.422)
I love that you talk about that focus of getting customer feedback because.

Rob Dwyer (32:51.854)
It is very easy for all of us in our lives or in our business that we're running or whatever it is that we're doing for us to lose focus of the ultimate goal. Right. And so having a North star and keeping your focus on that North star can really change how you go about managing your day, managing your activities, what it is that you're going.

to do in a day, in a week, in a month. And that's just a really wonderful reminder. And the success of companies that have gone through that process, I think, is pretty self -evident. I imagine as well that it helps not only for you to bounce some ideas off people, but there's some accountability that comes with it.

that regular meeting with other people that are doing the same thing that you're doing, right?

Alyona (33:56.59)
Definitely. Yeah. Having some sort of a mastermind session and with people who are going through the same thing is really important. We were, I remember the first session we had, we actually lost one of the customers that we signed before we joined the program. And so when everybody was sharing how they've increased their revenue by this much, ours has dropped and was so embarrassing.

And we felt like we let everybody down in our group. And we set really aggressive goals for ourselves. We worked six days a week. And thanks to my mom, who came along and looked after our little kids, two and four at a time. And we, yeah, when we did end up growing our revenue, 3X.

Rob Dwyer (34:47.982)
Oh no.

Rob Dwyer (34:52.462)
It's a lot to manage.

Alyona (34:56.718)
in three months.

Rob Dwyer (34:57.356)

That is fantastic. That is fantastic. And with a two -year -old and a four -year -old at home, I would imagine that it's hard to also.

Alyona (35:01.708)

Rob Dwyer (35:16.174)
Be present with your family as well. And I think that's one of those challenges where we can still get distracted. But at least when you return to work to have that North Star to keep you focused while you are in that mode is incredibly valuable.

Alyona (35:42.222)
Yeah, and this is completely unsustainable. We knew that we could only do it for three months. This is once in a lifetime opportunity. Ultimately, you don't want to burn out. You don't want to have any negative effects on your family. So ultimately, family comes first. And so we make sure that we spend a lot of time with our kids.

But we also think it's important for them to see parents do exciting things and love what they're doing.

Rob Dwyer (36:11.18)

Rob Dwyer (36:18.062)
I imagine too, once you have achieved a certain velocity of growth, that it becomes easier to spend less time focused on, I have to get growth, I have to get growth, I have to get growth. Because once you achieve some velocity, it starts to kind of come at you, doesn't it?

Alyona (36:48.238)
Well, the whole idea is that if you are focused on building what people want, eventually it becomes this, in startup land they call it product market fit. But a lot of companies don't get that far and they do get distracted and build things that potentially like many different people want, but it's hard. So.

It's a combination. So your offering and the market, it all needs to fit together for the snowball effect, where instead of feeling like you're pushing a boulder up the hill, things just happen. And I think very few people actually reach that stage where the business is just running itself without a lot of effort. Because even if it's like,

It can start growing like crazy and then your efforts and like, how can I find enough people to sustain this? This can also happen.

Rob Dwyer (37:55.726)
Yeah, it becomes then an effort in recruiting and making sure that you start building out all kinds of other structures to sustain the business.

Alyona (38:07.278)
Yeah. So when we grew 3X in those three months, we signed so many companies, we couldn't board them as two people. We couldn't onboard them. And I still feel embarrassed. After one of them churned, going to the, doing the customer feedback session and the interview and listening, and they telling us, Oh, your solution couldn't do this, couldn't do that.

And I had to do this and like, but it can, and it does. And obviously it wasn't good enough. The customers, we didn't have customer success. We didn't have training and it's, it's so important to, to make sure that customer actually get value from your product and solution.

Rob Dwyer (38:58.83)
Yeah, absolutely. Alyona, it has been fantastic having you on the show to share your story and all kinds of insights. Is there one thing that we haven't had a chance to talk about that you'd like to share with the audience?

Alyona (39:23.106)

Alyona (39:27.95)


Alyona (39:36.078)
Sure. What did we kind of talk about? All of the things we're going to talk about. And I feel like we've nailed all of the items on the agenda.

Rob Dwyer (39:41.966)

Rob Dwyer (39:45.71)
That's good. We've talked about all the things. Sometimes we just talk about all the things, and there doesn't have to be any.

Alyona (39:52.854)

Rob Dwyer (39:54.798)
Well, I thank you so much for joining the show. If someone wants to reach out to you, maybe they want to understand more about their customer feedback. What's the best way for them to get in touch with you?

Alyona (40:16.11)
I would love to connect with anybody who is listening to this podcast via LinkedIn. This is what I personally use to continue telling my story and things that I encounter on this journey of building a company in AI and CX space. I'm hoping that some of the things that I share will inspire other people to do similar things. So LinkedIn would be the best channel to connect personally with me.

But otherwise, thematic, we have a website if people want to learn more and they can request a demo on their data because this is how people have been evaluating our solution. I prefer to show instead of tell, put in your data and see if what you find surprises you.

Rob Dwyer (41:09.806)
I love it. We'll make sure that those links are both in the show notes. So feel free to reach out and thank you so much for joining us on Next. Thank you.

Alyona (41:24.078)
Thank you for having me and really grateful for the opportunity to tell my story.