This is the journey of David Čaněk and Memsource. Founded in 2010, the company recently sold a majority stake to a private equity firm. What will be the future of Memsource?
Find out in this interview with David, the CEO of Memsource. We also talked about:
- What does “Memsource” even mean
- Why the company invests into AI innovations
- The future of AI beyond MT
- What differentiates Memsource from others
- What are “Non-translatables”
- Why do we have so many MT engines
- The biggest failure of Memsource
- Company run by… a coder?!
PS: due to internet connection issues, David is sometimes out of sync
This is episode #31 of my social interaction practice, also known as The Localization Podcast 🙂 #localization and #translation insight delivered to you by the power of voice, this time with David Čaněk.
Andrej Zito
David Čaněk, welcome to the podcast.
David Čaněk
Thanks, Andrej
Andrej Zito
How are you, my friend? You’re in Italy? What are you doing in Italy right now?
David Čaněk
Yeah, I’m working from here. Very exciting. So, it turns out my you know, so my wife is Italian. So every summer she, she wants to visit her relatives. And also, obviously, our kids want to meet their uncles and aunts and and, and nephews and nieces. And and so you know, it’s, it’s this time that they have are in touch with with this the Italian part of the family. So when I can, I try to be with them. But I obviously cannot take a two month vacation. So they go to the beach, I stay here and work and talk to you or do something else.
Andrej Zito
How has your plans changed? Thanks to the COVID. Is it better now in Europe? Could you have traveled safely? Or did you consider the risks? I was just wondering, like when you told me just a few minutes ago that you are in the delay, I was wondering whether it has something to do with the recent news about Memsource that I just noticed over the weekend. So you sold the majority of the stake at Memsource, but I guess it has something to do with your wife and your family more than selling the stake in themselves. Right.
David Čaněk
Sure. And so you were thinking i i i got Carlisle on board and I and the first thing is I went to the beach, obviously, you know, anyways, yeah, so
Andrej Zito
No, that’s not what I was thinking. I was thinking that maybe they are in Italy you went there to finalize the deal or something.
David Čaněk
No, no, no, but it was a actually a funny story. You know, related to that, actually pre closing is I I was in Italy with my family already. And I wasn’t sure you know that, you know, finalizing this. It took still, you know, a few days, a couple of weeks, even, you know, the whole the three weeks maybe to really finalize everything, and then it’s stretched to four weeks as things kind of obviously take longer than then then you plan initially. And the plan was probably ambitious, you know, the three weeks. So everything went according to plan. But then I had to I had to go back to Prague to be there on the signing. And I was a little worried to actually if I you know, to make it because there were no direct flights with very little direct flights in June between between Italy and Prague. And so the signing was on the first of July. And so I took a flight that at 6:30am from Bali, and I landed safely in prizes on the first of July was the first flight there were only 49 people on the flight. And it’s as a little word to make it but I made it there on time. Everything worked out perfectly.
Andrej Zito
I’m curious if you can share us more details about the sale. Like Was it your original idea that you were looking for an investor or were you contacted by them?
David Čaněk
So no, this was something we wanted to do and and throughout the years, you know, as you as you get a little bigger, you know maybe when you hit the four or 5 million Your recurring revenue, you start getting emails from, you know, investors, and they seem like they want to just buy you the next day, you know, or invest in you the next day or you know, these are very kind of very salesy emails, and at some point, you respond to a few of them, and then you realize, oh, they’re just trying to, you know, figure out where you are, and where’s the market? And where are your competitors and so on. So so you know, we would get these any anyone in this space, anyone, any tech company, I think get they get, you know, get get a lot of emails like these, but this was so. So this was, it was not our first m&a kind of experience. So when but it was the first one that we really, that we really decided we want to go forward with and that we signed, close. And so it was a it was a process that took almost a year from the first day that we said, okay, we’re going to start this process looking for for actually, a new shareholder for memsource. And it turned turned out that a good, good setup was a majority shareholder.
Andrej Zito
So that’s where Memsource is today. I would like to go back to when you had zero annual recurring revenue, your very beginnings. When did you start Memsource and why?
David Čaněk
This was really the main reason
Andrej Zito
Did you saw like a gap in the market that you could fill in? Were you not satisfied with the tools that were available on the market?
David Čaněk
So okay, so So look, the main, I guess, for a lot of people, at least for me, dead time, the main motivator was to start a company I wasn’t, it wasn’t like, oh, there is this market gap, I’m going to start Memsource, so it was the opposite. So I said, Okay, I want to start a company, I want to start something, you know, see if I can do something on my own. And whether what, you know, people, I share values, and I want to work with them, and you know, have a team that you have fun with and don’t do something interesting. And then I also learned that was from my previous jobs where I was, you know, an employee really, I always liked to actually start something, start a job where I didn’t know much about it, I actually was it was, was a beginner beginner at that job, because that’s most interesting. But it’s it’s very risky also. So and so I guess it’s great for an employee like it, and it was this still this period, you know, after the Velvet Revolution was still kind of, we didn’t have a lot, you know, enough senior people in an any business roles. So it was great. It was still this period where you could start a career in pretty much anything. And it was enough that you that you were eager to learn that was that was and that you were smart, and you were able to kind of master it in, you know, in a year and then then do the job. So, so I love that. But then when I started Memsource, I was thinking well, I don’t have I don’t have the cash to to, you know, support to talk to, to really sponsor me and everyone to learn things. And so I was thinking well, I need to, you know, with the with the fonts I’d have, I need to start and we were bootstrap. So I need to start, you know, make being profitable very quickly. And I don’t have the luxury of just learning something for one year then deciding what to do whether it’s it’s a good idea. So we I started something that I knew. And that was translation technology. Because before I worked with a friend of mine at a small translation company, and I saw the translation tech that was being used at the time, I was thinking, well, it’s all desktop may be client server that was kind of, you know, starting that time or not that not that common either. So so you know, this will move to the cloud also, at some point. So I was thinking, well, let’s start a cloud based translation. app.
Andrej Zito
I mentioned the name Memsource started the company with this name. I’m wondering if there’s is there more to the name than just a combination of a memory and source?
David Čaněk
Yeah, I think I told you that right. When we were so most people don’t know what, you know, why, why this name? And, and at that time, when I was starting, the company was not already very difficult to find any domains, you know, that will be available on the.com, you know, the.com domain. So this was after some brainstorming, you know, I was, this is what was available? What was free? You know, and I registered that. And, and it was, I think it was, you know, international with memory source. Yeah, that this is how it, but most people, most people, you know, don’t immediately think of that sometimes, you know, initially I was thinking mate, well, it was quite similar to memo Q. Right. So well, at least the first three letters, well, not completely, maybe now you’re thinking it’s not but but when we were starting, it was funny, because? Because we would we would, you know, we would go to a conference, we would talk to someone and and then we would do ourselves page, and then we would call them back and oh, yeah, you were the memo queue guys ever, like know where that with immense sources, it’s actually not memory. So. So when I say hi to our friends at memo queue, in this way, but, but I think at some point, people realize it’s something different. And, and in the end, I think the name is pretty good.
Andrej Zito
You are bootstrapping from the beginning. And you mentioned that you didn’t want to take too much time to you didn’t have that much money to go and keep learning forever, until you figure out something. But I know from your websites that it took you around two years, to get to your first paying customers. How do you remember this first error? Where you like really running out of the funds? How much money did you have to to get to the first bank customers?
David Čaněk
Um, we not very much. I mean, we were very, very, you know, we had, we were very lean, we weren’t just, you know, a team of maybe seven, you know, in 2012, we were a team of three in 2011 2010, your team of five 2011. So, it was you know, we were very, very efficient. And I took I took a loan from a bank and use, you know, the, you know, my apartment, my family partner, as collateral. So, you know, me and my wife, you know, we were like, you know, I told Well, I didn’t share all you know, I had to get her consent, obviously. So I did that by you know, I didn’t go at great length. What’s going to happen if, if this doesn’t work out? Luckily, it did. So, so yeah, it was it was it was a long from from from a bank, and also had a, had a friend became a shareholder. Who was kind of an Yeah, a business partner. And got some cash from from that friend of mine. And the bed, you know, this wood. The funds were ridiculous. I mean, it was, it was really, I don’t know why I would have to guess but, you know, few 100,000 $100,000. You know, maybe $300,000. Need maybe four but fully enough more than we used to
Andrej Zito
Do you remember your first client?
David Čaněk
I think I do. I think it was a it was a Russian translation company. And it was our, you know, Russian Russian translation. We had a, we had a, we had a quite a few translation companies as customers. And initially, a lot of our first customers were transition companies. And here’s a hopefully funny story. So, so, Joseph, our first Joseph Grabowski, you know, some may know him or not, but it’s well known in the localization industry. So, you know, greetings to Joseph, and Big thanks for forever, for everything he did for memsource. But he, you know, he spoke, he spoke a lot of languages, he also spoke Russian, but he couldn’t, but he, I think he learned Russian when he was living with his parents in Russia. Because they were, I think, diplomats or something. And, and in the end, he was able to speak, but he wasn’t able to read or write well, I, I actually learned Russian at school, and I was able to read and write, but my conversational skills were not great. So when we would, we would have these calls, he would be, you know, on Skype, and he’ll be demoing Memsource, and then he’ll be speaking, and then sometimes they would write something via the chat feature, and he wouldn’t be able to read it, because it’s been several weeks, right. So I would read it for him. We were, you know, this is how we were kind of attacking the Russian market. I was reading and he was speaking.
Andrej Zito
First good reminds me i’m not i’m not sure if you know, but there’s like a chick chick or Slovak joke about the cops, you know, like, why they always go together into it, so that, like one of them can speak and the other one can take notes.
David Čaněk
Yeah, and there’s actually a third one, you know, a third column that needs to, you know, kind of keep an eye on the two dangerous intellectuals.
Andrej Zito
In the beginnings, Were you the one doing the selling? did you have to learn that seemed to me, like you’re more oriented on the tech side.
David Čaněk
Yeah
Andrej Zito
Is there something you would do differently? Now, if you’re starting a company again, maybe, let’s say in this era, that we are now?
David Čaněk
Yeah, well, first. I, you know, I would I would I still start a company. And, you know, if I yeah, I think I mean, I would do you know, many things differently, I guess, bootstrapping a company is really hard. So I think I wouldn’t bootstrap a company again, because, because it’s really hard. It’s really hard. It can, it can be extremely hard. And, and, and especially, you know, SaaS companies where, you know, you need to be operational 24 seven, and you don’t have all the resources, you don’t have all the on duty people and and not everyone is willing to wake up, you know, at night when there is an alert or some something critical on the application. So there’s things that may, you know, so yeah, I think, I think if I was to start another business, which is not something I’m you know, I’m still very much involved with with Memsource, so I wasn’t really thinking about it. No, but you know, they’re obviously I would do a lot of things differently.
Andrej Zito
So over the years, you have grown to I think you’re at 110 employees right now, right?
David Čaněk
Yeah, we have a team of about 110
Andrej Zito
Probably going to grow even further right with the investment. You just got
David Čaněk
It. It will It will I think it will,
Andrej Zito
Has your vision of the company changed over the years?
David Čaněk
I don’t know. Not Not Not very much. No, I mean, it certainly. Obviously, if you were asking me, you know, throughout, if you had asked me like that, you know, this entry, you know, we’ll be taking place in 2010. I don’t know what I would answer and if, but, but in retrospect, I think this is how I remember it, at least, is that we, we had kind of two milestones in mind. And, you know, very high level, one milestone was, how it was to bring all that functionality that was available in those desktop tools, like trade offs, you know, and other tools. How do we bring into the cloud? So this was our Okay, you know, how do we, and we would this would be the kind of what we would hear from from our early customers, oh, try this has this feature, when are you going to, you know, provide it to. And so our first milestone was to bring the kind of features that were needed the functionality to a cloud based platform. And then our next big milestone was, how do we once we have reached, you know, once we have it in the cloud, you know, the kind of what’s requested or what’s requested? How do we introduce innovative features that are not available in trials, and that are maybe some of them data driven. And so these were the two big milestones that we had, that we planned, you know, when that we this is our vision from the beginning. And, and I think in 2017, when we started our AI team, this is when we seriously started. innovating, right? Because Because everyone’s in the cloud now. And so now, it’s not an innovation to provide a cloud based platform or software.
Andrej Zito
Before we start talking about the innovations. I’m wondering, because you decided to start in the cloud. And you were sort of playing the catch up game with the desktop tools to bring the features to the clouds. Were you the first cloud TMS solution? And then the others were playing a catch up with you to bring their features to cloud?
David Čaněk
No, we were not the first. So I think SDN was before started before us also word v. And maybe someone else but so we were we were not the first for sure. But we were you know among the among pioneers.
Andrej Zito
What do you think, is what differentiates Memsource from the other TMS tools? For right now?
David Čaněk
So, Good question. And, you know, I’m sure a number of things that are different about Memsource. When. So, you may, you know, as we as there is, and there’s maybe just kind of the product level obviously. So, you know, how is the memsource product different than, and I think here you can see that some of the cloud based TMS is did not hop on the AI bandwagon, you know, so they did not start these AI initiatives, not not even as of today. So I think this is probably the main differentiator, and it will be more and more apparent going forward, because it’s going to be very hard to catch up. Because it takes quite some time and quite some changes in the overall architecture of the product to make these things possible. In. Yeah, and, and then I think, also, you know, what’s different was, maybe the company culture could be, you know, something that, you know, every company will have slightly different or very different company culture. So I believe we have we have in one way or another, some kind of, you know, unique company culture, maybe I’m wrong, but
Andrej Zito
Yeah, yeah, we will talk about the company culture a little bit later. So let’s focus on the innovations and AI. You mentioned the term bandwagon. So is that what you were thinking in 2017? When you set up the AI team that you heard AI from everywhere, and you feel like, okay, we need to get an AI? Or what was the initial ignite for you to actually create a dedicated AI team for Memsource?
David Čaněk
So back in 2017, I think, yeah, probably, you’re right, that it was it was kind of a kind of an important, maybe even marketing kind of keyword. But as I said, it’s been our, you know, this was our plan from the beginning. So I wasn’t thinking about it as a as how do we create a marketing? wave or, you know, splash or something? So, so it was no, it was, it was the plan from the beginning, I think it was just the next kind of the, the next, the next really means of innovation for this technology, right? It’s, it’s, it’s technology that allows you to do certain things that are not, there are not easy, or they’re impossible, through other technical means. It’s just a technical means. And then you have to, you have to, you know, think very hard how to make it useful for your customers, and you may need to provide features that are useful, and your customers really don’t care if it’s AI powered or not. Maybe actually, if it’s AI powered, they may even have more questions like, Okay, how, how about my data? Right, and, and so you have to have very solid, you know, legal and privacy and other, you know, overall framework to how you do make sure you’re, you’re, you’re complying with all kinds of regulations. And, and, and that you’re not doing something, you know, contradictory to, or something something mean, right? Because you don’t want to want to do good things for customers. So yeah, so it’s just a, it’s just kind of a tool, I think, but it’s, it’s a very important tool that allows you to do things that you wouldn’t be able to do. Otherwise,
Andrej Zito
When you set up the AI team. Did you already have an idea of what they will be working on? Like, what will be the exact features? Or did you just create the theme and let them think about what could be good for the localization community?
David Čaněk
So the latter, we didn’t have, you know, we didn’t have an exact idea, because also the AI team wouldn’t need to be part of that process of, of, you know, and providing input and, and then getting excited about the things that they want to do. So we didn’t want to kind of okay, this is our roadmap, this is the things that we’ll be doing. Now, let’s hire a few AI engineers. So, so the main source, and I think this is one of the kind of values that we have that we want to we want to make sure that every every team and every individual is really able to, to decide, you know, what is what they want to work on. And what what makes sense, what doesn’t make sense. It’s feasible, it’s not feasible. And so this was the same with the AI team
Andrej Zito
Were non translatable as the first main thing that came out of the AI team?
David Čaněk
Yeah, this was the first feature. So we we were thinking, Okay, what is going to be, you know, what is going to be our first feature, that is, you know, going to be unique, needs to be unique. But it may be our first feature shouldn’t be like, you know, overly ambitious, it should, you know, we should be able to deliver it relatively quickly. We want to have the results quickly, and we want to, you know, we want to meet we don’t, maybe we don’t want to tackle that hardest problem that we can think of with our first feature. So it worked out really well. So we we saw that actually, when we looked at our head and the Messrs, you know, the day our net source, we saw that there’s quite a bit of segments that are just not translated. And, you know, typically, these were short segments, they may be a sequence of numbers, some product codes or brand names, whatever, you know, or a URL and email address and there was a segment and so we in and so I think it was it was about fifth Yes. 15% of all segments in our data, which seems quite a lot. And actually, it’s in when you look at the word count, it’s much less, it’s 4% of the overall word count and the segment’s are very short. But still, you know, it’s something. And so we we develop this algorithm that learns, from data strained. And and it was, it’s able, it’s able to identify segments that don’t need to be translated full segments. Here, we’re not looking at like, specific strings or phrases within a segment. So it’s just the full segment, is this segment non translatable? It doesn’t have to translate yes, no, actually, we also provide some probability score. So and we got, by the way, we got we got this patented in the US. So it’s a US patent. So it was it was indeed unique. And it was, this was our first feature in Yeah, we probably the launching in 2017. From that mistake in 2017, beginning of 2018, Archer,
Andrej Zito
Still trying to imagine how the whole process, when did you start with looking at the data that you already have a Memsource, and then the idea came up? Or how does an idea like this originate?
David Čaněk
Yeah, I think this is how it how it came up, we were looking at the data, you know. So it is, it was this process, and I think this is a useful process to make sure that whatever features you develop, they’re kind of based on on the data of your, you know, in Memsource, because if it, you know, if if you, if you develop something that’s based on some public data, or someone else’s data, you know, it will probably mean, it may actually not work for our customers, because they have maybe other use cases, you know, and so it has, it’s this, it’s important, you know, to use. And this is this is important to realize, you know, and I think customers don’t always, you know, when you think, oh, you’re using my data, well, if we don’t use, you know, your data, it will not work for you, because it will be someone else’s take out. So, so this is this is an interesting, you know, an interesting kind of, attribute of, you know, of the whole machine learning,
Andrej Zito
How was the adoption by the customers, when you first rolled out the feature? Did you have something like a beta launch that you were pitching this feature to some selected customers of yours? Or did you just decide to do it, and then publicly roll it out for everyone?
David Čaněk
No, there was definitely a beta. And most, I think most of our most Also, our newer AI powered features, you know, we first launched them as beta, and we try to, you know, we look, we, this is very much about experimentation, and unique, and it’s about learning, and you don’t get it right the first time. Typically, even if you do a lot of testing, or you may get it, you know, you may end we’re trying to do it typically are able to get really good results. But that doesn’t mean you don’t need to iterate and improve. And I think as you know, as we go ahead in this direction, we see that it’s very important to work with, with customer, our customers that we have very, you know, very kind of very on their use cases, understand, you know, how they, how they use those features, learn from them, learn from the actual use cases, and work with them extremely closely. And we have you know, we have a number of customers number of proof of concept that we’re running on those, all those kind of latest features is extremely helpful. And it’s really critical to make sure we get this right at the end at the same time. And this is I guess the kind of the product editing that you have to have this has a few get, you know, a few get feedback that’s not completely you know, not all positive. At first, you know, you shouldn’t not should not discourage you, obviously, because people are conservative naturally and there’s something new then They need to know something some time also to, to kind of get get used to it. And maybe you need to ask them once, once more after, after two weeks or a month?
Andrej Zito
What is that? What is the first stage where you start interacting with customers about new features, especially when it comes to AI? Do you also involve them in conceptual discussions? Or is it only once you have something like real that they can see?
David Čaněk
So, there is, you know, if we can get some inspiration, feedback, and then use it to provide a new feature, we’ll do it.
Andrej Zito
I think the next thing that significant AI thing that came out of he was the machine translation quality estimation, quality estimate.
David Čaněk
Yeah, exactly. Yeah, this was our, I think this was our second major IVR feature of the non translatable segments. And this was already more ambitious. So it’s, this is a hard problem. And, you know, few companies are trying to solve this problem and not just memsource. But it’s, it’s the question of what’s the quality of machine translation output. And so in this case, we’re trying to estimate the post editing effort, effort that it will take to post edit a segment that’s pre translated from machine translation. So you know, is this Do I need to post set it a lot, a little bit, or not at all, and it’s actually perfect can go without any post editing, you know, be published, or whatever. So this is, this is a very important feature, I think, for the localization industry, because if you can rely on a score, that, you know, quality score that tells you, okay, these segments don’t have to be post edited, these need to be a little bit more certain. And these segments, while the machine translation quality is so low, don’t even bother the translator with it, because he or she will just waste their time reading through it, you know, so. So if we can kind of make these distinctions and and make these decisions actually in the workflow, then I think it should make things more efficient, and it should make, I hope, everyone’s life easier.
Andrej Zito
Yeah, when we’re talking about non translatable, as you said, the idea of that came from the data. I’m wondering how what data did you use for the MT QE thing? Because right now, I cannot think of anything?
David Čaněk
Well, it’s it’s again, it’s, it’s mostly it is the data that the direct source data so so we have, you know, we have a lot of customers that do post editing, right. So when you do post editing, you have the, you have the source segment, you have the machine translation output, and then you have the post edited. Target. And, and so you have these three. And so this is the data.
Andrej Zito
So trying to wrap my head around this. But how does the score count? I don’t know if it’s like your proprietary information, but how do you estimate it is based on like, previous similar strings that were posted at it?
David Čaněk
It’s it’s really I mean, look, there would, you could have a whole kind of technical discussion on this with our with our engineers and AI research researchers, but, but they’re, yeah, they’re, you know, we we basically look at What edits were made? And, and, you know, what was the source? What was the MT raw MT output? And what was the final kind of post edited target? And and this is the basis for the for the for the training.
Andrej Zito
think I’m getting it now. So when you talk about the score of the machine translation, this reminds me to translation matches the percentage that we get from the analyzes, do people actually start using this quality score to maybe pay to translate is more or less, depending on the quality score?
David Čaněk
I look, I think it’s definitely something that, you know, I heard, you know, some customers concern this, then I’m not sure if if in the end, they implemented something like this or not, I think it can, could be potentially used. Obviously, there’s some, you know, there’s some things you need to keep in mind. Like, you know, it, it is it’s not 100% accurate, right? I mean, just like, I guess, just like translation memory matches are not 100% accurate. Sometimes you have a real mistranslation stored in translation memory. So I think I think it you know, when you so I think this is still in kind of in evolution, so when we were we actually launched way early, I think it actually 2011 11, we launched the post editing analysis that actually provided you with the post editing score. But this was not an estimation, this was not before, this was not before the post editing to play, this was after that post editing took place. And so you didn’t know until it happened. And you didn’t know, oh, this, this is really low quality, MT let’s not even share it with a translator, or distance, this is actually perfect. It just no poor setting meeting. So you just had to push everything to for post editing. And then then find out what was posted. And then you got the score. So this is this is the difference. And I think it’s it’s convenient to know before and also for pricing, I guess for pricing and budgeting reasons you want to know what’s what it’s going to cost the for the setting? And actually, yes, you can use this for this for you know, for this use case.
Andrej Zito
I didn’t think about it until now. But if you think about it, it’s very similar to the TM matches, right? Because so far, I think I’ve only come across having like a fixed rate for post editing. So whatever you post that it, regardless of the quality, you still get paid the same. But if you guys can estimate or anticipate, like, what is the actual effort that will go into the post editing? It might make sense.
David Čaněk
Yeah, I think it could make sense. But then I’m not, you know, I’m not a translation company owner or, you know, I sigh I don’t want to make kind of, I’m not able to make these decisions. But I think it definitely came across a lot of kind of questions and an exploration in this area. I know, it’s still a relatively new field. I mean, it’s, I mean, we launched it in 2018. It’s not that new, but I think it takes why obviously what what you can see is that one thing is to launch a feature, but then in it but then another thing whole different thing is to change a business process, you know, so to change a business process, it’s really not to be underestimated, right, it’s really not easy. It takes time when it takes all the parties to agree on something new and, and so that this This, I think, you know, we will see more and more of maybe mt QE being used for also budgeting purposes.
Andrej Zito
This I think also ties to Memsource translate, if I’m not mistaking. Is there a correlation with the MT QE and Memsource translate?
David Čaněk
Well, commercially Yes, but not not technically. So. Not Not Not in terms of the technology so. So when you use Memsource translate to Memsource translate is a feature of also AI powered that that select The most optimal machine translation engine for for for your document. No. So it’s, it works at a document level. And, and it so we’re trying to, you know, with mt QE we’re trying to provide machine translation quality estimation at segmental level with memsource translate, we’re trying to make sure that if you decide to use an T, you use the best available and T engine, you know, not, not not just one that you think may be as, as good. And so it’s it’s very the two features. So so when you use ministers translate, we also, you know, we we also activate mt QE for you so that you can use that too. But so it’s kind of bundled together, I guess. But, but in terms of the technology, we don’t use and MT QE to power, the decision making algorithm of open source translate?
Andrej Zito
I see. Yeah, yeah, that’s what that’s where I was wrong. I was thinking that maybe you first do the MT QE using different engines. And based on the accumulative score of all the strings, you select the engine, but you said it’s based on the document level? So what is the logic behind? How does the AI select the right engine? Based on the document? What does it analyze?
David Čaněk
Yeah, so look, if you could do that you could use actually, you could use MT queue, MT queue is also available via API. So you could, you could have this, you could use this in this way. And you could test different MT engines and kind of look at their quality scores, and kind of in an aggregate way in which, you know, and then decide which MT engine works best for, for that specific documents, you could do that. But, but obviously, you know, downside is that it will be a little bit costly, if you want to run it against 10 MT engines or something, right. So so it’s going to get a little costly at some point. In so many search translators not doing this in this way. And, and again, when source translate is powered by data, so, you know, it is it is, again, powered by the data source. And, and the the, there are three, that our newest version of websites translate and take, you know, make the decision based on three inputs or three elements. It’s the source language, the target language, and it’s the domain of the document document. Question. So the domain could be I don’t know. Automotive Life Sciences, health care, you know. So, so this, so, you know, we’re actually, we’re looking at a document where we have some, you know, content understanding technology that tells us, okay, this is, this is automotive. And it’s, it’s English to German, and then say, okay, for automotive English to German, as of today, as of our current understanding, because it changes all the time, you know, there’s all these continuous updates to the different mt engines that are getting improved all the time. So as of today, as of now, this is the optimal engine, and it will be that this engine will be will be
Andrej Zito
explain to me as an MY noob does it make sense that there are so many MT engines in the world? Do they have do they serve a specific purpose?
David Čaněk
Yeah, it’s, it’s a good question. So, you know, we’re actually when we were starting the AI team, you know, we were, it was a, you know, definitely a kind of a big question for us. So, do we develop and an MT engine, you know, Memsource MT engine. And, in fact, we had people on the team that, you know, had that experience with developing and training T and developing MD technology and, and, and so it was a real, real question, we decided not to because we felt like there were just too many MT. A lot of MTproviders, and, you know, Microsoft and Google and Amazon and, and just so many small All our players and and, and so it’s, I think it’s always hard to compete with Google. And it’s, I think, very hard to compete with Google and Microsoft and Amazon, and everyone else in this space. So we thought, well, let’s not go in this in this direction. I think I think there’s a lot of MT players, because they’re new technologists, exciting technology, is there you see all these incredible improvements year on year, you know, in it, which we see, we see that in our data, you know, there’s incredible improvements in quality here, year on here. So So we, yeah, so, so I agree, there’s a, there’s a lot of there’s a lot, but you know, many of the MTproviders may be focused on us on on specific language barriers, maybe you have some, you know, maybe Japanese and D providers are really, really good at Japanese to English, you know, English Japanese, and, and then you have, it’s, we have a lot of local, local mt engines that have access to local data, and really spend a lot of time on, on, on, on refining that data, that they can produce actually amazing, amazing quality. And so the idea of message translate was rather well, let’s not compete with these guys, let’s partner you know, and let’s help our customers find their way to the most optimal technology that, you know, is so promising, which is machine translation and, and make it, you know, incredibly well integrated in memsource. To make it completely seamless, you know, you don’t have to test different MT engines, or which MT engine is good and isn’t good enough for my content, doing good leverage, you know. And so we this is really the plan that we have.
Andrej Zito
Yeah, yeah, I think I think it’s really smart. Not sure if I can paraphrase this in English. But you know, our Czech proverb is that if there’s a gold rush, you should be selling the pickaxes and shovels instead of going to look for gold. So I think that’s what you guys are doing. And I really like it, it’s a smart idea. I’m wondering, because you said that you’re partnering with this MT engine, so doesn’t mean that you go around and look for the MT engines that you integrate with Memsource? Or can like people apply? Or do I, as a user? Can I somehow, like, at a certain engine to the pool of the MT engines that you select?
David Čaněk
And it’s a good question, so, so look, we, when we, when we launched this feature, we had to, we started with just Google and Microsoft, you know, as the, as the two engines that were available, and we were just picking out between the two, that was, you know, early on, we added more engines, more generic engines, so you have a feel about it. And also, what you can do is you can add your customizable engines to memsource translate, and that’s, that’s a feature that we’re actually just about to launch public to the public. So it’s, it’s, it’s available to a few customers already, but we’ll have the general release within mid mid July. So pretty much as we speak. And until you’ll be able to add your your customized engines like you know, Microsoft custom translator or Google auto ml or other other engines that have this that support this capability of being customized and then you can kind of combine it, do you have it, you have it in a pool with those generic engines. And obviously, those customized engines are only available to you so, so you can mix and match your customized engines with the generic engines and we do automatically we do the matching, you know, which engine is optimal for for content, if and if you want you can also decide not to use some engines at all, you know, some generic engines, maybe you don’t like a specific engine for for whatever reasons, business policy or something. And, or, or you can also set some additional rules, you know, that for which language barriers who want to use those customizations So, so it’s there’s, you know, quite a bit of functionality that We’re launching around this this feature set.
Andrej Zito
So we talked about a lot of the MT innovations and the latest things that you guys develop. But I’m wondering if I’m a new company that just wants to start with MT? What do you think would be the first steps that I should think?
David Čaněk
Well, this, I think it really depends on the specific use case, you know, are you are looking to do raw machine translation, or are going to post edit the content or, you know, so that I think we would need to kind of be getting more a bit more specific around those use cases for real useful recommendation. But I would certainly, I guess, I would certainly recommend to first experiment and near not to invest too much, you know, and use maybe, SaaS version for sure. So you can subscribe and then decide if it’s going to work for you or not.
Andrej Zito
So all your AI innovations were centered around Mt. Not all, not not not all, not all. But but the majority.
David Čaněk
So we had, we had, you know, in fact, the first one, the non translatable segments, I think it’s not really MT, we also launched a feature that, as it’s kind of used in the background, we haven’t really publicized that much. But it’s, there’s a feature in Memsource that recommends relevant linguists for you know, your own linguist from your own user user pool for for, you know, when you want to assign a language to document for translation, then we would we would recommend that the, the most relevant linguists, and I think this is also partly powered by AI, some of that functionality when there is sufficient data. So, AI is in the background. So there’s, there’s a few things we we develop that are not related to IoT, but I think it’s a it’s a, you know, definitely true that a lot of this innovation that that we’re trying to do, they were working on, it is very much related to Mt. As you know, we believe this is where this this is where, you know, really, the industry is obviously gone.
Andrej Zito
Yeah, so there was actually my question like, what do you think is the the future with AI, in your perspective, like, maybe beyond MT
David Čaněk
Beyond MT? So I think we still need to, we’re still, I don’t know, what’s beyond MT, I think is, I think there’s still a lot of work on on the actual MT technology, and you can see all these big tech players and a number of smaller players and each players investing heavily into improving the core technology, really the and, and, and I think the smaller providers are, you know, also, it’s tough, and they also need to invest a lot into scalability. So one thing is to is to, is to get the Mt. quality that the linguistic quality to a certain level, that then another is to make sure it works under load. And that if we integrate it, you know, that it can, it can really, it can really support the traffic and the word volumes, it’s not always the case. So there’s, there’s a lot of work on tiesto. But I think there’s equally there’s also very much a lot of work to to automate the whole other workflow around MT the training, the quality estimation, right, the selecting the optimal MT engine, you know, you can you can think of it just like you know, you want to select the most appropriate to the most qualified linguist to do the translation. So in just, you know, very similar you want to, you want to select the most optimal MTengine to do the translation and, and it’s kind of parallel to that. All these big tech companies, obviously, they have a lot of very busy which is, you know, fine tuning and improving adding more languages to their MT technology. And, and, but there’s a lot of work also in, in in making sure that mt works in the, in the localization ecosystem as a lot of work. And to make it really useful for enterprise customers to make sure that when machine translation is used in, in an enterprise context that it really meets the kind of enterprise grade quality requirements. And I think there’s, that’s one thing, another thing is to, you don’t want to turn your localization team at a, you know, in into project managers of MT and you know, setting up MT engines customizing MT, and just figuring out why this data set doesn’t load into the, you know, MT system to for it to be customized, there’s just tons of manual tasks around MT currently, to make it useful, so that you have this paradox of this incredible, you know, latest kind of MT technology array that’s really state of the art, and then you have all these numerous myriad of manual tasks around it, then then No, no one thought of automating so this is what we’re trying to tackle. And this is this is, I think that the kind of next milestone that we have to make it a completely seamless, you know, to use mass, or she don’t have to worry a lot about, you know, which end to end and how to train areas that does it perform. And, you know, so this is, I think that still a lot of work there. And then what’s beyond, you know, it really the the ultimate goal is to provide more automation. And I think full automation is probably is the ultimate goal, but I don’t know if it can ever be reached, probably not. So let’s talk about the second differentiation of Memsource that you mentioned. So we will part away from the technology for a while. So let’s talk about company culture. I’m wondering company culture is such a fancy word, especially recently, did you think about company culture when you started the company? Or was it just like, these are bunch of friends that like to work with them, let’s create something. Now, this was definitely one of the main drivers for starting a company. So I wanted to know when to start a company and work in a company where I had some influence on the company culture, so that, you know, maybe, because I think the trouble with being an employee is that you don’t have the control about certain things. And maybe you have, you know, great boss, and this is how you get hired, and then you then that that person leaves and then you have another boss, and maybe that person is, is just completely different than and, and, and is, and so I felt, you know, I felt like if I that, you know, Memsource should be, you know, should be a place where people share these values that I share. It’s not for everyone, I think, you know, it’s not saying Memsource is perfect for everyone, but I think it’s, it’s, it’s built around company values, probably like, like most companies, right?
Andrej Zito
Yeah. So what are the values for you?
David Čaněk
So, in when we, when we hire people, we try to, we’re looking for someone who’s obviously very professional. Ideally, senior, can be also Junior, but we need to meet to see that potential, obviously, in them. So we’ve obviously hired a lot of junior people that grew incredibly quickly. And that that’s, that’s great also, but we need to, we want to, we want to so what’s really important is that they they like what they’re, you know, they like their work, they like what they’re doing, they’re not there just for the salary. And then they enjoy working. They, they, they like to you know, they really they, you can see that it’s, it’s something that they’re really interested in these challenges that we’re trying to solve. So this is another important aspect. And obviously what’s also very important is their character. You know, we they need to be able to work in a team they need to have their you know, integrity know, things Like I was, you know, I think just normal things that you want to be able to rely on your colleagues, you want to just be need to be nice people, right? And so, so yeah, you have, you know, this is you kind of have this ideal workplace in mind with these ideal colleagues. And this is how you’re trying to build the company, it may be doesn’t always, it doesn’t, it’s not always easy. And, and sometimes you, you know, it doesn’t work perfectly in all instances when, you know, when we made some mistakes, you know, hiring people and, and so on. for different reasons, maybe we read that experience, whatever, but, but this is, this is really the kind of goal that you have. And I think even if you grow, grow the company bigger, you need to you need to, you know, you can you can think, okay, now I’m 100 plus people, I can just hire anyone, or I can hire people that are, you know, you know, here’s a great, you know, marketing manager, but But okay, he or she doesn’t kind of share the same values, but you know, it’s going to be okay, so it’s not going to be okay, I think the values take precedence, always.
Andrej Zito
How does your leadership fit into the company culture? How would you describe your leadership style?
David Čaněk
So it’s probably look at, I’m not sure if I’m able to describe it myself? Because, because, I mean, I can try and but I don’t know how, how accurate is going to be. But I think it is, I probably, again, refer to these values. So you know, I, I am trying basically, I think, the, one of the main tasks, or one of the main kind of, I guess, tasks, or an important role that I have is to make sure we we we stay true to these values. And we, we don’t divert from the values, and then that this is, I think, one of the really important things.
Andrej Zito
Somebody ever told you that you broke your own values?
David Čaněk
No.
Andrej Zito
Do you think do you think you did it?
David Čaněk
Oh, you know, I see, you need to, you know, so there’s, you know, so, you know, life has not like, is not black and white, obviously, so. So the values are, you know, something that you strive to, but you’re not perfect, right? And, and, and so you certainly make mistakes, and sometimes you make a hiring decision that you later regret, and these are kind of the mistakes that are painful, right? Because you have to deal with them later, and people get affected. But, yeah, obviously, that’s life, you have to this is something if you’re if you’re, if you’re trying to build something, is, you know, there’s going to be always some mistakes you’re going to make on the way and also, because you I was not born as a CEO. And most of our managers were not born or nuts didn’t start as managers in Memsource and had to learn and successfully actually grew into these roles. And so most of our, almost all of our engineers, you know, that I started the company with so the early kind of hires that I made the serve almost all of them are, you know, top managers in the engineering group that we have and it’s been amazing to see them grow and into these roles
Andrej Zito
On your CEO journey. What do you think was the biggest failure that happened to you and what did you learn from it?
David Čaněk
So the biggest the biggest failure look, I think, I think there’s been maybe a few on the product side I think side of it, I don’t know if anyone kind of remembers it still but but we started when we were when we started developing the the translators workbench. Our first attempt was to use Microsoft Word and develop a plug in I think it’s called it was called added add in memsource add in And that really is a plugin. And, and we invested a lot of time into this. And the reason why we took this path was that we had very limited resources and to develop a translators workbench is really very resource intensive, doesn’t mean, maybe it doesn’t look like that. But it’s really hard. It’s really hard. And so we took this to this path, and we spent a lot of time on this, and then it wasn’t the right, the right way. And, and we had to completely, you know, ditch that. And then we, we develop the main source editor for desktop and message editor for web and then much later associate editor for mobile. And so this was, this was a big mistake, you know, probably one of the biggest kind of one of the biggest product mistakes that we made that very early on, and we were able to recover from it. And then I think the, you know, maybe the mistake that, you know, I, I regret the most is when you hire someone, and it’s the wrong the wrong person, or is the wrong hiring decision, you know, it’s, it’s not the fault of that person, like you made that hiring decision, right, as a manager, or as a CEO. So these are the these are the outside the mistakes that I would that I would definitely regret, but I’m not sure if they can be prevented 100%? Probably not.
Andrej Zito
What is something that people seem to misunderstand about you?
David Čaněk
Well, it’s, no, it’s, I look, I don’t know. But there’s one thing that I can remember. And again, it’s kind of tied to our very early years, when I was, I would, I would go to all these localization industry shows, right, all up all around the world. And I would go with with Joseph Boesky right away, when mentioned earlier, who was really the sales, the first sales person in memsource. And, and, and later our head of sales. And so we would go with him, and I would, you know, he, you know, he’s very, he was, and he’s very extrovert. So, he’ll do a lot of the talking that I would maybe explain some of the more like, the technical things. And then, you know, and then we, I think, a year later at a conference, you know, we bumped into this person that this customer said, Well, you know, Joseph, you’re really great, you know, you’re great sales guy, or, you know, and, and here, you’re the decoder, you know, that, you know, pointing at me, you know, that this guy that, you know, it’s also a really, really knowledgeable quarter that you have, that you brought along. Thanks. So, you know, this was maybe one of the first things that I can think of.
Andrej Zito
Do you think that people look at you as a, as a tech guy as the coder?
David Čaněk
It happened to me a few times, so I guess so. But yeah, I never I am not a I’m not a coder, you know, but I think, I think, and I never told anyone, you know, so. But I’m dumb, and I’m the product guy. So so I was, you know, the, the product guy for many, many years and memsource. And before we had any product team, so. So it’s not that far. Right. So you, you know, if you know, their sales and marketing, and there’s product and engineering, so I was kind of on that side. So in that sense, maybe, you know, they were right.
Andrej Zito
What would you do if there was no Memsource? Of which is we just started a company or?
David Čaněk
Yeah, I would start Yeah, yeah, I would start it, you know, if there was, I would have a plan B in areas, another company, another Memsource.
Andrej Zito
At your position, or like, at your age and experience? Would you ever work for someone? Or do you always have the drive to be the leader? and be in charge of decisions?
David Čaněk
So no, no, I don’t have to be always kind of, you know, to dominate everything. No, I think at Memsource you know, we work very much as a team. And so it would be a misconception thing that, you know, the person answers that runs everything. It’s true that I found it in Memsource, and I took this risk, you know, and, but but it’s, it wasn’t certainly there were you know, we’re 100 plus people so, so I work together with many, many, many colleagues, you know, either from management or from across the company. So I like you know, I like to I like to work I like to work with with other stuff.
Andrej Zito
What are you curious about, right now?
David Čaněk
I think it’s this new state with Carlisle. Yeah, no, I think look, I’m I’m kind of if you’re looking. So, you know, I’m kind of outside of work, I really have very little to share. I’m kind of if you’re not interested in localization, probably the most boring guy that you’ve met, because I don’t have a lot, a lot to share. Because I, you know, I don’t, I don’t really, you know, I’m sure that, you know, there’s going to be time when I’m going to be doing other things too. But for right now, it’s very much focus on what we’re doing at memsource. And it’s very exciting. And, yeah, that’s not only what I’m doing, like I said, I definitely my family is very important to me and my kids and my wife and everything. But But, you know, obviously, that’s not the topic of this conversation. But it’s and in fact, I can hear my my kids are just getting back. Now may hear some background noise.
Andrej Zito
Yeah. So you mentioned the Carlisle Carlisle Group, right, that you’re curious about? Are you expecting changes within how the company is running? Or did you talk about some plan together? Or are they going to be more of a passive shareholder?
David Čaněk
So yeah, so the Carlisle Group is the majority shareholder, but, you know, I remain remain and, and except for one of the previous Memsource shareholders, everyone remains in board. And, and remains in management remains a shareholder. So, you know, the Carlisle Group is not management substrates, and that needs to be the management. And, and so there’s not going to be, you know, obviously, time will tell and, but, right now, the plan is to, is to kind of to continue what we’re doing, just doing it maybe even better, if, you know, and in an accelerate certain things, and suddenly, what, what I want to what’s really important is that we add more senior talent. So really, if anyone is listening, you know, we’re hiring, for sales and marketing, especially. So, we we need to, you know, I as a, as a CEO level, you know, I worked as almost like, you know, they all these roles and wearing these different hats, you know, head of sales and marketing head of product. Well, luckily, we have now, a Chief Product officer, we have a chief marketing officer starting very soon, and but there’s, you know, I need to, I need more help from others. We need more more leadership bandwidth. Definitely.
Andrej Zito
Yeah, I still have, have it on my backlog to create some videos about memsource that you guys gave me the academic edition. So that’s going to do something? What do you think is wrong with our industry?
David Čaněk
Nothing? Nothing.
Andrej Zito
You’re the first person to say that,
David Čaněk
Like, I think there’s this this, like, there’s something there must be something wrong about us. Right? I don’t know, why should we be so hard on us? Like, I think we have a very, I think we have you know, we’re living you know, it’s really exciting, exciting period for localization. Really exciting times, you know, a lot of innovation, you know, sometimes you know, what, maybe, you know, we have these obviously, it’s it’s an industry where you have the the linguist that do the work really write the translation work, and other linguistic work, you have the lsps that sometimes are hard on the linguists, and then you have the enterprise customers that are sometimes hard on the LSB I think, and on us also, right, and LSP so hard, not some linguists are hard on us also, because we maybe, you know, maybe we’re not their most favorite tool, you know, for some of them, although they’re trying really hard. So, so I think sometimes we we it’s a it’s maybe a little bittoo kind of confrontational. And I think we could, we could maybe learn how to work more together, you know? Because the industry is growing. I mean, it’s not like we have you know, the the, the industry is growing, we should be able to share more, even with all this innovation. automation, I think there’s going to be enough work for everyone for, you know, all the kind of experts that we have. And, and I don’t think this work will be disappearing. So it’s just there’s just more and more content for localization. So I think we’re trying to deal with that, with that increasing amount of content. You know, everything is moving digital and, and so this is what we’re dealing with. We’re not trying to get people out of work, I think this is this is this is important, too, to say.
Andrej Zito
What are the things you changed your mind about? Maybe not even like during your memsource? But let’s say maybe compared to when you were younger, like some hardcore fundamental things about the way you view the life?
David Čaněk
Yes, I think just like anyone, you know, people change as they get older. You know, you you, maybe you’re kind of less radical, you know, you you learn that sometimes, there’s different ways of how to say things, right, so that you can say them in a way that that insults everyone, even those who think you’re right, it’s not going to help. So, so I think you at least, you know, you try to be more kind of, maybe not diplomatic, but really, really be more constructive, you know, be more kind of team oriented. And, and I think this is, this is the kind of development that I saw with, you know, at least with me, and tried to work on that a little bit.
Andrej Zito
Are there any absurd or stupid things that you do? Still, even after your development? Yeah, sure. Can you think of anything specific?
David Čaněk
Yeah, you know, I think I still do, I still like to, I’m still kind of this perfectionist guide written. So yeah, I still like to, you know, like to have, you know, I still like to make sure everything is pretty much, you know, perfect, you know, as I would like to have it, but, and then I try to maybe comment on things internally that I then realized, I don’t know, the food understand. And I think so I, and, you know, and I think I still do it, occasionally, I think I’m doing it much less so. So and it’s a little bit kind of related to the growth of Messrs right. And as you know, ice you start with and still with the team of 3040 is still pretty much know everything about the company, when you grow beyond the 80 people, I think in our case, it was really 7080 people, you you don’t you don’t have that visibility anymore, into everything. You still need to keep that visibility into into a lot of things, but not too into everything. And and you need to make sure you have as many people as possible that can fill that that gap.
Andrej Zito
Okay, though, the final words from you, if you could speak to everyone in the industry, what would you tell them? out of them? Please submit your resume to Memsource if you’re seeing this?
David Čaněk
No, I think I think I’d like to this, I already said it, and I can think of anything else. I think, you know, we shouldn’t be too hard on ourselves. I think this is a really exciting industry. Sometimes, you know, you, you hear, okay, this, this industry is too small, or it’s not growing quickly enough, or it’s not innovating, you know, it should be more innovative or whatever. I think we need to I think there’s are some true, like, we need some, you know, we you see that sometimes there are people that are conservative and stick to old ways of working, but I think it is just in any industry. It’s not about localization. So, so I think we need to we need to become a bit more self confident maybe. And then we also need to, maybe we need to work work more together in we need to work more as partners. And this is what we’ll be trying to do and what we’re trying to do all the time. And I think this is really, it’s really been thanks to our customers that funded the growth of Memsource, right. And thanks to the translators, linguists and lsps that are willing to work in Memsource and support us So it’s a I think, and I think we were we were always trying to be really partners and friends with whatever one you know. And and and not not confrontational, although we are obviously kind of wanted to had our vision in mind and we’re going in that direction. And so I hope we can you know, we can we can work together as, as, as industry and the different stakeholders. That’s, that’s my wish.
Andrej Zito
Thank you. Is there anything I should have asked you, but I didn’t?
David Čaněk
No. I think i think i think you you, you exhausted everything you exhausted means is exhausted all the question? No, I’m not exhausted. But, but it’s, it’s been No, thanks. Thanks. I’ve been great questions. I cannot I cannot think of like some some question that I’d like to kind of suggest. free to ask and so.
Andrej Zito
All right. Well, then thank you very much, though. It’s for the interview.
David Čaněk
Well, thanks, Andre. Thanks. Thanks for your questions.
Andrej Zito
I’ll let you get back to your family. Enjoy your time with iItaly. And we’ll talk next time. Bye bye.
David Čaněk
Yeah, definitely looking forward. Thanks a lot for the interview. Thanks for your time.
Bye
