4 weeks ago
March 26, 2021
Language I/O Founders Heather Shoemaker and Kaarina Kvaavik on USD 5m Series A
Language I/O Co-founders Heather Shoemaker (CEO) and Kaarina Kvaavik (CBO) join SlatorPod this week to discuss their newly announced Series A round, channel partnerships with CRM systems, expanding beyond their core multilingual customer support offering, and the future of subscription-based pricing models.
The duo join SlatorPod fresh off their USD 5m funding round, which was announced on March 23, 2021. Kaarina and Heather discuss their journey from founding and bootstrapping Language I/O to bringing on angel and VC investors in early 2021, and their plans for deploying the latest investment.
Heather talks about the challenge of working with user-generated customer support content — which is often littered with typos, neologisms, and idioms — and Language I/O’s approach to identifying and dealing with “problematic” content. She describes Language I/O’s CRM integrations, and their approach to supporting major enterprises with their MT-based multilingual customer support offering.
Kaarina talks about the company’s tiered-subscription pricing models for both machine-only and human post-edited translation, their use of channel partnerships with CRM systems, growing the sales team, and their plans to expand beyond the customer support arena.
First up, Florian and Esther discuss the language industry news of the week, such as the role of the “translators” (read: interpreters) that facilitated the recent US-China talks in Alaska, and the differences in the coverage they received from Vice, South China Morning Post, and Reuters.
Esther shares highlights from video game services provider Keywords Studios’ 2020 financial results. The company grew revenues by 14% from 2019, while its Localization segment — which provides text-based translation for in-game content, marketing, and packaging — was the only business line to report declines. Keywords’ Audio segment grew its footprint in dubbing and subtitling for video games, TV, film, and streaming platforms as Keywords said it has an “active interest” in providing services for the adjacent film and TV market.
Florian applauds the success of his latest SaaS find: Copy.ai. The startup raised USD 2.9m with its multilingual copy-generating offering and is built atop OpenAI’s GPT-3, the world’s largest language model. Copy.ai follows in the footsteps of OthersideAI (and no doubt a swathe of other projects), which raised USD 2.6m in 2020, and attracted investors with a multi-language content generation pitch.
The two also talk about Slator’s newly released Pro Guide on Translation Pricing and Procurement, which dives deep into established and emerging translation pricing models, and lifts the lid on translation and localization procurement practices among buyers — with a dozen charts and graphs to help readers conduct their own pricing and performance benchmarking.
Stream Slator webinars, workshops, and conferences on the Slator Video-on-Demand channel.
SlatorCon Remote May 2021 | $130
A rich online conference which brings together our research and network of industry leaders.
Florian: First tell us a bit more about Language I/O. What are the company milestones so far and give us the elevator pitch so people understand what you are all about?
Heather: Today Language I/O basically enables large companies to use their monolingual or English speaking support agents to support customers in any language. We have apps embedded in the major CRMs and proprietary technology that handles real-time translation in a very unique and accurate way. The legal entity of Language I/O was formed back in 2011 when we had the idea for the company. Kaarina and I at the time were both engaged in other projects so it took a little while before we were able to focus wholly on Language I/O. We really got up and running after I sold my previous company, Cheyenne Technology, and at the end of 2016, beginning of 2017, we did have customers before then, but we were not wholly focused on Language I/O until about that timeframe.
Kaarina and I had both exited from other projects that we were involved in and bootstrapped the company. We did not seek outside funding other than a very small friends and family round until later when we were approached by a Wyoming angel group called Breakthrough 307 who are focused on growing the economy and in Wyoming, outside of the oil and gas industry, which is the focus of the state right now. They approached me at this tech event in Cheyenne and said, we understand you have this company called Language I/O that is SaaS, and B2B, we would like to fund you. I thought that was easy and so we raised the angel round of half a million dollars and that did not close until early 2020. Then after that round closed, that was a convertible note round, we started to go after a larger A round and we wrapped that up recently, $5 million and an institutional VC round. That is going to allow us to grow a lot more rapidly than we have been in the past.
Florian: Tell us a bit about your backgrounds. What got you into this space, language and maybe career before?
Kaarina: I am not sure that I have had a career before the language industry. I have been in localization for probably 25 years but on various sides of the fence so I have been on the client side and I have been on the vendor side. I left it for a while and was on the peripheral side and then came back and basically owned a smaller localization company which was after I had met Heather because we were both working at an internationalization globalization company on the software side so that is where we originally met.
Heather: The first decade of my career, I spent as an internationalization engineer. My education is in linguistics and then I have a master’s from the CU Boulder College of Engineering. I was able to combine my love for languages and writing code as an internationalization engineer, traveling all over the world, helping big companies refactor their source code so their software could support multiple languages. The interesting thing I discovered as an internationalization engineer was that when a company is going global, the biggest challenge they face is not the obvious things such as translating the user interface of their software or Unicode enabling their platform. The biggest challenge was the messier challenge of multilingual customer support so during that decade, working as an internationalization engineer. I started thinking about this problem because there was no technology solution. You had to travel around the world, staffing up native-speaking support agents for every language that you needed to support and it was a problem, it was hard, it was not scalable.
After I was able to exit lucratively from one company that I was part of, I went off on my own to think about this problem and that was exactly when Kaarina contacted me. She was with a localization firm where she was part owner and they had a customer, one of the largest survey platforms that needed this multilingual customer support solution. They needed a more automated way to support these customers in many languages. Kaarina knew I was out on my own at this point and she asked if I could build this solution? Could I automate this process for this company? I said, yes but let us not just do it for this company, let us turn this into a subscription, a productizable solution because I bet they would be interested and we could sell it to all other companies that have this exact same problem. They were absolutely interested and subscribed to our solution and then shortly thereafter, we decided to go to a conference, see who else would be interested in our solution. One of the world’s largest social media platforms approached our booth and were like, I understand what you did for the survey company, we need something similar, could you do that for us too? Of course, we could and it just blew up from there.
Esther: Tell us more about this A round. Why now? How did you connect with the different investors and how did you run the process from start to finish?
Heather: Raising money is a process so after we closed the convertible note rounds, it is expected that you would move on to an A round. We did not have to necessarily, but we know that our space is a growing market. Customer support, the CRM market is huge. The multilingual customer support market all by itself is large enough to start attracting competitors. We knew that if we were going to stay on top of this market and continue to be competitive just in multilingual customer support, we would need to grow faster and so it was time. I will say that it is tough for a woman to raise money in the VC world. I will not deny that. I am already familiar with the challenges of being a woman engineer and the bias you see there, and there was definitely a bias in the fundraising process. I will not tell you how many VCs I talked to all over the country and Silicon Valley and elsewhere.
We had a Boston-based angel group called Golden Seeds that were part of our seed round and we made some great connections there who ultimately connected us with Bob Davoli, who is a well-known VC nationwide. He immediately understood the opportunity. He has already invested in companies in the customer support space and the AI agency space so we did not have to convince him of the size of the opportunity and he did not have any issues with two women running a tech company, which was refreshing. That was not an obstacle for him like it seems to be for others historically. He was full steam ahead. He pulled in other very influential investors into the round, for example, Omega Venture Partners and Gaurav Tiwari who is super influential on the West Coast and Silicon Valley. He pulled in Bruce Clark, who he invests with quite frequently, who is great in our space. Then he also pulled in a guy named Tom Axbey, who has been very successful in the SaaS world and growing companies, companies from our size to much larger.
Esther: Day to day, what do you expect the involvement of the investors to be in operations and what do you hope to get from them, other than just the capital?
Heather: Kaarina and I are very confident in our own skills to do what we have done. I wrote the original code, version one of the Language I/O platform. My background is in technology and linguistics and obviously, I led the fundraising process and Kaarina has massive experience in localization, sales and marketing. While we are very comfortable in doing what we do, we have never taken a company of our size and grown it to revenues of hundreds of millions of dollars. When it comes to growing and guidance in how to scale, they are definitely going to be and already have been influential.
Florian: Kaarina, what is your current sales and marketing approach? How do you get your leads? Now with the funding, what are your plans to expand? Is that one of the focus areas for deploying the funds?
Kaarina: Absolutely and we have already started hiring. Obviously, our previous approach has been a little bit more narrow, but one of the things that we have done extremely well is having good and strong partnerships with our CRM partners so the companies where our software integrates seamlessly. Salesforce, Oracle, Zendesk, we have fantastic relationships with the people that work there. They know us, they know our products and they recommend us to their customers so a lot of them are bringing us into conversations that they are having when their customers are having this pain point that we address. They do it because they know we are going to take care of their customers. We have a fantastic reputation in the industry. We are reliable. Our technology is the best out there. That has been a core part of our strategy and will continue to be a core part of our strategy as we move forward. Obviously, having more feet on the ground, like having more salespeople that can actually do this work helps a lot so that is where we are growing this year. I would say that we are keeping it in the US but we are certainly expanding internationally going forward as well.
Esther: Tell us a bit more about your physical presence, you mentioned wanting to have on the ground sales and marketing. Are you all remote? Do you have an office set up? What is the footprint across the US right now?
Heather: We are headquartered in Wyoming, which is where most of our engineering staff is. Especially post Covid-19 we have been much less concerned about where folks are. We just want to hire the most talented individuals for the roles that we need to fill. Kaarina and I have always worked remotely. She has always been on the East coast. I have always been here in Wyoming and we have not had any issues. As we have started to expand, especially during Covid-19 and hiring salespeople in Ohio, engineers in California, we already have the framework and the culture to support that environment.
Kaarina: I think we are a bit of an anomaly. A lot of previous or earlier investors had a problem with this, like how could you possibly work together? How can you and Heather run this company? Now, it is a given so we were set up from the get-go exactly like this, which means that we had zero downtime when it came to converting to remote work so the way we see it right now is the world is our oyster. If we find awesome talent in rural Nebraska, great because it is the talent that we are looking for, not necessarily the physical location. I have hired people in my team, as has Heather, who we have never physically met and they are brilliant. For us, this was easy, for others I know it was not as easy.
Florian: What are the key challenges in translating this type of customer service conversation?
Heather: Let us focus on chat as a support channel because it is one of the more challenging channels. It has to be instantaneous. There is no time for a human in the loop when we are talking about chat translation, it is absolutely a machine translation. What makes chat especially challenging and for us very interesting is that it is UGC, it is user-generated content. It is messy content. The source content is not good. We start off with misspellings, with every day new acronyms being made up, slang, abbreviations, not to mention the company’s product names and industry jargon. All of this is challenging from a real-time machine translation perspective and that is really where Language I/O shines. We have developed proprietary technology to tackle that style of translation.
Esther: I think you have already begun expanding beyond this customer support area. What new markets or what new content types are you looking at? What is the rationale for that?
Kaarina: When we started doing this, we started on the customer support side and we made sure that we nailed this. We had absolute laser focus on this type of content. Where we are going now is outside of that. We know how to do this, so now we know this can be replicated across pretty much anywhere where there is conversation going on. This could be, for instance, a gaming platform where a bunch of gamers get together in a game and they are chatting with each other. That is one space, there are going to be gamers from all over the world.
Internal Slack channels, we ourselves use Slack internally. Larger companies could have Slack deployed globally so they are going to have all of their employees on this channel. It could be learning management so basically anywhere where there are large bodies of content and conversations going on that is messy. There are other areas. User-generated content, people actually believe that when they get reviews from other users, they are going to believe this much more than if a company puts it out themselves. UGC is used hugely in marketing these days as well. We totally get customer support. We understand it. Wherever these conversations are going on, that is where we are going.
Florian: Tell us a bit more about your CRM integrations and also how you integrate with the MT. How you do the engine selection, if you build your own engines, if you just select the best of the breed. How does that work for you?
Heather: Let me start on the CRM side because that is the easier side to talk about. Obviously, that has been the focus and it is not just that we have API level integrations with CRMs. We have turnkey apps on their app exchanges. From a usability perspective, our apps are amazing. They often support agents who are using our solution, inside of Salesforce, inside of Zendesk and do not even realize it is a third-party solution. It is a seamless integration so that when you are a support agent working in Salesforce and you receive a chat support request, our app just pops up in the workspace that you always work in and it recognizes that your profile language is different from the language of the incoming chat, instantly starts translating back and forth. You do not really even have to do anything, it just happens.
People think of CRM integration as just an app, anybody can build an app, but we put so much into what support agents need from a UX perspective to make it super easy and seamless for them that I do not like to underplay the role of that piece of our technology. Then of course the core technology is on our server. That is where all the heavy lifting happens. That chat app sitting in Salesforce, makes a call to our SaaS server and there are a number of things going on there. Before we even consider translating any UGC, we focus on security and that is a huge piece of our offering. We would never be able to reel in the large murky fortune 500 customers that we do without security being a major piece of our technology. When any piece of UGC hits our server for translation, we are first going to scan it for personal data because they expect us to protect their customer’s personal data. We look for hundreds of different types of personal data across all locales because those patterns are different. When we find personal data embedded in a chat or an email, or just an API call sending us UGC, we encrypt it and tag it so we can reverse that process on the way back in, wherever it came from.
Once we have handled security, then we are of course going to attack the translation piece of it and if it is just an article or a batch of articles, we might send them to our human translators who work inside of our system. There might be an MTPE process. We handle all of that, but our focus is on real-time UGC translation. It is a chat that needs to be machine-translated. There are pre-processing and post-processing aspects of our technology, but let us first talk about the machine translation engine side of things. We did consider building our own engine. There are all sorts of open-source code bases out there that we could have started with, but in our research process, what we discovered is some NMT engines do a better job with certain language groups. Why not just intelligently select the best NMT engine for that language group and other aspects, other features that we have engineered from our data?
Let me just use an example, I do not want to call out a particular vendor as better than any others, but DeepL is an up and coming engine, Systran is a great engine. Google is still important because they support such a broad set of languages so for the edge languages that nobody else supports, it is going to be Google or maybe Microsoft. Ultimately what we are doing is using the data that we have been gathering over the years to intelligently select the best engine, not just for that language pair, but for that type of content, for chat versus email, MTP for articles, and then the industry, the company, all sorts of features that help us make that decision.
Furthermore, we are not training those engines via the traditional process. Again, we have gone down that road. We have a company in, for example, the diesel engine manufacturing space and they were one of our earlier customers and we went down the path of training an engine for them, and it took so long, it was very expensive. You all know how this process works. You have to feed it a huge corpus of high quality human translated content. Then you have to iterate on the training. It requires linguists experts to validate the training. You have to maintain the training for one language pair. Now spread that across the 20 languages that your company needs to support. We decided early on that it was not a scalable process to get company specific accurate translations. That pushed us into building our own solution to this problem where we intelligently select the best engine to get the best general translation for the content. Then we have proprietary technology that is going to impose a company’s preferred translations for problematic terms. It is both pre-processing and post-processing, so I do not want to just say that it is in one direction, but there is a normalization step for misspellings and slang, so on and so forth.
Then there is post-processing after the general translation comes back, some things are imposed ahead of time, others are imposed post-translation, but ultimately we can promise our customers that if you have a product name that always needs to be translated, it will be. It is not based on the mood of the NMT engine on that particular day. We all know how that works. We can promise that certain terms will be translated the way that they should be. What one of the things we are focusing this fundraise on is expanding the machine learning component of what we do, especially in the realm of what we are calling our self-improving glossary. It is one thing when we onboard a customer to make sure that we load their glossary into our system, that those terms are properly handled according to what they have provided us when we onboarded them. We can also generate glossaries for them if they come to us with no glossary. However, it is the maintenance piece that we are focused on now. They do not have to reach out to us and say, you did not translate this term right, you need to translate it like this. Rather we can detect anomalous content automatically in the UGC that hits our system, but also in their public knowledge bases, any public marketing content that they have got out there. We are constantly scanning and updating that glossary proactively.
Esther: How do you run machine translation, post-editing, how do you interact with those translators who are doing the post-editing? What is that relationship like? What is the UI for them?
Heather: We have a separate platform that the human linguists work in and of course, like when we are talking chat, there is no time for a human linguist, even for support tickets, even though technically there is a little bit of time. If, for example, the support agent wanted a human translation of their response to the customer, we could do a human translation and we do offer that to all of our customers. Over 99% of the ticket translation traffic that goes through our platform is machine translated because once they see it working successfully for chat, they are like, let us just do it instantaneously for emails/tickets as well. We do offer the option to have a rapid human translation of the agent’s response to a ticket. That of course goes to our human linguists who work inside of our human translation platform and that is your standard UI.
Then for support articles, for knowledge-based content, which is still a huge piece of customer support and will be for the foreseeable future, we offer just your traditional human translation everybody is familiar with. The translation step, the review step, so on and so forth. Again, the trend is not that, the trend is toward MTPE, machine translation with a human post edit. This is awesome for us because that MTPE process provides us with another form of supervised learning back into our system to help us improve translation quality via edit distance metrics. MTPE for knowledge base is becoming the norm and the human translation component is less but it is still a critical component of what we do from the learning perspective.
Florian: First, what is your channel? What is the channel partner/partner universe like? Who do you partner with in a channel partnership way and what are some of the supplier/end clients? What is that partner universe like?
Kaarina: We have a number of different kinds of partners so we have the CRM providers themselves, so they are big partners and within those CRM companies, Salesforce, Oracle. We work well with the solution engineers and the customer success managers because they are the ones that are out there in front of their customers, seeing this pain point, knowing about it, knowing us and saying, here is a fit. We worked very closely with them. We also work with the large CRM implementation companies so, both Salesforce and Oracle, Zendesk does not necessarily have this but Salesforce and Oracle, the larger installations handoff professional services to a lot of different implementation companies so we partner with them as well. Now we have the CRM and the implementation companies.
In addition to that, we partner with some of the technology providers out there so I had a call with a company that sells into a space called BPO so that is business process outsourcing. A lot of the large fortune 500, fortune 1000, and even smaller than that, outsource their customer support to a company in the Philippines or a company in India or somewhere like that. They need our technology really badly because generally speaking, what they have are English speaking agents and the trend is that people buy what they can read. You need to provide customer support in a local language for your customers to be happy, for them to give you a high NPS score, Net Promoter Score. They talk about you, they mention you to their family, their friends. Now they are going to buy more services from you.
BPOs traditionally have been throwing bodies at this problem but it is hard to find a multilingual customer support agent and imagine when Covid-19 hit what that did. Suddenly, they cannot interview them, they cannot find them, they cannot hire them, if they do, they cannot retain them. It was a huge problem where our technology came into play in a way that has just been astounding. BPO is a big part of our strategy as well because what our software does is take their single language-speaking agent and turns them into a polyglot. Suddenly they can get chats in over a hundred languages and just respond to them as though they are a native.
Esther: Tell us a bit about the subscription-based pricing, how does that work? What are some of the challenges and opportunities?
Kaarina: Our model is a tiered approach so it is going to depend on how many chats or how many tickets a company needs to translate on a monthly or on a yearly basis. We have a monthly or yearly subscription model because we are a SaaS company. The larger companies, generally speaking, will move into a yearly subscription that gives them a little bit more flexibility in how much content they are pushing through. It could vary from month to month, but then we have an entry-level that gives you all the bells and whistles even if you are an enterprise customer, but it is just lower volume. We do not charge per language that they support. It is a straight usage type model so they buy a subscription and then if their usage goes up, they can upgrade to the next level. We have standard subscriptions and I will give away that our point of entry is $499 per month so it is unlimited agents and unlimited languages as in the number of languages that we support. On the article side, for instance, it is a little bit different.
Heather: That is where I can jump in on the MTPE side. We are moving to MTPE, the whole industry is, but we have also been developing a subscription model for human translation which nobody else that we know of anyway does so far. What Kaarina mentioned on the real-time translation side, the MT side, basically there are these levels and then there is a word cap on each level. If you are going to exceed your cap, exceed the usage, then you need to bump up to the next level. That is easy in the MT world because you just say, the cap is 50,000 words, the cap is 5 million words. On the human translation, even when you throw in MTPE, it is a little more difficult because you have got the concept of fuzzy matches. You have got different rates for these various tiers, and it gets complicated to just throw all that into a cap, but we drastically simplified the various charges associated with MTPE. We have developed a subscription model for MTPE that throws it into various tiers so if you expect to translate 100 articles a month, that would throw you into a particular MTPE cap or level, and then you can move up from there. It is important as a SaaS company that everything be a subscription. What you want is MRR, we will take services revenue but that is not what builds the valuation of the company. What builds the valuation of the company is the MRR and so we are working to fit everything into that model.
Florian: Generally, where do you see the language industry and localization industry as a whole heading, and then for you more specifically? What are the plans for the next six, 12, 18 months? What are some of the key priorities?
Kaarina: What we are seeing is that everyone is moving towards MT. There is no other way of handling instantaneous translation requests. You just cannot do it unless you have this massive staff of basically interpreters and that is all that they are doing. That is just not cost-efficient so we are absolutely convinced and we have been seeing for a while that MT, MTPE is the route where everything is going. Content is getting messier and messier, that is the other thing that we are seeing. The challenge that we have figured out how to do is to be able to handle that content in a way that the two parties that are communicating understand each other. It is the difference between a complete break in the language barrier and actually understanding and being able to communicate with the other person, but it has to be real-time.
Heather: The buzzword nowadays is conversational. That is the future of marketing. It is the future of customer support. When you look at search results, around 25% of search results now take you to UGC content. It is not branded content. It is not professionally written structured content and this is important on many levels because conversational content has to be tackled differently than the content that the industry is familiar with, which is professionally written structured content. Not only is it UGC now, whether that is real-time or not real-time but a conversation is real-time, so we are headed in that direction. It is happening, real-time, conversational, B2B translation, and that is where the whole industry is headed. We feel like we are a little bit ahead of the curve there.