Episode No:63

Insights on the Text Prediction Industry

David Eberle

CEO, Typewise

Listen & Subscribe


Ep#63: Insights on the Text
Prediction Industry ft. David Eberle (CEO, Typewise)
Ep#63: Insights on the Text Prediction Industry ft. David Eberle (CEO, Typewise)
  • Ep#63: Insights on the Text Prediction Industry ft. David Eberle (CEO, Typewise)

Episode Summary

Welcome to today’s episode of the podcast! We have a special guest, David Eberle, CEO and co-founder of Typewise, the AI-powered keyboard app. We discussed his journey into Y Combinator, the challenges faced by Typewise, their unique approach to text prediction API, and his advice for people in the industry. Tune in to hear David’s insights and experiences!
Key Takeaways Time
Journey of getting into YC 4:08
Initial challenges that the company faced 5:23
Text prediction API’s approach to differentiation 10:33
Advantage of in-house models for real-time typing updates 15:26
Challenges involved in serving SMBs 20:20
Company’s expansion strategy 23:11
Advice for people in the industry 26:35

Empower Your GTM Strategy

Monthly expert advice and top GTM insights in your inbox.


David Eberle 0:01 Probably there will be a lot of drop in usage for a lot of these products if they're not very deeply embedded into the workflow. Taylor Kenerson 0:11 Welcome to the Hyperengage Podcast. We are so happy to have you along our journey. Here, we uncover bits of knowledge from some of the greatest minds in tech. We unearth the hows, whys and whats that drive the tech of today. Welcome to the movement. Adil Saleh 0:30 Hey, greetings, everybody. This is Adil from Hyperengage podcast, I have my co-host Taylor Kenerson joining from New Jersey, and a very, very special guest, David, he's CEO, co founder and co founder of Typewise, the emerging AI based writing platform for customer service, you know, engagements teams, would would today, learn more about how does that work? Thank you very much, David, for taking the time. David Eberle 0:54 Thanks for having me. Adil Saleh 0:57 Absolutely. Love that. So you guys started back in 2020. I'm sure that you later went into the Y Combinator and you know, your, your great journey there as well. A lot of partners strategic partners there. So if we just scroll back a little, how easily did you guys found the initial team, especially on the tech side, the technical team, the technical co founding team, alongside us of how was that initial journey? And how did you navigate this, you know, addressable market for such platform as turquoise? David Eberle 1:29 Absolutely. I mean, my, my co my co founder, he was a head of data science at our national broadcaster SRF here in Switzerland. And he was do like, during a longer sabbatical, he started playing around with, with language models, and, and and build like the first, let's say, iteration of our predictive text engine. We use that first on like mobile devices. And so we benchmarked against, you know, Google keyboard and Apple keyboard and which was like the benchmark back then. Not that many years ago. But still, things have moved very quickly. And we got very good results, sometimes even outperforming, you know, apples, like, like we're predictions. And so with that, in the pocket, we then applied for a government grant, here in Switzerland, jointly with the Swiss Federal Institute of Technology, which is, you know, the kind of the Europe's equivalent to the MIT team that we approached, was led by a former Google director of engineering professor, and he then supervise the research, and under his kind of, let's say, academic leadership, we then started building a team. And with also his name, we were able to attract great talent across Europe and started building our AI, research and development team that way, and I think, that has really helped us both getting initial funding to work on longer term r&d efforts, which then obviously paid off later on, but required some initial invest, and also getting getting the people you know, on board, like a no name, startup, which is to, to weird guys. Taylor Kenerson 3:41 Can you Can you unpack that a little bit, because you really took a unique approach into how you came to fruition with an actual product and like giving it to your market, you kind of took more of a research and development? And can you walk us through the thinking behind that strategy and kind of how that allows you just like you said, to bring on these unique personalities where you might have not had that opportunity given you are just like a startup coming like everyone else into the into the market. David Eberle 4:08 I mean, we already had a product, which was a consumer product. And so that I think helped even getting the grant because we were able to show, hey, we already have something and it's a real product. And I think that also got the professor excited knowing that the r&d would be like we could actually test it with real users. So we already had that consumer app. But then the actually so the story was like consumer app, we then started developing AI within the consumer app. And then we realized actually, that AI would be put to better use in a like in enterprise software. And so we pivoted towards enterprise software. And then that story got us into yc. And then we We really started accelerating on that trek. So it was like a three step. Journey. Taylor Kenerson 5:09 Yeah. That's amazing. Talk to us a little bit about that fork in the road, almost where you pivoted into enterprise, and then how that actually led you into yc? And walk us through that journey a little bit. David Eberle 5:23 Yeah, so I think we, with with a consumer keyboard app, they're like, very early on, it was clear that we need to monetize the technology in some sort of b2b setting. Even like our pre seed investors, they only came on board because of that kind of vision. But it was very unclear what, like, what what we would sell to whom we would sell. That was absolutely not clear. And I think if I look back at our first thoughts, we had to write some business plan, which is still I think, common here in Switzerland, if you want to get like I'm investors, you need to write this 50 Page plan. And I think it was totally like, yeah, it turned out very differently, I would say. And so the first idea we had was to license that technology to other software companies, and saying, Hey, we have this AI and we then build an API out of it. And buy it like everywhere. Where do you write a lot? But that turned out to be very difficult? Because I mean, yeah, you have to fit onto a roadmap, then it's a question, how much can you even charge because it's more like than a feature as part of a huge system. And that didn't really seem like it didn't really gain traction. And so we then had the idea of building our own product on top of the API. And there, we then approached that that was in summer two, releasing our consumer app. And we tried out different segments. And I think customer service was one of that and trying out mentors, sending like the like, couple 100 emails, seeing how many calls he gets taking those calls, kind of pitch it, but just getting advice and say, hey, if we had this product, you know, any work that in that way, what would you say? And quickly turn out that customer service definitely had the need for efficiency and quality that there were a lot of customer service teams, every every like enterprise has one. And that they mostly used web based applications. So with a browser extension, we could very easily fit in, and there was no huge IT integration required, which with an API, it's very different, right? If you want to set sell this to, you know, like Zendesk, they would make to make this huge evaluation. And you would, then they have a roadmap of 5000 things they want to do. And maybe, yeah, you get dropped again. And it's it's it's super hard. And so it. And now looking back, I mean, probably it's good. We didn't go that route, because now there's a huge, it's Yeah, I think building your text Prediction API today, it's probably not a good idea. And so having a full featured, like fully fledged product, I think that's like the differentiation. Adil Saleh 8:35 Exactly. So sometimes it's it becomes so complicated. You mentioned to have native API integration, I'm sure you have your SDK on rapid API built as well. So you know, these b2b customers of yours, they can definitely use that as CK and integrate across their processes for the meeting system. I've seen. I'm sure you already familiar with Chad GPT that's been open sourced a few weeks back. And now software's like big CRM, like HubSpot. They're trying to build their own software on top of it, you know, using their models and intersecting it with their existing database that they're trained, was listening their message, the CEO of HubSpot, they build their own sort of CRM that that people can use their own customers can use for sales for marketing and all of that. So how do you see yourself as a business emerging? The chat DVD been open source a few weeks back, what kind of iterations are you going to be making? Because that's a massive infrastructure that they have trained. And it's it's always better to not reinvent the wheel and use these these existing models that have been open sourced. So are you guys thinking of doing anything such like that's one and second, you know, talking about customer service, and sales and engagements like of course, it becomes very, very critical when when it comes to talking to the customer. You need to make sure you're using the right places, right to tell us up bit about how efficient type was the number one without chatty behavior integration, as of now using the open AI three, like GP three model with a previous version, now it's going to be chat tivity that, then it's going to be another update activity for that's coming up. So that's going to be trillions of data points that they are trained. So how do you see is as an industry evolving in the in the natural language process? And how do you think you can apply all of this towards your work workflows to make it a better whole lot better product experience? David Eberle 10:33 Absolutely. So like, it's almost about the user, the user experience, like the user, the user flow, or like the workflow, how we integrate into the workflow of that, for example, a customer service agent, or like a sales representative. And a lot of, let's say, more generic consumer, you know, focused applications have the disadvantage that they're not really embedded into the workflow. Like if you were just to use that interface to to generate your sales messages, what you could do today, you have to go to another window, you have to copy paste. And then, you know, if you just have to answer like a three sentence email, maybe you do that for the first couple of ones, but if you have to answer 100 emails, are you going to really do that for every single email? So probably there will be a lot of drop in usage for a lot of these products, if they're not very deeply embedded into the workflow. Then you have obviously usage restrictions at large companies, data privacy, and so forth. Now, the second, let's say group would be Yeah, like a HubSpot offering this directly to their customers. So they would be able to, like integrate it into the workflow or like very well, I think we couldn't do it better. From a technical point of view, we could do it better, because maybe we're more focused. And we see also clients from different CRMs. And we really own the work and customer service and HubSpot has to do customer surveys, sales, marketing, newsletters, so probably also their TPT integration, at least in the first iteration will will be much more generic across their different CRM modules. And we're really going into email writing and live chat writing only. And so that's already more more tailored. Now, there are two, I would say two, three big, like, you can do differently that a large, like, like, let's say, large CRM company, um, in this case, maybe struggles. Number one, is certain. So today was a TechCrunch article about Microsoft's like launching some features in dynamics, so similar case, and one discussion topic, there was, okay if like if dynamics pre writes an email for you, how much time are you going to spend to review that email? So are you as a as a service rep even faster? Clicking the generate button, which is great, and you get the super cool email, but if like one or two words or even wrong, that can make that can make all the difference? Yeah, definitely apply you to credit or it doesn't. And so you have to carefully actually review it at least, that's how it works today, maybe actually be those issues can be resolved. But and so we take the approach of going a bit the other way we're saying right now, we're giving you like sentence suggestions, or paragraph suggestions. And, and you go kind of one by one. So you're not with a single click building the entire email, but you're building it maybe with 10 clicks. There, we actually see time savings of, like, 30% plus, that's already substantial. But it actually works. Taylor Kenerson 14:29 I was gonna say like, when people talk about integrating AI, it's like, oh, efficiency, efficiency, but people forget, like, you got a copy and paste it into the chat GPT it's providing this great, you know, piece of content, but then how long are you actually spending to review it and there's a Adil Saleh 14:44 cost. There's a cost to it. You have to spend more time to get directions. Maybe you have to go fetch it from chat. GBT maybe credits from you know, there are tools like Jasper rider Yeah, loads of tools that you can use. Cross to your marketing or sales operations, but again, you know, you got to, you got to make sure you find the tools that work inside your environments inside your, let's say, work inside your CRM, let's say, let's talk about Salesforce for sales. So they want integration inside Salesforce do where they can have all of those custom objects, all of those customization and sites it for the test their source of truth. So I'm, I think this is what you're trying to refer. David Eberle 15:26 Yeah. And so our stack is that we have buttons, I think that's that's your question what you know, what about the future? And does it make sense to have your own tech versus just leveraging what's out there? There is for us there is value in like, in parts having our own, like, predictive models and having our own correction models. Why? Because we can apply them real time. So when you type one character, we can update the prediction in real time we can. And so there's, if you write a message with 80 words, that's 400 characters, we make 400 calculations for that message, using GPT would be slower, it would be more costly. So I think for this, let's say problem of like real time updating as you're typing, I think GPT is an overkill, and actually wouldn't work well. It works well, though, to generate an entire message to maybe say, Please shorten this message, or please, maybe rephrase the message. So it better reflects the sentiment of my customer. And so that's something we will I mean, we're also leveraging GPT. exactly for that type of problem. So think we're saying GPT is not the thing that does 100%. But if you have to use different different technologies, for kind of different problems as part of the pictures, sometimes they even use templates that the client can just predefine because they have a communications department that once you know, certain things are mandated. This is how this is our boilerplate. We don't want anyone to rephrase this boilerplate This is approved, or some legal statements, you cannot rephrase this, it's it, you know, you will, yeah, it would be no longer valid. And so I think there's a combination of using different technologies to like complete the picture. And I think that's, that's where a lot of value comes in. And that's maybe also lengthen the third and last point, what I think is an advantage is that, I mean, that's why we also focus on on like large, let's say larger enterprises as clients, because they often have very specific requirements, we only want sentences to be 15 words or less, it will be very hard for like a HubSpot to then start offering all these customization options, as part of their feature, you maybe have to, like, retrain, like kind of reconfigure the grammar correction and say, you have to for this client, they want this for that client, they want that. And then of course, because we go after larger companies, they can pay the premium that requires that, you know, to kind of build that configuration, but that are really beliefs, then that gives us also like a longer term advantage, or like differentiation of why the enterprise would take a type Y solution. And not just take the built in feature mean even if you look at like Outlook to hat they have been having all the correct grammar check forever, but you cannot configure anything as a company just out of the box Standard English, what if I, I don't know I want to be funny or formal or whatever, I cannot change this. And so that's why Grammarly also succeeded, right? Because that was possible with them. And so I think we're it's a very similar strategy. Yeah. Adil Saleh 19:13 So David, I would love for you to, you know, explore this, or this moment, right at this moment, touch on the SMB side, you talked about enterprise, this can be a big use case for support service and sales team and Energy serving enterprise segment, talking about SMB segment, we have we've heard this a lot because we have CST in from all range of different technologies, regions and all. So the newer the newer emerging norm is to do more with less, make sure you have you are absolutely taking care of your bandwidth, and you're not increasing the technology or tech stack, but at the same time you're trying to optimize your time. So you can have like one person serving like 30 accounts 40 accounts and the surveys and the success of the sales. So from that state unacquainted. How do you think this platform like pipelines can come into play? Because there are loads of platforms in your industry, competing you they are more targeted towards selling the SMB model like SMB segment, companies that are more self serve, they have, like hybrid sort of touch. And they're using technologies like this. David Eberle 20:20 Yeah, I think for us that that's not the strategy at the moment, the thing we're really going after, after the larger accounts, I think, yeah, if if you serve SMBs, you have to automate everything from from the start, for example, with with a browser extension is that you have to support so many different web pages, that in the end is like you integrate into like a text editor, which is the field where you're like writing the message. And there are about 10 text editors in there. And there are different versions of those. And if you're a small team to to, I mean, even grammerly. That's, that's a Decacorn, they have a lot of bugs. So you can test them on like MS teams webpage like web version, and it's buggy. And that just shows building a browser extension that integrates into your text editing, and making that available for for, for for all sorts of, like CRMs, I think that's, that's quite a big task. And that's why we said, look, we need the customer to pay a reasonable amount of money that we can even ensure that compatibility, and that's why we go to larger companies, I think if you want to do SMBs, and for example, one of our one one, there's a new YC company. He was previously the founder of co founder of Socialcam, he now launched a SMB targeted but only on on the Shopify. So he said I'm limiting one platform, and Dara do SMB. Customer. That's, that's what you have to do. And on the enterprise, you can't really because then your market is like Danny gets a bit hard if you say I at least the beginning if I only do Zendesk clients, large companies and I in enterprise, we find it makes more sense to limit in geography and by industry, rather than like by by CRM. But the SMB approach, I think, should be then you really do one plug in for like one system and try to scale in there. Adil Saleh 22:43 Yeah, just like a lot of, you know, data integrations platform data analytics platform, they're using Salesforce, they're using segment, they're using one to one like, it takes them like three, four years to you know, add integrations, like on top of the so I mean, that that definitely makes sense. So in terms of market positioning, David, you got quite a bit of traction other than in the States. So how are you trying to on the commercial side? How are you guys trying to penetrate and expand from this point? David Eberle 23:11 Yeah, I think that's a good question. I mean, that we're predominantly Europe based at the moment, although we're like a US company. But I mean, we were born in Europe. And so our strategy is to rather expand quickly in the German speaking with, you know, like a more limited team. And then, once everything, like let's say, once all the onboarding processes are really working smoothly, and we know we can very quickly add add at new companies, then I think that we can, we can double down on the US market. And I like what I wouldn't do is focus too long on Europe, because it's quite fragmented. And, you know, you have to go country by country, even for large fortune 500 companies, they usually make different decisions for different countries. And so it becomes a very long process. So I think using our home base to get to certain first, let's say first level, but then being like a US company, I think it's the right, it's the right move to then in phase two, double down on the US market, but I think doing so early, you know, then you have a lot of complexities with timezones and us is expensive compared to here, and so I don't I think it would be like if you cannot get to a certain level in your home market. I think we have another problem. Yeah, I think expansion is not the right solution. Boom. Adil Saleh 25:02 Absolutely. So you're in the near future you're trying to expand within Europe. So now we've seen that you're trying to hire you have some position opening number one, you need to share what what kind of roles you have open, what kind of skill set, you would need experience and experience with location with the remote not, because we've got some recruiters that that can help you find the right candidate. So you're gonna stick about it? David Eberle 25:23 Yeah, I think it's mainly to two positions, like we're looking for, like mL mL ops slash dev ops engineers. It's kind of combination of both into one which makes it a bit tricky. And the other one is, like, senior JavaScript. Developers. Okay. Okay. Mostly for for for for that reason, mostly, like Europe. Europe, based. Yeah. So because we also let you know, we're remote, but we also like to get the team together once a quarter physically got it. And that makes it just easier. Yeah, yeah. Adil Saleh 26:06 You gotta you gotta need your, your team and sell your art. So whenever you invite quarterly meetings or lunch or get together, you have people available, they have access. Great, wonderful. Do you have one advice for people in the industry? Trying to, you know, kickstart their startup? You know, especially going into yc? Is it right? Or not? In the first two years? If yes, why just give us a brief recap on your journey. And what are takeaways for, for people listening? David Eberle 26:35 I think yc is always a great, you know, a great experience, I think we were quite late. Others, you know, barely met each other, like a couple of months ago, and they're, they're still in undergrad, and and they're pivoting their idea, like day one, and spend YC, figuring out what they should be doing. And I think it's like, regardless, I think it's, it's always great. And I know also many that that applied multiple times before getting in. And so I think I wouldn't wait for the right moment. You just didn't apply again. And, I mean, if you could show progress, of course, then I think that's, that's also part of the, like, part of your of your story. is probably like, it's try try something out. But yeah, I mean, it's always a balance, right? Is sticking with with it, because it's probably always hard regardless, consumer, like, if that doesn't, yeah, and then you go to enterprise, you think, oh, enterprise is easier to get revenue, but then it's also super hard, because every single account is like a grind, then you think we should do product like growth, but then that's also hard before for like other reasons. But if you also realize something is not working, I think, like, reading the signals correctly. And I think YC really helps to say, what are the signals? And how do you read them? And it's usually very simple. There's like one metric and you focus on that. And don't find excuses of why, like, if your metric is revenue, which is it should be for most companies, and that's not moving, don't like go to another metric and say, oh, you know, like, now we're tracking I don't know, usage or like something else. And yeah, and I think stick stick with it, read the signals well, and if you have to make a change, make a change. And I think having, like such a program, that Yeah, I think it really puts you in the right mindset. But absolutely not. You can also succeed without I mean, Adil Saleh 28:48 yeah. So because in the start in the beginning, know your customer, it's very easy to say, but it's not easy to, you know, basically implement, you need really need outside voices and outside noise to get that point and knowing your customers and making sure you're making the right decision inside of the product. And sometimes you get you get the wrong sorts of customers. In the beginning, you hand it over to the wrong people, they try to drive your products elsewhere. So that's where you got to be very smart to having having strategic partners. Like YC, you know, there are people very smart people and you know, specific to industries and categories that can help you get to the right customer, knowing your customers, as you mentioned, knowing your segments. So I really appreciate David for taking the time. It was real nice conversation with you. I love that you've been so genuine about about your story and what you want to do with the disruptors. David Eberle 29:39 Absolutely was great talking. Thank you. Taylor Kenerson 29:42 Thank you, David. Adil Saleh 29:45 Thank you so very much for staying with us on the episode please share your feedback at adil@hyperengaged.io We definitely need it. We will see you next time in another guest on stage with some concrete tips on how to operate better as a Customer Success leader and how you can empower engagements with some building some meaningful relationships. We qualified people for the episode just to make sure we bring the value to the listeners. Do reaches out if you want to refer any CS leader. Until next time, goodbye and have a good rest of your day.

Keep Listening VIEW ALL EPISODES >>>