Watch Omilia CEO, Dimitris Vassos, discuss how conversational AI is moving the needle for companies across sectors with Omilia state-of-the-art technology. Stick around until the end to hear about a success story from a Fortune 500 company that receives over 30,000 calls per week!

Transcription for the video

So, what I’m gonna do today is talk a little about Conversational AI and where I think as a technology where it’s going and to do that I’ll take a step back a little bit and look at how the technology has developed.

First of all, it’s important to define what I mean by Conversational AI and what’s the mystery behind it because there’s a lot of people and customers that I talk to and they’re somehow all tasked to do something with Conversational AI; but they’re looking (to see) exactly how to do it and what to do sometimes. So, I’m going to focus on Intelligent Virtual Assistants so I’m going to focus on a little bit of technology that allows you to provide customer service to consumers automatically with the technology rather than a human, but still have a human-like experience.

When I’m asked what Omilia does, the best response I’ve given was last year and I’m quoting myself here, that “we’re in the business of destroying IVR” as the legacy experience" that we all love to hate and “we’re rediscovering IVR or the voice interaction as a new digital channel". That’s what Conversational AI I think is enabling today.

So I’ll take a step back with technology. I’m old enough to have seen several technology eras in my time so when I was just in university I think it was the time where the PC, the personal computer, was being born. So you know the history of the PC and how that technology era evolved is very close to me and something that I witnessed firsthand. And I remember PCs at the beginning– very clunky and very expensive, the performance was- the memory you would get was in the kilobytes. Now it’s many many zeros beyond that.

So I’ve seen a timeline of the personal computer from the trigger point, the innovation triggered back in 1981 with the IBM PC through to when processors became much more powerful with the Pentium. And then I saw an explosion. I was still a student there or maybe just off university. It was a big explosion of all these different components that you could go to. And I was studying in London so I would go down to Tottenham Court Road which was a street with all of the electronics shops and you could literally buy any component you wanted and put them all together to compose the PC of your dreams really or supposedly because as soon as you composed it and you tried to put software or the operating system on it you had to go through this device driver hell which some of you may remember where things started to break down because of compatibility issues and testing. And it was really the democratization of technology seemingly but really what was happening is that everybody with all the consumers were forced to become an IT specialist.

And of course, this didn’t last for long because people saw the gap in the market, they understood what was going wrong and people like Steve Jobs with Apple, they came up with integrated, best-of-class packages and said “Well, don’t bother with becoming an IT guru and putting together your PC … here’s one which actually works pretty well for most people.” I think that paradigm then was followed by a lot of companies and not just in the personal computer market but also with smartphones like the iPhone and all the explosion that happened there.

And I parallelize that historically if you like, a timeline of the personal computer era with what’s happening in AI today. And I think the trigger point in my opinion for AI was just about in 2003 when there was a little bit of, I think, a breakthrough in terms of the neural network training algorithms plus the fact that computers were becoming fast enough, that made neural networks actually start to work. Neural networks were definitely not invented in 2003 but earlier they didn’t work that well.

So, that was a trigger point and of course, IBM threw a lot of marketing money into making sure that people knew what was happening with IBM Watson. If you remember the Jeopardy tv series where IBM Watson beat the best human in Jeopardy and that was a big deal I think from a marketing perspective because people started to understand what AI means for the consumer and others followed with Siri and Watson in 2011 and then I saw this again- this explosion of all the different vendors, components, services, a big proliferation out there which led to a lot of confusion because there was so much and so approachable technology out there that it was really again coined… I heard the term “democratizing AI.”

So the way that I experienced that is I saw a lot of the clients I’m talking to- they were spinning internal projects to try out Google, try out Amazon, try out Microsoft, all the AI and cognitive services and see what they can do by themselves and what they can put together and experiment really. And again although this is still happening I think we are seeing now people having done a lot of experimentation having almost become gurus in putting together these services and of course understanding the shortcomings of DIY assembly of a personal computer really they’re now understanding there must be a better way. There must be something that somebody has thought of which is a best-of-class integrated package and I think that’s where we are today.

There is of course a great deal of platforms out there that allow you to develop voice bots. There are tens of thousands of voice bot developers at least. There are at least 150 bot platforms that I’ve counted very very quickly on the internet and I bet there are more than a thousand voice developers today around the world that are spending their time developing voice bots for financial services. And if you think about what they’re doing- all of them are doing exactly the same thing- they’re rediscovering the wheel. We have 1000 people doing exactly the same things every day. They’re building intent models and they’re providing utterances like “what’s my balance? I lost my card. I want to pay my bills. I want to transfer money.” All of them are developing exactly the same machine learning models with very very little differences between them.

So that’s an exhilarating experience I think because people are feeling like they are empowered and they can put together the motorcycle of their dreams and it’s a great experience; but at the end of the day where does that get you in terms of business benefits and where do you go next so that you can take that into the market with real customers with the complexity of the real world, and get results. And I think there is quite one way to do that and that’s to really understand what is the real recipe for putting together something that will eventually have user adoption at a wide scale.

Speakers before me talked about the voice user interface and it’s definitely a different animal. It’s definitely, I don’t think experience to design something and to implement something that’s going to be able to understand the chaotic behavior of humans. Doing that also- a lot of speakers before me said it doesn’t just take technology, it also takes art.

So the VUI experience is super super important and it’s definitely something that’s right at the top of my recipe. Strong algorithms? Yes, of course, you need to have strong algorithms that can provide good machine learning models with good accuracy but we also touched up on general AI and specialized AI and I’m very much of the same mindset myself because of the name of the game I believe is specialization. You have to if you want to be good at something or good enough for primetime, you definitely need to specialize. And how do you specialize; today it’s all data-driven so you need to have a lot of real data. Not just any data but real data.

A lot of the times I talk with customers and they say but you know Google has the most data in the world and nobody has more data than Google and that may be true but there’s data and then there’s data. And when it comes to a specialized task you really have to have specialized data. You really need to have the right kind of data.

So if we’re talking about, for example, a voice bot in the contact center, you need contact center data and there’s no way you can do it otherwise with other data. There’s no substitute for the real thing. And of course, once you have data, you need more data so I cannot stress enough how important it is to really have a lot of the real right kind of data.

And having an integrated platform I think is also quite an important ingredient. I think speakers before me also touched upon this, “it’s a tough nut to crack,” the call center integration. So when you are putting together your DIY solution you definitely do not want to go near telephony and integrating to telephony that’s- it’s not fun. It’s a difficult task with a lot of pitfalls so you really need to know what you’re doing there.

And of course, once you have all that you need iterations, you need to mature the whole proposition, the whole solution and you need to be exposed to the real world with real data, with real customers. See how the whole thing works and go back to the drawing-room, improve, get more data, tune, and by doing this process continuously, you gain a good understanding of where it is you have to go.

And, of course, love and care I think, is super important. You have to have a real love for customer service. You have to really want to be in the shoes of the end-user on the other side of the AI and understand how they feel, what they want to do, and whether you’re really developing or designing something that’s right for them.

But you know, the way we solve this problem- and don’t get me wrong- this is a recipe here that it may take 10-15 years to put together, right? So, this is definitely not something that you do in the course of a single project. It’s something that requires quite a lot of investment over the years. The way that we solve this in our recipe, is of course to go to the cloud, because offering something from the cloud means that you can offer a single solution to many customers and when you’re tuning a single solution, everybody is benefitting; so the idea of having this, building this AI and then sharing it across the market is something which enabled us at least to focus and get to the next level.

Our proposition to solve this problem was you don’t need to DIY. You don’t need to even be a voice user interface expert. You don’t need to join the DIY wave and you don’t need specialized services for someone who knows what they’re doing to do it for you.

We package technology and know-how into reusable components which we call OCP miniApps and they are specialized; so each one of these blocks is a specialized AI that’s so specialized that it’s extremely good at doing one specific task and no more. And then really, what we are doing is we’re saying “with these components, you can now put together any Conversational AI experience you want and you don’t need to worry about the voice user interface, you don’t need to have designers". It works itself out based on best practices.

And a key moment in that journey for us was not long ago, a few – a couple of months ago with one of our customers where we built a voice bot and deployed it for them. It’s one of the US’ largest car retailers, a Fortune 500 company. They wanted natural language understanding for their IVR, and we built something very very quickly for them based on OCP miniApps. Basically, we went from hello to going live in a matter of a few weeks. And the accuracy we got out of the bot was above 90% and this is writing zero code. We didn’t even write a single line of code and the customer of course didn’t have to write any code themselves.

So the big validation to this journey came a little bit later when we deployed that where this customer, not accidentally, we had given them access to monitoring the solution and seeing how the voice bot is responding to their customers’ requests, and they, I think, just poked around a little bit on our platform and they found out how we had developed this experience for them.

And because I’m based out of Athens and this is a US-based customer so we’re in different time zones. During their business day, we may be sleeping. Well, I work in two time zones but most of the team is sleeping when the US is doing business. They had the opportunity to go into the platform and just look around while nobody from our team was there and they decided they had a good understanding and enough understanding of what was happening to just change a few things and deploy them. And they did change the voice bot. They cloned it. They changed a few prompts, they changed a few settings, they changed the error handling of the Virtual Assistant, and then they took it to live!

Live traffic. Just like that. Hot-plug without ever asking any of the Omilia folks. And the next morning we came into the office and we saw the application we had developed was not taking any calls so we were naturally alarmed and were trying to figure out what had failed. And of course, nothing had failed. It was just the whole traffic was redirected to a different application that the customer had themself developed. And that was a wow moment because at that point I had understood, that the platform we had given this customer didn’t require any designers, definitely didn’t require any developers. It didn’t require any speech scientists. It didn’t require building any specialized speech recognition grammars or any intent models or specialized NLU models. Nothing.

The customer was able to just go in, change, all the magic happens in the background and they even felt comfortable enough to deploy in production. In the words of the client, what they found very very and highly valuable from the platform is that the interface is very intuitive and easy. And it is so because they don’t have to deal with any of the AI. You don’t see the AI, you don’t have to. The AI is advanced enough to be able to understand and adapt itself to whatever the business user or business analyst is trying to do. And you know, having simplicity also helps you have a better understanding of what happens because you get these very advanced Conversational AI platforms with a thousand buttons and a thousand fields, a thousand screens. And it becomes very daunting. You need to really get trained and understand how to use those very advanced tools.

So, having something simple, a lot of the time gives you a much better understanding as to what is happening and gives you more confidence that what you’re doing- you know what you’re doing really and more confidence to take it into production.

So, to close off, having seen the evolution of different technologies over the years, what we are aiming to bring to Conversational AI is definitely the best of class, human-like Conversational Experiences. That is definitely the #1 target for Omilia. State-of-the-art accuracy, because when you deal with customers they need to be understood and customers are very demanding. We are providing within, built-into our service, all the terabytes of data that we have collected over the years and trained expert models. You don’t have to worry about getting data because that’s all what we provide. And of course, when you have all this, your speed-to-market is in days rather than months. And no coding approach. I think that’s also an important differentiation.