Events Add an event Speakers Talks Collections Sign out
 
Duration 20:54
16+
Play
Video

What’s next in CDD: Intent Clashes and Selective Confidence | Rasa Summit 2021

Sebastian Pedavoli
CEO & Co-founder at Proxima
  • Video
  • Table of contents
  • Video
Rasa Summit 2021
February 11, 2021, Online, USA
Rasa Summit 2021
Request Q&A
Video
What’s next in CDD: Intent Clashes and Selective Confidence | Rasa Summit 2021
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Add to favorites
250
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Sebastian Pedavoli
CEO & Co-founder at Proxima

Leading the team at Proxima, a software development and conversational user experience team headquartered in the beautiful Sydney, Australia. The vision of Proxima has always been to connect brands and people through software. We are the creators of the Iris Conversational Intelligence platform, a Conversational AI platform powering meaningful, real-time conversations to engage, empower and support customers at scale.

View the profile

About the talk

Clearly understanding your users’ intent, sentiment and background information is critical in helping them achieve what they want. Hence, we built our platform Iris around Conversation Driven Development even before we knew what it was.

The challenge now is the need for a huge collection of data to get close intent matches. In fact, you have to design, build and test your chatbot with INTENT CLASHES in mind. We’ve used Rasa to build SELECTIVE CONFIDENCE, addressing intent clashes. If we know that the user is going through a sales pathway, we won’t match them with content from, let’s say, customer support. Or if the user says something irrelevant, we can still engage them with the identified flow without losing the conversation. This enables you to build a much richer user experience.

Global brands such as Twitter and Flight Centre are taking advantage of this. We’d be happy to share these creative and innovative projects using Rasa.

Sebastian is the Co-Founder and CEO of Proxima, a global software and conversational AI developer headquartered in Sydney, Australia.

With a vision to help brands engage their customers at scale through conversational AI technology, Sebastian co-founded Proxima in 2014. He has since overseen the creation of the firm’s Iris Conversational AI platform, which powers intelligent, real-time, omni-channel conversations to help companies better serve their customers whenever, wherever.

In just seven years, Sebastian has led Proxima from its inception to become a global business supporting major brands across multiple industries around the world, including Twitter, Flight Centre and Lifeline.

Today, he continues to oversee the firm’s strategic vision, product development and global expansion, and is also a regular speaker at industry events.

Away from work, Sebastian is a keen cyclist and regularly competes in cycling events.

Presented by Proxima CEO & Co-founder, Sebastian Pedavoli at the 2021 Rasa Summit.

- Learn more about Rasa: [https://rasa.com](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqa292TWs4eUVTMWp1NTJuYlM5NWRPM3ZjUkhGd3xBQ3Jtc0trQldlM3ozV2szQjNBeDd0VE5HRVR1MmhRWnFNWjNqS2kxMGhJTk04Q21LWkZLWWNNWHJPSzRRXzFyUnFtaURYbkZFYWJLUkxXaFRyZTQtQXBsdG1VV2tCa0ducHpYQ3FHdVVoY0lZcUZ0TVY2eHhQdw&q=https%3A%2F%2Frasa.com)

- Rasa documentation: [http://rasa.com/docs](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqa2Z5TkJraEdEcG9mS0hKelVHUjZoSzZGaFNFUXxBQ3Jtc0tuTTJxQ0MyZzVPdTR5R3cwbmF2aElfSF8wTW5IN0dUSG5ZbGVLRVpFYkNCX2ZoMklNYWJudXNsQnJpTEpULWNGYkRuaXdjNEhTZUdyMjd6cG9EeUJtRV9HUkZsZFdKcjliVExnQ2NONXpYV3JBNWsxdw&q=http%3A%2F%2Frasa.com%2Fdocs)

- Join the Rasa Community: [https://forum.rasa.com](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqbDRCcFpBLW96SUZ1Y1FJWXhHeGpISUdCMFh6d3xBQ3Jtc0tsNG1MRmVtVnY1d0N4a1pUYTJTa1hMRTRBcUxPdkVyaUZQZkc4RkIyenp1aHZDYTg0WnlHS0dOaktDR3pjRkpMS29JQ0NKTjg5M2dKckNEd3JQeVBkWGUxWEQzUkpybXBnNFJWYVA2X3ZqSTVoYlBSTQ&q=https%3A%2F%2Fforum.rasa.com)

- Twitter: [https://twitter.com/Rasa_HQ](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqbTNVNW00cXJUSkM5b2RUbVdZQnZrYnVma2s2d3xBQ3Jtc0ttcnBuN0ZfeU1vUG5VY3ZwbmJpcjRtN2hCNVNzcGkxMmV6T3JXTTh6cDdnSHVTYzN6NHFuUG9JV09xTUNUbFBrUHdIeXhBbFBJZkZHUnpNQVF1cFVqdVQwVE1veUgzZjRmYXVuUHd4QTNQdkdFWTBBdw&q=https%3A%2F%2Ftwitter.com%2FRasa_HQ)

- Facebook: [https://www.facebook.com/RasaHQ](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqblJUSWhmOHFJWFpPZE9RY3BuQVpNOTlwRWdqUXxBQ3Jtc0ttMllOdjdfcE9zVU5lSjA2VjFSQ0c4eGRpVjdrXzBrbzVqcWZaOTdHcDl4T2tWX3RmNHdkQWtGa3BKc3diaE1adWo5SlFJNS1HLVBVSGlhWWFaSEJ4elZWdzB3MWZNdVBQZUNZUHhsYjNQOWVaTUIxOA&q=https%3A%2F%2Fwww.facebook.com%2FRasaHQ)

- Linkedin: [https://www.linkedin.com/company/rasa](https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqa2czR3UtUTJyNldHNjdZQ2pjRzlxZmM3UkZOUXxBQ3Jtc0tuTktyamZOZWhBX2lKZWotRml4THd4cGg3alBId1VLNWIzamlKd3M4SG9iRE5fSjRFM0FoYmtmWnctOHh3SE9vSE0xbXVxV2YwdTRDOGVvNUN1aTZJcFlhYk1wYUJxNmtZSDNWLTNjTWdwbnV4NTNZVQ&q=https%3A%2F%2Fwww.linkedin.com%2Fcompany%2Frasa)

#conversationalAI #aichatbot #CDD

Share

Talk about intense clashes in selected confidence. So going to go through this morning, I'm going to try and get through everything relatively quickly. Give enough time we spoke at the end of run to dive in and just ask anything of me doing so running through a little bit about who has come from. And then a little bit of a used case, Sarah Live customer Valley Market, and how that's relevant to what we're actually doing around trying to sell French and clashes and select competent and then going through a little bit about the whole process. So what it actually looks like an

imitation of Raza which might be a little bit different for some people of how we can put mentioned it there and touching on a few things and how have used different training models? Also, how we have experienced that we can either choose whether to use intense aware, they would go completely Atlas and then also, which is Experiences in the hands of customers and uses both background. So I contact the state. So I thought I'd just get a little bit of background of still, kind of where we've come from. And a little bit of an understanding about

who we are. So we have been building conversation experiences since 2016 and we are identified early on the reset of conversational experience that ran between an organization and a costume. Is it had to be a channel like no sleep. We had to have a system that could run behind all channels, add event, only Channel solution. And it also had to be able to handle everything from sales through the support we win. We never, we knew we never wanted to build an experience that was on one channel. That was very

specific with the sales are very specifically support. We were looking to solve true business problem. So I actually meant that we had to be quiet. We wanted a system that could be generic. Men not focusing on a specific business case for industry which was probably a little bit risky. It's not your traditional mode fractured launching company. You tell to sort of think about to go pick my pocket and got hot after the way we kept a little bit different we went to Market, but I think in the early days, we definitely saw ourselves as the Middle where faces we were connecting a lot of these

pieces together. Essentially resource else's conversation. We were basically pulling a lot of the pieces together and then just trashing traffic. But what was interesting? The very beginning when we started building some of these Solutions, as we didn't actually see a product on the market if we sought Solace fit. What we were trying to achieve with this Vision across the site where we could go, but we were trying to do and then towards the end of 2016, the actual Stars, Fine for us on Twitter approached us to build a boat to the u.s. election behind their official Twitter

handle and that was the first version of Iris that we actually launch site. Glen Cove into building conversation with appearances. It was a huge amount of scale. They push that bought out to surround the 33 million people as it went live and it was rolling into the election day so it was very states of Atlantica. Thought that was the beginning of the inside of the Genesis of virus, which is a conversation on a iPhone. The way we look at Iris were really building, conversational ASL for the future Business Without trying to build for now, we're trying

to build for where we think the pack is actually going rather than focusing on here now and it's all around. So, the old man seeing how businesses work streamlining, how they communicate with their customers and also have a key point for us is an you say this, so, touch-up throughout the talk is we're looking at Seaman in the live concert. How do we pull it Parts? The people that were building at 4 and the people who got to use it front and center in the car, talk to any experience that drive the CD process is well, and put some

front-and-center. It means that we get immediate some feedback from the perspective of Iris. We are focusing on conversational marketing, building virtual assistants customer service at customer. Service customer support. Location and then also disruption notifications. So, from a server business, use case, when they coming on board they bury for the mystery. But everything from notifying engaging empowering and that she's supporting customers as well. So we're a relatively small team, but we're bicycle sydney-melbourne in London. We have some fantastic Channel

Partners which support friends in different markets across the world from marusia Zealand. And we get to work on projects from everything from Twitter off lights until we do the launches of epic games. We recently just woke with the US, Democrats are on the u.s. election as well as nonprofit face from lifeline and we're actually have a handful of Transport operators as well. We've been rolling out with that have been able to use utilize the software and we say that the problem that way. So I'll be in the Mach is very complimentary to what

without to rasa so you'll actually See how little she got robbed the running in there, and a little bit of a different type on it, so I guess and one of the questions at the house, but how will we work in conversational marketing? How, how we running that? It's Kyle. And essentially, what we found is that it's a conversation to go to run conversational, marketing Skylander, post, Malone station on, which means it was really effective and that sound of a space that we've been able to kind of corner and globally. One of those

make of it and classes and selectively confidence. I saw a couple of questions that are always are popping up when we're thinking when we think about building an errand, especially at that complex, wanting to see these guys there about coming up that, you know, when you have a lot of scenarios, you have a huge amounts of DACA, what do you do when you get them 10 classes? How do you build an experience where you can avoid those scenarios and do it? Where is not going to solve your problems anymore. It will only get you so far and

one of the key problems in this, use case with light since I'm going to talk about is the refund problem. Where do you want a refund? And do we need to process your refund and pre-qualify you before we hand you over to an agent or do you just want information about it and you want to be the background and it's that actually within an experienced. How do we make sure we're asking you right now the right question, then How do you do this with a certain level actress, accuracy and consistency? So when your building in Skyline, you have thousands of people. How do you do it time

and time again, really quickly and consistently and consistently across a conversational experience as well. And then that will not from that perspective. You know, if we have this in mind, when thinking about Sadie day, we use this to actually drive on the road with thinking about how we can actually use a set of selected companies, thresholds, and only classified and sudden, intense particular points. India, since the very beginning and not replying back to

2016, we had to figure out what was actually called. Conversational driven development dollar design. Absolutely. Same thing it's essentially and approaches of a methodology that we used We have templates, we have processes but essentially I think it all comes down to its customer Centric development and testing. How do you put people in the middle of what you're doing? How do you wrap a private hot Bill. Get it in front of them in jest oughta and it's that she even in the lives of conversational development and we use that across,

we use that across to do everything that we're doing. So for assets, it is proactive element but because we're by the park company and the service companies with building the solutions as well on top of that as well as I that keep me in the loop. He's a product Focus for us and it's also a development and a customer experience. How do we deal with the customer experience from booking flights caused and whole days? I think they bigger in Europe have been in the Australian founded company. So we we look up the experience for them and it's

really interesting use guys. So they have thousands of customer interactions are coming through the channels that was 44 them really interesting because it's almost certain that Marquis customer for us because they have customers segmented into 3 raised to the different types of customers that have sales. I have support and a general inquiry and customers reaching out. They're they're coming in with a general inquiry that giving in to support building II style or they have a very distinct intent when they coming through. Another

interesting part about actually running on flight center and being on a messenger channel, is we get by the organic and pay traffic. So some people coming through, I just reaching out, organically replied center with a support request. Other people actually driven by Facebook ads, so that's where the system comes in to be able to do different entry points. I'm detecting add flows coming through and I keep pushing papers open can use all of his experiences bow because it's such a big organization to be able to actually integrate into their existing systems as well. And it's also quite complex

as a great example, because we They're very complex business rules in place by about five, different order numbers across the different types of different different order numbers across the organization, is all that people can have this 10 different scenarios through that. And then there's also a complex business rolls and pie. So as we're building the solution and we're working on it. So they are running side-by-side with their agents. How do we take into consideration? Those vegetables or not but we'll be talking to

a little girl. So, kind of out of it, how do we actually go about the cement and dive a little bit deeper into this, and that human in Loop is, is he under starting a bit deeper on that one somehow perspective? It's the product is designed where people are in the loops from the very beginning, and it makes everything. It makes everything more efficient furnaces. Well, bye-bye, having a product that is designed to have people in months, they are designing and building it from a testing perspective, as well, from a development and then going

to be at you as we transition customers into business as usual. How do we make sure that training process that we take them through? How do we tighten? That's a detail as well. So I bet the feedback that's coming through as well as supporting customers globally. This point in time is only 12 people on a tank but, you know, we are rolling out new projects. Anywhere between two or three new projects of wait. What? Can be different Channel? Partners customers globally. Efficiencies K Frost. So it was almost out of necessity. We've had to figure out how to tighten that Luke

and then also be able to build very flexible experience. We have an interesting implementation to Brasa where we can selectively design into it. Different confidence threshold based on where you are. We can selectively choose which intends to classify and then within an experience so when you drive inside experience with flight center is not just a single Rosen Mobile in there as well. We have two different ways of Mortals are cheating, two different things within the experience and we have, we can then choose different, our confidence specials at different

points in time and we, we do that part. We can either do that programmatically or we can actually do that 5 to sign as well. So this is a little bit more into the process about how we actually go about it. But I do actually have in the next couple side. Visually what this looks like. Touching a little bit on the actual hours platform as well. Because this is where, you know, we have that resume service biking today and this is what it looks like when it actually when it actually runs in, this is also the dog

designer allows copy-editing and rapid prototyping. And that means that we did not see the dates of the methodology and porous as we're able to share what we building very quickly and get it in the hands of customers. And, but it means that that dollar designer, he's also makes is a developmental for our guys internally as well because the load programming language. And it also has only integrate integration in space as well then allows us to build out the conversation before and this is where Action cutting the business logic. So it is his to the Aquatic Complex

pacing but allows Bus to go to share of you and Savannah State and test it really quickly to go to roll out those experiences and select your blade design in a visual sense, what it actually looks like to adjust your confidence, special or selective are within this particular point in the experience. I only want to classify. I can't stay in tents then Launchpad. Is that you know the price I said actually design Stage production of review, assess repaid, go through that process. It is it like a visual interface will you can quite literally attesting, and then push into production

and then going through that price. Then that sort of cool conversation engine is where we've got Raza and the two different limitations of a traversal implementation where this is us using intense and natural language classification to push people through an experience and moves on true conversation and the second one is another device. Look up. Sorry that we're able to. The different model and what route do there is just so that those articles of the actions to achieve

move to different parts of the experience but we've got to Immortal sitting side-by-side at the same time and then natural language interface as well allows us and also the customers to get in there and see the see the training daughter and also be able to adjust confidence threshold. Seefeld interview quickly, we trained at the Scott at simulated load test it, and then constantly just go through that cycle, quite quickly, bouillabaisse of everything from the designer to the launch pad to the conversation ends and then the natural language interface come

together to say, they look. But then also allow us to have that very sort of fine-grained, granula ability to adjust so things and this is where we really get into them made of This so selective confidence threshold selective intent. So I looked at before we've got these two with inexperience. Generally, we have to buy the motor was running and we have this one for traversal, that moves the user through the concession stand sign. Another one for natural language that and because we going through this, and it's a visual programming

language and it is low programming. So they executed in real-time the Lori that actually run by a system because it's a conversation flow. So if you're thinking, when you building conversation parents has you need a decision tree, right? So we're we're literally doing the decision tree and then we're coming back around and replying copy to it. Then what connecting the natural language experience to your them where I kept testing it and running it in writing on it, and everything is done, everything is done here from a design perspective and the cake eat it. You

can say, want that purple hour that I have to, that's the ability to select. We link an unlink intense that we don't want to classify Pacific point in the actual experience as well. So we can actually decide that contextually you'll halfway through a support flow and you're answering questions of around FAQ and support. We don't want to classify a gangsta, all of the sales. I'm intense this, what we can actually tell you that is that's what we programmatically, quite literally Lincoln on the knowledge. I mean, with all of that means that even though the Raza Motors training, against the

programmatic laid there, we can adjust that confidence threshold, in that call to Raza on when it returns and choose which intense, we actually classify back. So if another route is only connected to support request that because we want to move someone around from a support perspective. We can have an individual, like a style change of mind in Tampa. Maybe the only sales in tent that we trained against that and that can be along the lines of my returning to the Is a Halo Hopper 3 in a sole support request. Are you sure you want to push of it at the sales and things like that? Like, it's

just a lot of flexibility, but we can also do is, we can change the priority and which would classify as well. So that the nou knowledgebase Kohl's, who does the nou traversal? We can choose a different times to classify one higher than the other based on whether we think you're at the top of the experience where you might be asking more general question. And then, always your further down experience. We actually think you understand how it works. You're awesome. So fine, grain questions during a sales glow, you know, maybe I'm at that point we don't even want to be classifying against

the net for language. And then at a checkpoint when we do that, we have this ability contextually to understand where you are and what we think you're doing with the night and adjust the conference trash. All is well and we can do that with such buying grain sort of ability because it's contextual I know you and we know we're not trying to classify again. All of the intense whipping, very selective about how we actually do this. Chesapeake Ripper kind of rapping

rapping that often and so in conclusion a little bit so that's kind of a really high level and more than happy to take any questions about how we do this. But that's kind of has a high-level. How are you selected different intense based on context and contacts? This is kind of key to designing conversations and we use that context to reduce the clashes knowing that you're only so much. I was actually going to help us. And then how do we actually leverage that that contacts to adjust Comfort stretch out and all of this kind of comes together with that sort of human, and Luke

process to tighten that say today was slow. And I guess I'll be doing theirs and implementation the Roswell and Naples is sort of flexibility and versatility that we weren't able to achieve just by using Raza by itself. So we put this layout of the top of it, which gives us his Find rain, will we get in cards that are brandvoice business logic? And, and kind of ability to stop to Rum to be very selective about the type of experience that we use at a particular point. So that's how we design and build experiences.

Cackle comments for the website

Buy this talk

Access to the talk “What’s next in CDD: Intent Clashes and Selective Confidence | Rasa Summit 2021”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Ticket

Get access to all videos “Rasa Summit 2021”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Ticket

Interested in topic “Artificial Intelligence and Machine Learning”?

You might be interested in videos from this event

October 7 - 20, 2020
Online, Mountain View
19
5.41 K
google, googledevs, it, machinelearning, mlsummit, network, platform, tensorflow, tfx

Similar talks

Hans Van Dam
Co-founder and CEO at Conversation Design Institute
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Nikhil Mane
Conversational AI Engineer at Autodesk
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “What’s next in CDD: Intent Clashes and Selective Confidence | Rasa Summit 2021”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
742 conferences
30258 speakers
11303 hours of content