Hi, everyone. My name is Marina Goplacheva. I am a product manager. I've been working on a few AI products for several years now, specifically enterprise applications in HR tech domain. And today I'm going to share with you my professional experience and what I've learned from building conversational AI experiences in HR tech. First of all, just briefly, I would like to say a few words on what exactly I mean by conversational AI. I think it is important because the terminology in this space is somewhat fuzzy. Although, it's not that new. Some years ago, it was mostly chatbots. It is the application of machine learning that enables people to communicate with mobile apps, interfaces, devices, et cetera, using your own words. It's written off spoken. The focus is also on producing natural and seamless conversations between people and computers. And usually those interactions are designed as human as possible. And voice has become a very common modality in this space for several years now. And I think the reason that in this space we do love voice so much is because how we perceive something that can speak. as somewhat human and intelligent. That's why I love this quote by Judith Shulevitz about our minds respond to speech as if it were human no matter what device it comes out of. But I'll talk later of, about how it's sometimes difficult to live up to these expectations when you design conversational AI experiences. I think conversational AI. is one of the best ways to deliver value in HR tech. I've been working in this space since 2019 and at that time it was not something common to use conversational AI in HR. And I do think at that time that was quite progressive thing to do. If you look now at biggest vendors in HR tech, you'll see companies like Oracle, Workday, SAP SuccessFactors, ADP, all trying to move into this direction of conversational interfaces as well. why do I think conversational AI is the best way to deliver value in HR? first of all, the ability to use your own words make a huge difference. Employees do not need to remember a specific term or the name of a service or a specific app to get to what they need. And it usually is the case with HR services. As someone who worked in HR tech for years, and I love HR, I always saw this gap of understanding between HR and employees in terms of terminology. So removing this obstacle by giving employees Ways to get to what they need using their own words and understanding is, I think, a great thing. The other thing is the one interface, which is no interface in case of conversational AI, and how we usually say that it's an interface less UI or in case of conversational AI. And I think it's another big thing if you work in any big enterprise. I'm sure you know the mess most companies have in terms of, how many tools they have, systems, apps. and that makes it very difficult to mitigate for employees. So removing this mess and giving them to use something that has one interface. That's one. Interfaceless UI is also, I think, a great thing. The other thing is the ease of use in terms of UX compared to traditional HR platforms. You usually have to invest time and effort to teach employees to use these systems. Also, if you've ever been a part of any HR transformation, Associated with new implementation of a new HR platform, you will know that a lot of change management is required. Conversational AI takes a lot of the change management hassle out of the equation. You also can build conversational AI experiences on top of what you have without re-engineering existing systems and. Finally, these ability of what we call meeting users where they are so that they don't need to interrupt their usual workflows. We call those, systems, apps, messaging apps where employees, which employees use in their everyday work surfaces. So integrating conversational AI. on all those surfaces seamlessly is another great thing, that makes Condensational AI great to use in HR domain. I've designed Condensational AI skills for nearly every aspect of HR, whether it was onboarding, lead management, compensation, performance management. I'd say not every single aspect of HR processes can be easily designed in a conversational interface. There are some things that are still maybe better done in a full UI content rich interface. But even those can be incorporated and I'll talk about it as well. So how do you build conversational AI experiences? First of all, I think no matter how great your tech stack is, good design practices are essential. In my view, it is foundational for designing any kind of tech product which is used by humans. And the experience of interacting with the AI product can be fundamentally different than with traditional software products. Hence, you have to design for that. If the design is not up to the level of the technology, that itself is a barrier for making the most out of these products. It is a barrier for adoption and overall results in poor experience. And also I think part of this problem with design practices for AI is that there is a large knowledge gap and perhaps cultural gap between people of a UX background and people of a machine learning background. So we need to think of ways to minimize this gap. Try to find ways for UX designers and ML experts to collaborate effectively. I'm certain that UX designers nowadays have to understand a lot better what's happening in the backend of ML products. You also have to design conversational experiences with an understanding of how conversation works. There are some scientific foundations of how human conversation works. We also need to remember things like models tend to degrade and the performance of the algorithm goes down with it. So how do we design for that into the product itself? To make sure that we are still delivering the performance there over time. Researching and understanding your users is also still necessary. I remember when I was designing voice skills for AI voice assistant for internal employees. For I was biased towards asking questions within a voice flow as if it was a logic flow within the UI app equivalent of that flow. And there were users who were experienced at using assistants like Amazon Alexa a lot at home. So those users interacted with voice skills as they were talking to a human. But those two are hugely different mental models. And it's a huge difference in how you could, you would go about designing that AI product for two different groups of users. So you have to research how people use your products or use AI products in general. Just having great technology does not mean it doesn't need to be designed and applied properly. In order to build personalized. conversational AI experiences at scale, you need to leverage the context of the users. To do that, you have to invest in fixing your data. And I'm not just talking about data quality, but in building an additional layer. That is, you have to have a high quality codified descriptions and dictionaries of the data of your employees. because your LLMs have to query that context. Another thing I would suggest is to maybe supervise the process of AI development, not just the outcomes, especially when you are designing complex compound conversational experiences. and in cases when you are not, Using public models, but starting from scratch. Think whether it makes sense and it's feasible to set up experiments involving people in the development process. It makes perfect sense to do that when trying to AI, apply AI to complex decision making. It will give you invaluable insights into what expert human does and what is the ideal process. And the more granular your understanding is, the better. My next point is on the levels of anthropomorphism. When I started working in conversational AI space back in 2019, I quickly realized that a lot of time and effort in that space was spent on trying to find ways to make people feel comfortable with this technology by putting a human face on it. I think I spent hours every week in conversations about how to design the conversational interfaces, trying to mimic how humans speak. And it sounds logical, because when you introduce language or speech into the equation, you immediately have this perception of these things as intelligent. The question for me is, that level of anthropomorphism is needed. Is it? necessary? Is it going to speed adoption or make for a safer adoption? Or is it just setting bad expectations right away? And for me, the answer is, it depends on what you're trying to achieve. When, for example, I analyzed how people used the AI assistant we developed for our internal employees. We realized from the data that people generally use conversational AIs for utilitarian goals. And their wording and language is reflected in that. AIs are tools for them. So they cannot be more than that, because humans cannot make emotional connections with them. And frankly, I don't think people should make emotional connections with things. Moving to my next point. I think you do not have to limit yourself with modalities and concepts like AI thirst, voice thirst, anything thirst. Yes, conversational AI is about natural language, but it does not have to be pure natural language. Later in my career as conversational AI product owner, my team and I did a great job Incorporating micro UIs into conversational AI skills. And so combining natural language models with graphical UIs, you can have the best of both worlds. Especially in HR, it just makes sense for main things to be able to have both. Again, meeting employees where they are. On devices, platforms, systems where they work without the need to break their usual workflows. I think it's essential. For many years, people like me in HR tech space struggled with this one big thing. That people do not spend their time on HR, in HR systems. They primarily spend their time in other systems, in case of knowledge workers. Or in case of blue collar workers, they don't spend that much time in any system at all. this context in which interactions occur shapes the perceived utility of success of that experience. there was this one example of an experience, we developed for some of our employees some time ago. And I won't go into specifics, but it failed miserably just because we were so focused on the problem itself. And we missed the most important thing, that in order to use that conversational AI skill we developed to solve one of the pieces of work they were doing, and it was some tedious stuff. We didn't know, we did know they didn't like to do. But, we ended up with people not using it because it required these people to interrupt their usual workflow. To go to open the assistant on some other device, and it was just not worth it. This interruption just made no sense to them. I think this thing of meeting employees where they are is also crucial. And the last thing I wanted to mention is that the important thing to remember is having AI doesn't mean we need to ram it end to end AI everything. We still need to recognize what is good for and what is good for and use it to power the most effective experience. Especially in a regulated domain like HR, there may be cases where it may be better to leverage rigid or more traditional technology. For example, with issues around payroll, you may need direct controls there. So there will be cases that will live outside of the AI components. And you can do a lot by how you design the things that you do and how you create the user buttons and the mental model for the person using conversational AI. So I think the AI first craze should not get in the way in cases where we have to put security controls in place. In conclusion, and to wrap it up, I think yes, as I said. Design practices are crucial and they are key no matter how powerful the technology is. And you have to remember that conversational AI or any AI is a tool that should be aligned against a specific business objective of what you are trying to achieve. And it's critical to consider the state of your AI. HR data, tech stack and risk tolerance when it comes to implementing conversational AI. That's all from me today. I hope you found something helpful and useful in what I've talked about. And thank you for listening. If you're interested in maybe talking or reaching out to Just find me on LinkedIn. I'm always happy to talk all things conversational AI or anything you want to talk about. thank you. Bye bye.