AI-Powered Customer Service: Insights from Roman Lutsyshyn on the Future of Support and Engagement

15
Mar 24, 2025

The Analytics Inside podcast explores how AI is transforming customer service amid budget constraints and rising customer expectations. While predictive AI has been around, generative AI has recently taken center stage, prompting businesses to develop strategic approaches while managing costs. In this episode/article the discussion delves into AI-powered customer service, its impact on customer experience, and expert insights from Roman, including an introduction to CoSupport AI.

Block 1. CoSupport AI & Visionary Leadership

Can you help us understand more about your company and its specialization? Is there any particular service or solution the CoSupport AI is based on?

Roman: Yeah, sure. CoSupport AI is not only about support. It's more about the combination of customer experience, customer support, and optimization plus automation. So these three main topics, these three main core ideas are underlying in our company. So basically, we try to make the customer support process as much optimized and automated as possible. At the same time, we want and believe that we can improve customer experience also pretty much.

I’m the co-founder and CTO of CoSupport AI, and my journey into this space started with my background as an ML engineer. Back in 2020, my partners had a business problem, and I was just an enthusiastic engineer who believed that no technical challenge was impossible.

What was the challenge? Customer support automation and optimization - how to reduce response times without compromising quality. I proposed a vision, built the first proof of concept, and from that moment, we knew we had something worth pursuing. That’s when we decided to establish a company.

Initially, we launched as NLP Lab.ai - a natural language processing research lab focused on advanced AI solutions for customer support. Over time, we saw the potential to build a standalone product, and CoSupport AI was born. It integrates our most advanced NLP technologies to enhance customer interactions, automate processes, and improve response quality at scale.

I’m always exploring cutting-edge technologies and applying them to our product. In fact, we’ve already secured a patent for our content generation architecture, and our second patent - for preprocessing - is currently pending. We live in a time of rapid innovation, and at CoSupport AI, we’re committed to pushing the boundaries at every step.

Block 2. AI’s Impact on Customer Support

How do you see AI transforming the customer landscape today or how do you think the customer support landscape has evolved? What role AI technology has played in customer support?

Roman: Time is crucial. A few decades ago, customer support via phone calls was the norm, even if it was time-consuming. But the world is changing rapidly. Today, more and more people prefer instant support through chat, email, or automated responses. Convenience is key. Many customers would rather interact with a chatbot or send a quick message than wait on a call with a human agent.

To what extent are AI chatbots used in customer support? Are customers more inclined to converse with a chatbot rather than a human representative? How do AI-powered chatbots and virtual assistants improve response times and reduce the workload for human agents?

Roman: This is a great question. I would like to start from the point that humans, from their nature, are pretty lazy. Usually, many people just don't want to read the FAQ section or documentation at all. Let’s take a step back to our experience with the NLP lab, where we weren’t just building a product - we were analyzing the entire customer support domain. Through this, we identified a few crucial insights, let’s check examples.

First, on average, more than 70% of customer requests fall under Level 1 (L1) support. These are common, straightforward inquiries that can often be answered using an FAQ section or documentation - if those resources are available.

Second, even Level 2 (L2) technical inquiries can be addressed with AI-based solutions to some extent.

Now, coming back to your question - how can AI-powered chatbots and virtual assistants improve response times? The short answer is that AI chatbots can handle 70-80% of incoming customer requests, allowing human agents to focus on complex, high-value issues instead of spending time searching for answers to simple or repetitive questions. AI is an ideal solution for this.

Additionally, AI assistance inherently reduces response times. Take, for example, our Co-Agent solution within CoSupport AI. Rather than responding directly to customers, Co-Agent generates suggested responses for human agents to review, proofread, or modify before sending. This drastically cuts down on response time without removing the human element.

Beyond end-to-end automation, even AI-assisted support significantly reduces operational costs and response times. And, of course, fully implementing AI in customer support can lead to a dramatic reduction in both workload and response times for human agents.

What strategies would you suggest for businesses to ensure a smooth AI-to-human handoff in complex cases?

Roman: Yeah, I believe the entire AI domain relies on two core factors. The first is computational resources - CPUs, GPUs, and other hardware components. But the second, and in my opinion the most important, is data.

Now, getting back to your question - what strategies should businesses adopt, and how should they prepare for AI integration and implementation? The short answer is data.

The long answer? Businesses need to collect, validate, and clean their data as much as possible. In the AI era, achieving precise and accurate responses depends on providing the model with a solid knowledge base. Whether you’re fine-tuning a model or using techniques like retrieval-augmented generation (RAG), structured and clean data is essential.

There’s a simple principle in AI: Garbage in means garbage out. This was true five years ago, and it’s still true today. Yes, advancements in large language models and state-of-the-art techniques can reduce the impact of bad data, but at the end of the day, if you feed an AI system low-quality data, you’ll get inaccurate or unreliable responses.

So, to sum it up - businesses need to collect, clean, and structure their data as much as possible. At CoSupport AI, we offer solutions for data preprocessing, cleaning, and anonymization using a range of AI pipelines and neural network ensembles. But that takes time and money.

My recommendation? If you have the ability to collect and improve the quality of your data now, do it. You’ll thank me later.

Block 3. Adopting AI-Driven Customer Service Solutions

What ethical considerations should businesses keep in mind when deploying AI for customer service? How should companies address these ethical concerns when adopting AI-based customer service tools?

Roman: This is somewhat of a philosophical question, but I would separate AI solutions and AI-generated models into two main categories. The first category includes solutions that are accessed via APIs, such as ChatGPT, Anthropic models, Google Gemini models, and others. The second category involves in-house or semi-in-house pipelines and models, which are trained and deployed on data within the client’s infrastructure.

From an ethical standpoint, I would say that sharing personal data with companies like OpenAI or large language model providers is not ideal. Even though OpenAI states on its website that user data may be used for retraining or fine-tuning models, you still don’t have full control over how your data is used. You can trust these companies, but that doesn’t align with the level of control businesses should have when implementing AI solutions internally.

The second option - using in-house or semi-in-house models - is, in my opinion, much better. It provides more control. With this approach, businesses can ensure that their data is being handled responsibly. They can decide where the model is deployed, who has access to it, which data is used for training, whether data is anonymized or not, and whether personal information, such as first and last names, is used.

Having this level of control means you can have more confidence that your data will remain secure and that there won’t be any data leaks.

Can you suggest some tips or advice on how businesses should prepare for the integration of AI into their customer support strategies?

Roman: Yeah, that’s a great question. As I mentioned earlier, data is crucial. Clean and structured data is fundamental. This is point zero, the core element that businesses need to have in place for AI systems to be successfully implemented and integrated.

The second point is that, in the next several years, I believe we’re going to see the rise of autonomous AI agents. Right now, we hear a lot about AI systems, AI agents, AI experts, and similar concepts, but in the near future, we’ll see AI agents that are truly autonomous - able to interact with each other, integrate with multiple systems on their own, and perform tasks without human intervention.

To quickly explain, an AI agent is essentially a system built on a language model that can take in requests or inquiries, process them, and take action. For example, an AI agent could go to a weather website, retrieve weather information for a specific location, and return that data to the user. That’s a simple example of an AI agent in action.

However, I believe soon, we will have semi-autonomous or fully autonomous AI agents capable of interacting with each other. They could, for instance, approach another AI agent to retrieve information it might need. This means that businesses will need to prepare their APIs (Application Programming Interfaces). Why? Because if you want an AI agent to check the availability of an item in your warehouse, for example, it will need access to your warehouse’s API. Without an open API, the AI agent wouldn’t be able to retrieve that information.

To clarify, APIs should be publicly available - though still secure with authorization tokens and other security measures. This will allow the AI agent to interact with your systems, just as a human would, to fetch information, process it, and provide a response.

To sum up:

  • Data: It should be clean and structured as much as possible.
  • APIs: They should be prepared and ready for integration with AI agents.

In short, these are the key areas businesses should focus on.

Block 4. Future Trends

How do you see AI evolving?

Roman: The first point is that I believe AI will handle a much larger volume of chats, chatbots, and communication interfaces. The second point is that I am confident we will see a rise in voice-based solutions, which will also manage a significant portion of voice assistance tasks.

The third point is that I’m sure AI will continue to focus on making models smaller and more efficient. Currently, large language models (LLMs) require substantial GPU, CPU, and overall hardware resources. However, I believe we will see a reduction in resource usage, leading to more efficient AI solutions that can be widely integrated and implemented into businesses.

As a result, I expect that interactions with AI chatbots will feel just as natural as conversations with humans. In fact, I believe the accuracy and precision of AI responses will continue to improve every day.

Summary

AI is transforming customer service by shifting the focus to engagement, improving overall customer experience, reducing costs, boosting sales, and maximizing long-term value. Now is the ideal time to leverage these technologies to enhance customer support and deliver more value to customers. Thank you, Roman, for joining us and sharing these valuable insights. It’s been an enlightening conversation, and I’m sure our listeners have gained a deeper understanding of how AI is reshaping customer support.

The full version of the podcast can be found here.

Want to learn how

leverage the power of AI for your customer support and business?

Please read our Privacy & Cookies Policy