Impact of Generative AI on Customer Experience — How Talkative Is Revolutionizing Customer Service

Talkative — Digital services software for contact centers

As a provider of customer service software for contact centers, Talkative makes digital customer contact more profitable, productive, and personal.

An interview with Talkative CEO and Co-Founder, Felix Winstone

Tell us about Talkative and how the company has evolved over the years.

Talkative is all about helping organizations deliver great digital customer service.

While the technologies and tactics have changed since the inception of Talkative, this goal remains constant.

When we started the company, we wanted to bridge the gap between digital journeys (websites) and contact centers. That led to an initial offering of video chat and web chat, to provide human assistance through a company website.

We were perhaps too early with video — but then the pandemic changed the perception of video chat and its adoption. As consumers became comfortable with video in their personal/work lives, the adoption of video in the contact center followed.

We’re now seeing another consumer perception shift driving technology adoption in the contact center.

Chatbots used to be associated with consumer frustration — we’ve all had frustrating experiences where we get “stuck in a loop” as consumers when we just want to speak to someone!

ChatGPT’s arrival has changed consumer perception of what chatbots can do. Consumers now know what the technology can achieve. And we’re subsequently seeing a massive shift in how organizations are using generative AI to improve contact center operations.

You mentioned the impact AI has had and still has on your organization and offerings. Could you please elaborate?

Generative AI, specifically Large Language Models (LLMs), are trained with the fundamental goal of understanding language at the deepest level possible.

LLMs are quite unlike traditional computers. They are probabilistic, rather than deterministic. They’re not that good at calculation. But they are excellent at understanding and using language. 

Fortunately, this makes them great for customer service applications.

At Talkative, we’ve used LLMs to radically improve the effectiveness of three key areas of our offering:

  1. Responding quickly and accurately to customer questions
  2. Helping agents be as effective as possible
  3. Helping supervisors effectively manage a team of agents

Our product offering has therefore improved in all key dimensions. It’s been supercharged by AI.

Luckily, most organizations’ decision-makers have used ChatGPT themselves as consumers, so they “get” what is now possible.
Another big change for us is internalizing how neural networks work, and how they are replacing traditional software. For the last 50 years of software, we’ve become accustomed to rule-based, deterministic software written by humans. You write some rules, and you get a predictable output. But neural networks are a complete paradigm shift. Instead of writing code, data is now most important. Instead of having software we can interpret and debug, we have a black box that can give different output answers to the same question input.

What are the benefits of using generative AI for customer support?

The promised land of AI is that, in a not-too-distant future, AI can take care of all of your customer support. 100% of it. Instantly.

Imagine your best human customer service agent, with 10,000 years of experience, working 24/7, speaking every language, and answering every question in one second. That is what is coming.

In the short term though, we are obviously not there yet. However, there are still significant benefits to be had today, such as reducing customer wait times, increasing efficiency, and enhancing the customer experience — all while maintaining a consistent and high-quality output. Other specific benefits include: 24/7 availability, scalability, consistency, cost-effectiveness, speed, and efficiency gains.

For example, chatbots leveraging GenAI can now automate over 50% of interactions, compared to 10-15% on average for traditional “intent-based” chatbots.
GenAI excels in translation, so an added benefit is enabling instant multilingual support and real-time translation during interactions, breaking down language barriers and catering to a diverse customer base.

Data insights are important to note too. GenAI can gather and analyze customer interaction data, providing valuable insights into customer behavior, preferences, and pain points, which can inform business decisions and improve the overall customer experience.

AI in customer service can also aid agents during conversations and analyze vast datasets for real-time insights and decision making.

Through automation, AI enables businesses to operate more efficiently by saving time and reducing reliance on human support agents. It also enhances the customer experience by optimizing self-service options and ensuring faster, highly accurate human-powered support.

Earlier, you mentioned conversational AI, Felix. What exactly does it entail?

It’s admittedly a bit of a buzzword! In essence, conversational AI is an umbrella term for all AI technologies that simulate natural conversations between humans and machines. It’s primarily associated with automating customer service interactions.

Are there limitations when it comes to conversational AI?

As with any technological advancement, deploying conversational AI in customer service poses certain challenges, including data sovereignty and AI hallucinations.

However, these limitations can be addressed and overcome with strategies like integrating with an AI knowledge base. For organizations to be successful, they might need to give users the option to bypass their AI and speak to a human agent when needed. This not only ensures that customers are receiving the support they need in sensitive or complex situations, but also helps in maintaining customer loyalty and trust.

AI hallucinations refer to instances where AI systems like chatbots output incorrect, nonsensical, or misleading information in their responses. These shortcomings can easily be overcome by optimizing training data and implementing fallback mechanisms that trigger when the AI is unsure about a response. I’m pretty sure we’re all familiar with responses similar to “I didn’t quite get that, let me connect you with the team” when talking to a chatbot.

How are your customers applying GenAI? Could you give us some examples?

There are currently three main categories of GenAI in the contact center:

  1. Chatbots
  2. Copilots
  3. Supervisor assistants

Chatbots are the most obvious use of GenAI. Our Talkative GenAI chatbot enables organizations to import URLs from their company websites into their personalized AI knowledge base, along with articles, documents, and additional resources. Subsequently, the bots can assimilate this information and address inquiries about our customers’ business, products, and services, whether received via their website, app, or social media platforms.

The key advantage of basing responses on a knowledge base is to minimize hallucinations and ensure you’re only sending business-appropriate responses.

GenAI chatbots offer two key advantages over traditional intent-based chatbots:

  1. Greatly decreased implementation and set up — you just need to add your website/data. Intent-based chatbots require you to list out every single response, and (even worse) every single way a customer might ask that particular question.
  2. Greatly increased response rate — GenAI has excellent understanding of user questions, even if they contain typos or are in another language! A good example is if a customer asks two questions in one message. A traditional intent would fail here, but GenAI responses excel. This increased response rate in turn leads to increased resolution rates, which means happier customers, and a more efficient organization.

We’ve deployed GenAI chatbots across retail, financial services, and local government, and the benefits are clear and immediate.

Copilot/agent assist tools are very popular as they provide customer service teams with real-time information and recommendations during interactions. At Talkative, we provide an AI copilot, which uses the same AI knowledge base to suggest answers to agents. Agents can also ask the copilot for recommendations or for answers to questions. The AI Agent Rephrase tool also helps ensure that agents provide accurate and consistent information. It works by offering improved or revised versions of agent messages during chat interactions. Instead of having to manually edit and proofread each live chat message, agents can choose a preferred tone/style and have the AI do all the work for them.

Where AI chatbots can be thought of as doing an agent’s role, Supervisor AI is designed to do the work of a supervisor. Gradually at first, through summarization and insight reports, but doing more and more of this function over time as capabilities improve. With Talkative, every interaction (chat, video, SMS, etc.) is instantly summarized, both individually and collectively. This is valuable for quality assurance, training, and compliance.

AI excels at processing vast amounts of information very quickly in order to extract meaningful patterns or trends. Our clients are using our platform to automate reporting and to gain detailed insights into analytics, efficiencies, and 
the performance of their customer 
service teams.

Another supervisor tool we recently introduced is AI Agent Training. It trains agents by pitting them against chatbot simulations of customer interactions. Agents can practice customer communication, troubleshooting, issue resolution, and navigating difficult conversations — all within a controlled, risk-free environment.

GenAI is here to stay, and organizations worldwide should embrace it to remain competitive. It is going to play a pivotal role in the future of business and customer service.

While GenAI is already capable of a lot of things, it’s not at a point where it can fully automate your contact center, at least not yet. You still need humans. This is where we try to add value with our platform. We make it easy for organizations to try, use, and review AI alongside existing operations, and give teams the insights to understand its effectiveness, and how it can be improved.

Would you like to add anything else, Felix?

One common misconception we see is that organizations underestimate the acceleration of capabilities in AI over the coming months and years. They assume that GPT-4 is as good as it’s going to get.

The quality of LLM improvement is highly predictable and is a function of i) the amount of parameters used in the neural network, and ii) the amount of training data. This is why companies like Microsoft are spending $100Bn on data centre build out. It’s because the advancements in intelligence come “for free” without any progress in algorithms, solely through increased computation and data. There are currently no signs of this trend topping out.

This means that the models in the coming months and years will be substantially improved.
We’ve seen this firsthand. In late 2022 we used OpenAI’s API for Agent Assist with an early version of GPT-3. It was not good enough for customers to use (they tried). Little more than a year later, GPT-4 Turbo provides highly accurate answers, and our customers give it rave reviews. Putting something measurable around this, the context window has grown by 32x!

Today we see analysts saying things like “GenAI is not ready for handling customer interactions” but we believe this is short-sighted.

The current limitations (e.g. hallucinations) are transient and will no longer be issues as models become more powerful, more controllable, and more reliable.

Now, the question for many businesses is whether to integrate GenAI. Soon, the question will be why they didn’t adopt it sooner.

To learn more, visit