KiaKia Chatbot Virtual Assistant 1.0

At Bernadette, digital innovation is at the core of everything we do. That’s why we put time and talent development into those tech disciplines that we think are going to define the digital landscape of tomorrow.

Conversational AI is a key part of that vision. So, when Kia approached us with the intention of enhancing their existing chatbot solution, we were confident we could set them on a path towards something truly special.

Digital conversational experiences, or chatbots, come with a set of unique challenges. Crucially, there’s a perception that modern chatbots simply aren’t functionally capable enough to satisfy their customer and business needs — seeming too stupid and ‘clunky’ to carry context-rich conversations all the way through to fulfilment. And they lack any form of unique brand specific personality - despite them being a virtual member of the team.

The advent of sophisticated language models capable of leveraging NLP (Natural Language Processing) is already causing these rock-bottom expectations to gradually shift. However, this new wave of deep learning, generative AI brings has given rise to another challenge for brands to consider.

The impressive nature of language models like ChatGPT and Gemini is rooted in their ability to make clever predictions. But, paradoxically, that means they also come with a tendency to output responses — both in terms of tone and reasoning — that are sometimes hard to definitively predict.

In real-world scenarios, that could mean a chatbot going off-script in ways that are detrimental to customer experience at best, and damaging to brand capital at worst.

Part of our challenge then, was to develop a robust strategy to ensure that Kia’s Virtual Assistant operated within strictly defined parameters so it always stayed on-brand and on-topic. At the same time, the chatbot needed to exercise just enough flex in order to be fresh and adaptive so as to provide natural, non-repetitive answers in real-time. This meant we needed to demonstrate a delicate balancing act and a nuanced approach to our conversation design strategy.

Specific to Kia’s business objectives, they understood that their existing live chat system was in need of a significant overhaul. This solution was using basic logic and pre-scripted replies in a way that was primarily reactive. By functioning more like an answerphone, it meant that only a small percentage of traffic was able to engage with human agents, with the rest being forced to leave messages. And the metrics revealed this inefficiency: only 0.66 to 1% of site visitors engaged with live chat, and of those, only 18-23% left messages.

Our initial challenge then would be to deploy a chatbot to replace this system and hit the ground running — with a capability of engaging with every customer, all the time. It needed to stay true to Kia’s brand guidelines, and deliver a near-exhaustive range of responses regarding Kia’s products, particularly the EV6 and future models.

And, beyond simply answering queries, the assistant needed to be proactive and versatile, steering conversations towards key brand features, guiding users to relevant specific destinations on cue, like booking test drives and connecting users directly to dealerships.

Our shared vision with Kia was to create a best-in-class, customer-centric virtual assistant that could:

• Intelligently respond to a wide range of intents while handling customer queries autonomously.

• Uphold a consistent, brand-aligned tone of voice that feels human-like and trustworthy.

• Go beyond traditional FAQs by offering proactive recommendations and seamless handovers to human agents when required.

• Provide end-to-end resolutions for key use cases such as connecting users to dealerships, booking test drives, or sending brochures, all within the chat interface.

Our approach was a collaborative effort between Bernadette and LivePerson — Kia’s Conversational AI platform and chatbot implementation partner. Their robust technical foundation and scalable platform, gave us the flexibility to define and apply a precise tone of voice that felt uniquely aligned with Kia’s brand values.

Having been closely involved in crafting Kia’s tone of voice across other, more traditional comms channels, we were well placed to take the lead in this area. So, we formulated a combination of 4 principal cornerstones that would help contain conversational output. The first three, speaking to a sense of optimism, flow and essential purpose, were already inherently on-brand because we’d applied them as part of Kia’s CRM and Web copy guidelines. The fourth and final cue was specific to the conversational nature of our AI: making it demonstrably capable of impressive knowledge and reasoning.

We went on to define each of these guidelines in granular detail, writing pre-scripted answers that simultaneously conformed to each principle in a way that was balanced and well phrased. These would then be used as baseline sample responses to help train our language model.

Beyond defining tone, we also wanted to reinforce our Virtual Assistant's behaviour by embedding it with a set of complementary operational ‘persona traits’. So, while the ToV principles inform the assistant’s manner of speaking, these traits were designed to reinforce the chatbot’s behaviours within conversation flow. They’ve proved super helpful as both back-end prompts for the AI component to better understand its own core directives, as well as informing guidelines for manual conversation design.

Both the ToV principles and persona traits were integrated into LivePerson’s platform and, to ensure our risk mitigation strategy had proved successful, we implemented a rigorous test plan to verify the chatbot’s output. This test and learn approach involved continuously adjusting prompt structures to fine-tune the output until we were confident enough to deploy our first iteration of the Kia Virtual Assistant to its live environment.

But, this is just the beginning — and we’re continuing to look at this long-term comms solution with future scalability in mind. While the initial release centred on core use cases, our prescribed framework is ready to support more advanced features down the line including:

• Proactive recommendations, where the assistant anticipates user needs based on interaction history or preferences.

• Expanded knowledge domains, integrating deeper insights from Kia’s marketing and CRM systems.

• Personalised experiences that recognise returning users and tailor interactions based on previous engagements.

This strategic approach was underpinned by our passion for immersive conversation design not to mention the dynamic, adaptive potential afforded by a bespoke, well-trained language language model.

This project has offered an exciting opportunity to blend Bernadette’s trademark technological innovation with creative strategy and copywriting. When it comes to exploring the details surrounding a new Kia EV, we’ve really enjoyed mapping and optimising the various flows a dynamic conversation might take.

It’s set the stage for a Virtual Assistant destined to meet the needs of Kia’s customers and hopefully exceed them. We’ve aimed to explore the baseline potential of conversational AI and asked ourselves just how far we’re able to push it. This has sketched out an exciting strategic roadmap that would see our chatbot take more human-like behaviour as its functional capabilities evolved over time.

Conversational AI is a truly exciting digital space to work in and this initiative has proved to be much more than simply adding a customer support channel. It’s really transformed the way Kia is engaging with consumers online. As customer behaviour increasingly trends towards self-service and real-time assistance, Kia is now well-placed at the forefront of this shift toward smarter machines capable of serving as reliable brand ambassadors.

We soft-launched the Kia Virtual Assistant in June 2024, marking the beginning of an iterative journey towards establishing a best-in-class conversational AI. The impact of this initial phase has been multifaceted:

• Improved User Engagement: With the new assistant, user engagement levels saw a noticeable increase, with more users completing interactions within the bot and fewer requiring hand-offs to human agents.

• Tone Consistency: The assistant’s responses consistently reflected the brand’s tone of voice, creating a more cohesive digital experience that users found engaging and trustworthy.

• Operational Efficiency Gains: By resolving more queries autonomously, the assistant reduced the reliance on human agents, freeing them to focus on more complex interactions.

• End-to-End Service: The integration of key features, such as direct dealer connections, test drive bookings, and brochure requests, allowed users to get the information they needed without leaving the chat interface.

• Future-Proof Foundation: The architecture and conversational design principles established during this phase ensure that Kia’s virtual assistant can continue to grow in sophistication, integrating new features and capabilities as required.

With this strong foundation in place, future iterations of the assistant are primed to focus on enhancing its proactive capabilities, expanding its knowledge base, and integrating more personalised interactions. The ultimate goal is to set new benchmarks for conversational AI within the automotive sector and for Kia’s Virtual Assistant to become a mainstay brand experience for years to come.