Chatbots seem like the perfect assistants we’ve longed for, but they might run loose and run wild. As leaders and innovators, you’ve probably seen both the impressive and the not-so-great. Whether you are leading a team or steering a company, understanding the chatbot rollercoaster is key to unlocking their true potential. Buckle up for the ride!

The Evolution of Chatbots 🤖

The evolution of chatbots marks a significant shift in artificial intelligence, profoundly influencing business-customer interactions. This journey started with ELIZA back in 1966, a basic chatbot that mimicked a psychotherapist using pattern matching, and has since undergone dramatic transformations.

The last decade's advancements have been particularly transformative. The technological breakthroughs in ML, NLP, sentiment analysis, and contextual AI have enabled chatbots to better comprehend human conversations and to continuously learn, facilitating their application in various commercial settings such as predefined answer matching and often serving as a navigating tool in customer services.

Recent breakthroughs in transformer models and Large Language Models, notably OpenAI's GPT series, have taken chatbot capabilities to new heights. These AI models facilitate more nuanced, human-like conversations and drastically expand the use cases of AI. Combined with audio processing technologies, today's AI chatbots can do much more than their rule-based predecessors, engaging in more dynamic and responsive interactions.

These advancements bring significant benefits: AI chatbots work at all hours of the day, handle complex information retrieval, aid in data management, contribute to process automation, and improve efficiency, often with reduced costs. But they are not without limitations. Issues like inadequate grasp of complex contexts, absence of empathetic understanding, and struggles with ambiguous queries highlight the areas needing development.

Now that we’ve seen how far chatbots have come, let’s dive into some real-life examples and see these digital agents in action.

The good, bad, and the ugly 🧐

Klarna's chatbot boosting customer service 👍

Klarna’s AI chatbot, impressively handled two-thirds of customer service chats in its first month, managing 2.3 million conversations and showing improved efficiency and matching satisfaction rates. It is available 24/7, in 35 languages, and resolve queries in 2min rather than the previous 11 minutes, estimated to drive a $40 million USD in profit improvement to Klarna in 2024.

Klarna teamed up with OpenAI in 2023 and invested a year in fine-tuning its chatbot, leading to significant breakthroughs. The chatbot goes beyond a basic online feature, representing a complex, tailor-made AI integration in its system workflow. With continuous data gathering, the chatbot is expected to improve its performance over time continually.

Chatbot agrees to sell a $58,159+ Chevy Tahoe for $1 💸

The integration of AI Chatbots for customer services is a growing trend and American auto dealers are embracing them. However, some of them are getting a hard knock. Being prompted by X Developer Chris Bakke, A car dealer’s chatbot agreed to sell at $1 and recognise it as a legally binding offer - “no takesies backsies”.

Imagine this being legally binding (is it?). To avoid unexpected answers like this, proper AI compliance design and continuous monitoring is essential. This incident is a call to action: to harness the power of AI with caution, ensuring that these advanced systems are not only efficient but also reliable and aligned with the business’s values and operations.

DPD Chatbot writes defamatory poems about itself 🚚

In a rather unexpected twist, a DPD chatbot took customer satisfaction to a new level by calling itself “useless” and using swear words upon a customer’s request. So although the chatbot can’t connect the customer to a human advisor, it can surely write poems to acknowledge its bad performance.

Even without professional prompt engineering, an angry customer can still push AI to its limits. This instance of AI ‘candour’ not only raised eyebrows but also underlined the complexities and customer impacts, indicating that more emphasis on empathetic interactions and preparing for unintended use cases could enhance chatbot performance.

Your Enterprise AI Chatbot Journey: Where to Start? ✅

Knowing the potential of AI chatbots for great success or setbacks, how should we start dipping our toes in the water as enterprise organisations embrace the new era?

• Start trying with prototyping chatbot tools: Using tools like Botpress or Intercom can be a great way to start prototyping and getting internal stakeholder buy-in with demos, but be careful about relying on these too much. These will fall short of custom-developed chatbots like Klarna’s.

• Build your in-house or external talent pool: Start thinking about your team to build, maintain and upgrade your chatbots. Custom builds will enable you to extract much more value and are more likely to pass internal infosec and governance gates.

• Prepare your enterprise API suite: The best chatbots will do more than just relay information but also perform actions (like a customer service agent). You need APIs for this - start preparing your enterprise APIs to give your chatbot the widest toolkit of actions.

• Implement a QA, evaluation and monitoring system upfront: Set your metrics, conduct tests and obsessively monitor your outcomes. You would do no less with a human team, it’s no different with a digital agent.

• Start Internally: If your organisation is new to AI chatbots, begin by testing them internally. This helps identify potential issues in a controlled environment.

• Contextual Understanding and Specificity: Ensure that your chatbot is designed for specific purposes and contexts. It should understand and respond accurately to the specific needs of your users or business area. This is something Koodoo emphasises. For the mortgage industry, we tailored our LLM with industry-specific knowledge to the extent that our AI passed the CeMAP test.

AI Chatbots sound great but they need to be carefully designed and monitored - and that means time is required. Isn't it time to start thinking about it now?

✅ Follow Koodoo on LinkedIn today for GenAI insights and real-world applications in financial services.