During recent years, we’ve seen a lot of chatbots trying to help us find our flight details, get answers to our questions, or argue a faulty bill with the teleoperator. At least for me, 90% of these conversations lead to a place where I try to get the bot to call a human to the chat. But could GenAI software actually help here if implemented differently?
Learnings from the robots of the past
Back in the “good old days,” I remember a time when I was faced with bots on customer service phone lines all the time. Usually, these interactions proceeded with me trying to use single-word voice commands to get past the 10+ steps and reach the actual human after the IT-bot obstacle course. After some time, these bots ceased to exist mainly because customers' frustration was way more significant than the hypothetical saved money in customer service representatives.
I see many of the same ideologies, hopes, and dreams happening nowadays in the AI scene. Everyone wants to have an AI bot chatting with customers and helping them out. But what is the actual bang-for-the-buck that these things can offer? I don’t see profit in a model where chatbots are used as a first step to waiting in line to be served by an actual human. There must be a better way, and at Vincit, we’ve been involved in multiple such projects.
AI capabilities in customer service
GenAI software can help in multiple ways, a few of which involve the direct customer interface.
GenAI software can:
-
Summarize what is said in a longer text
-
Detect emotion and what is needed from a message
-
Quickly go through a lot of data to find answers to a question – semantic search makes the results much more likely to hit the spot
-
Ask for missing details to make sure all data is available to be able to solve an issue
-
Give step-by-step instructions based on human answers, making complex sequences of actions more easy to do
Out of these, semantic search is the one that can be used in direct contact with customers quite easily. Leveraging this opportunity requires a lot of security protocols and instructing the chatbot to answer in a way that respects the brand and always responds according to the company tone of voice. Therefore, I wouldn’t start there.
Out of the many things that GenAI software can do, where does the benefit come from? I believe the fastest way to profitable usage of AI tools lies in the hands of internal users having their own assistant to ease mundane and repetitive tasks. To get more time to focus on the things that GenAI software cannot solve independently. Not starting from scratch and relying on a single person's memory or ability to search for the right thing.
Vincit’s approach
At Vincit, we usually approach these things by mapping out the current customer service process and systems. Customer service is typically handled in a ticketing system that has structured data. Also, customer service agents have databases to search for information (for example, an intranet, a database of earlier question + answer pairs, manuals, or instructions of use) and some CRM or ordering system where the company orders and customer data live. So, there is a lot of information we can give GenAI software to process.
In customer cases, we’ve seen in the past that a common problem is ensuring uniform quality of service. People’s abilities to search for needed information fluctuate from day to day and different people behave differently no matter how much you train them. GenAI only improves its results over time. It never gets tired and the service level is always up to par. So using GenAI tools to for instance search for customers’ details and order history before giving the ticket to a human is beneficial.
We can also make GenAI tools search through data of previous answers and generate a response based on those. Humans can then determine if the response is good or if it needs tweaking. This tweaking and tuning is way faster than starting from scratch and reading multiple Q&A pairs to form one new answer.
Before determining if the GenAI tool’s generated answer is any good, it's important to have a quick look at what the question was all about. Capabilities like summarizing and detecting tone/emotion and the basic need expressed from the original message can speed up the process a lot.
Let’s discuss
From my experience, I see that AI can be truly helpful in customer service duties – but I recommend that it be used as a helper for internal users. This will realize the value much faster than going through the hassle of creating a public chatbot for users. Internally, people are doing their job and are mindful about what to post to customers and take personal responsibility for their actions. Externally, a GenAI chatbot can be a brand risk, although this can be mitigated to some extent through extensive testing and instruction. The bot is limited to its instructions and database. It has no responsibility for its actions – “no brain” to think things through.
If you’re wondering where to start on your AI journey, may I suggest we meet up and think things through together? The topic doesn’t need to be customer service, but I see a lot of potential in that area. Our experts have seen and tried multiple solutions with our customers and within their own test benches. We have over 100 Azure AI-certified cool and fun individuals who can be called on to help you succeed.
Email me or contact me via LinkedIn to chat more about AI possibilities.
Taina Arjanmaa,
Design Lead, Senior Consultant, Vincit
