Why chatbots are bad, and how to make them better
15 Nov 2024
6 min read
We hate chatbots.
Pretty weird thing for a company that builds AI chatbots for ecommerce businesses to say, right?
But it’s true. Chatbots have traditionally just been barriers between the customer and the brand, doing almost nothing for customers.
So, why did we build a business around chatbots?
Because we think they can be better. Just like humanity didn’t give up on electric cars after the disastrous Sinclair C5, now we have Tesla.
Let’s recap the history before we explain the future.
Current State of Chatbots
The current state of chatbots is a mixed bag. On one hand, chatbots have become increasingly popular, with many businesses adopting them as a way to improve customer service and reduce costs. On the other hand, many chatbots are still struggling to deliver on their promises, with customers often finding them to be frustrating and unhelpful. According to a recent study, 47% of organizations will use chatbots to automate customer care processes by 2024, but many of these chatbots are still plagued by preventable errors and a lack of understanding of customer intent.
Button-based chatbots
To call these chatbots is a bit of a misnomer, because there is really no chat involved. It’s a series of decision trees that a business has built to try and direct customers to a limited set of answers.
Think about how you approach customer service when you have a problem. Are you someone who looks to jump on the phone at the first opportunity? Or are you someone who looks around online for the answer first, only using the phone as a last resort?
Whichever camp you fit into, most chatbots are no good. They lack sufficient conversational intelligence and natural language processing capabilities, making them ineffective at handling nuanced or complex inquiries. If you want to talk to an agent, they get in the way. If you’ve been searching the website for the answer, the chances are you would have found the answer.
We’ve labelled these as “early” but the truth is lots of them are still around.
Buttons + Conversational AI
The next stage was to add a degree of “chat” to the chatbot by having some ability to describe your problem. The bot would then try to understand your problem and send you the relevant information – either by linking to a help article or some other tool.
Too often the “AI” was not all that intelligent here. It simply used keyword matching to try and bring you to what it thought was the right answer. For example, if you asked “Do you do next day delivery?” the bot might pick up on the word “delivery” and send you to an article about how long delivery takes. You could probably find the answer in that article but it wouldn’t be perfect.
Just as common was to see that the chatbot was unable to understand your question, and so you were stuck. Integrating robust business logic into chatbots can enhance their effectiveness, minimize unnecessary interactions, and streamline processes, thereby improving the overall customer experience.
The key thing was that this was still a generic chatbot. That is, the answers it would give would be the scripted, so it was unable to do anything unique to you, like look up an order.
Buttons + Conversational AI + Integrations + Proactive
Soon, there became much more of a push away from buttons and towards Conversational AI as the ability of the AI models to parse and understand the questions has become better. Buttons still exist of course, and can be useful as shortcuts for customers who don’t want to type their question.
The key step forward was to integrate the chatbot to other technology platforms, such as CRMs, Ecommerce platforms, and Shipping platforms. Adequate computational power is crucial in enhancing these AI systems, ensuring they perform reliably and effectively. Now, a chatbot can look up a user or an order based on an email address and order number and start to answer questions based on that information.
Now, with rules-based flows, customer service can start to be proactive in the background. For example, if deliveries are running late, the AI can do a daily scan to check on status and issue an alert if it appears the delivery has been stuck, offering the customer a solution.
This suddenly makes the chatbot personal rather than generic, and enables it to start actually solving problems. These are the roots of DigitalGenius, turning chatbots into something that can actually answer and resolve queries for individual customers.
Generative AI
In the past few years, artificial intelligence, particularly Generative AI, has been a big breakthrough, and suddenly up sprang chatbots that were built on LLMs. These turned what were more scripted conversations into much more free-flowing chats.
The downside was that if they were not tethered to anything, they just invent information, however mundane it seemed.
There is a temptation with platforms built on generative AI to just let the AI try to answer everything regardless of whether it “knows” the answer, or has access to the right answer. So we have heard horror stories of generative AI platforms giving customers insane offers or telling them plainly false information. All of which is hugely damaging to the brand’ reputation, and occasionally its bank balance.
But the ability of generative AI to be creative and move beyond the formulaic answers is incredible when used effectively.
Generative AI + Deep Integrations
Now, this is the forefront of where things are with chatbots and at the heart of what DigitalGenius does.
By combining generative AI with deep integrations (i.e. going deeper into complementary platforms to be able to access relevant information and perform actions), we know have chatbots that have free-flowing and engaging conversations, while actually being able to solve customer problems.
For example, it can generate new return labels, or find the exact status of different shipments, or even amend orders in the order management system. All things that are a far cry from button-based decisions trees.
Buttons can still be used at various stages in the process, but most brands will start to find that they just get in the way.
Take a look at air up’s chatbot and you can see what we mean. Or ask On to recommend you a pair of shoes. Or ask where your All Saints order is (assuming you’ve made one). You’ll see the difference.
Deep integrations are the key to making chatbots better. By better we simply mean that they have access to information and can resolve your queries. And more than that, they know when they are not able to solve your query, so they get the hell out of the way and let smart human agents fix things. Sentiment analysis is also crucial in enhancing chatbot effectiveness, as it helps in understanding customer emotions and addressing their needs more empathetically.
This achieves what chatbots have not been able to do. Solve most of the problems for people who don’t want to speak to agents, while making it easier for others to get through to human agents. Everyone’s a winner.
Agentic AI + Human Agents
The future is agentic AI. We’ve spoken about this before, but as we advance our AI capabilities, the ability for an AI to be able to do any action that you or a customer service agent can do grows and grows.
AI powered chatbots will be able to perform almost any task for you, even ones we haven’t thought of yet.
You might even have an AI agent asking a brand’s AI agent to do the work, saving you the effort.
Combining that with historic data about you, your preferences, and other information then leads to the possibility of a true AI concierge. An AI concierge has the context of what you ordered, and perhaps why you ordered it, enabling a fully personalised experience. On top of that, the concierge can be solving problems in the background that you didn’t even know were problems. AI chatbots powered by large language models, such as ChatGPT, have found widespread use across various applications despite being notoriously unreliable.
It’s closer than it might seem. Already our chatbots are able to resolve 88% of chats conversations for air up. Just think what might be possible in a few years.
To get your own effective chatbot, speak to our team.
