Chatbots have come a long way from the days when Microsoft’s Clippy would pop up, unannounced, offering 'help' but actually just disrupting whatever it was you were doing. As Mark Zuckerberg and Elon Musk know well, the latest generation of conversational AI chatbots promises a more personalised and relevant experience for users.
When done well, AI chatbots can deliver a faster resolution to issues, reduce costs and provide an “always on”, 24/7 service (and deliver zingy one-liners, of course). They are now being used for everything from website navigation support to supporting users with their mental health.
However, as the adage says, you never get a second chance to make a first impression. According to a report from Forrester, nearly 40% of all interactions with chatbots are still viewed as negative. One bad experience will, at best, lead your customers to try to use a human channel. At worst, they will buy from a competitor and tell all their friends about their terrible experience, with 30% of customers saying they will take this course of action.
So what are the biggest challenges when implementing a chatbot? And how do you overcome them?
01. Set the right expectations for users
Most customer facing chatbots are designed to work in a limited range of scenarios. Rules-based chatbots have clear limitations, particularly when users use language that hasn’t been pre-defined. It’s also important to understand the limitations of AI chatbots. AI chatbots are most helpful where they have been properly trained to deliver specific outcomes. Model training takes time and good data to deliver results, and the narrower the scenario the model is supporting, the better the experience is likely to be for the customer.
Setting clear expectations for what a chatbot can and can’t help you with, and how to get the best out of it, will help you to improve the experience for your customers. It’s critical, however, to make it clear how your customers can get help elsewhere.
If your business has high value or very vulnerable customers, you may want to consider a priority route to speak to a real human when needed.
02. Keep it simple
The internet is full of tales of chatbots gone bad. Many of these problems are caused by having a wide range of scenarios, trying to be too clever and giving the bot access to the wrong data. Take some lessons from Facebook’s chatbot, Blenderbot, which has provided hours of fun for people looking to bait it into making racist comments or having off topic conversations about Genghis Khan.
Even the financial might of Meta can’t manage all the potential questions that a user can throw at it. Giving the bot access to external data sources will increase the risk of erroneous and potentially reputation damaging conversations.
Be really clear what problems your chatbot is trying to solve, get the right data to deliver the experience and train your AI to deliver the best outcomes.
03. Use the right channel for your problem
Live chat? Facebook Messenger? WhatsApp? Website? The channel you use will depend on the problem that you are trying to solve. If you want to increase your audience on Facebook, then Messenger is probably for you. However, if you sell through your website, Facebook is absolutely not the place if you want to help your customers to find answers to product questions. It sounds like common sense, but you would be surprised at how many businesses direct their customers away from the channel they are already engaging with.
04. Contextualise the experience
The holy grail for Customer Experience technologies is to provide a personalised, contextual experience. This experience should be one where the technology understands the history, value and preferences of each customer and combines that with what the customer needs right now. One of the most common frustrations of chatbots is that it’s too hard to get a relevant response.
Ensuring that your chatbot has access to accurate, high quality customer data in your CRM so it can personalise the interaction is critical. Combining this information with what the customer is doing right now will deliver the best experience. This could include browsing history or reaching a point of friction in the online experience.
05. Consider security
With the UK under constant cyber attack, its critical that any public facing tool that has access to customer data is as secure as possible. The National Cyber Security Centre (NCSC) recently warned that chatbots are highly susceptible to prompt injection attacks. Simply put, a prompt injection attack is a vulnerability that allows a hacker to deliver unexpected behaviour within the bot.
Make sure you have clear policies in place about the use of chatbots. Work with your legal and security team to ensure that your chatbot meets the standards of regulations such as the GDPR and PCI DSS.
06. Measure the impact of your chatbots
There is so much potential for improving customer satisfaction, reducing costs and increasing sales when using chatbots well. Make sure that your chatbots deliver on the promise by consistently measuring the impact of the chatbot on user adoption, performance and business value.
Read about AI in our how AI is changing graphic design post and our round up of how AI changed art and design in 2023.
