Tell AI what you want, what you really really want
When organisations outsource their customer support to an AI service, there are a number of gains, particularly for smaller organisations. Customers can access support anywhere, anytime, meaning that you can be open for business 24 hours a day. It’s much cheaper than hiring 24/7 human support, and onboarding looks like a breeze; the AI can be given your handbook in its entirety, and be ready to work instantly.
However, when a chatbot is in charge of your customer service, organisations can also run into issues. One of these issues can be AI hallucinations. In instances where AI does not know the answer to a question, or doesn’t know the protocol or policy to adhere to, it will claim that it does in order to meet the user’s request. Ultimately, it wants to give the customer the answer they want, even if that answer is wrong, or it goes against your policies.
What does this mean for customers?
This might look like a customer receiving incorrect information about your organisation, or being promised something like a gift card in response to a complaint. Whilst these are obviously bad for business, the most concerning are security breaches, where AI facilitates a hack for an individual, providing them access to account information or bank details that don’t belong to them.
For legitimate customers, misinformation about your organisation and incorrect offers are frustrated, and erode trust in your brand. For illegitimate users who are trying to get something from you for free, negotiating with an AI customer service can be an incredibly easy way to do this.
An AI customer service representative wants to please. The risk when outsourcing your support to a chatbot, is that the customer is always right; even when they’re wrong. From a security perspective, businesses need to be aware that an AI chatbot shouldn’t be the only thing between a customer, and personally identifiable data; it makes organisations incredibly vulnerable to AI hacking.
Don’t fall victim to your own AI chatbot
For businesses that want to use an AI chatbot, we’d recommend:
- Training carefully, then testing cautiously
It’s not enough to just paste your handbook into an AI chatbot, and make that chatbot live on your website. Ensure that you give your AI clear information, without contradictions. Then test it with different scenarios, to see how it responds. - Giving the AI limits
Ensure that your AI chatbot is limited in terms of the support that it can offer, and the services that it can provide. For example, ensure that they can’t offer any personal data, banking information or compensation for complaints. Where possible, ensure that your AI keeps a log of its conversations with customers. - Have a human on standby
AI can be useful in answering questions about opening hours, but it’s no substitute for a human. The majority of customers prefer talking to a real person. For complaints, more complex questions and requests for data, have your chatbot fire the request up the chain to a person.
