There are a lot of great uses for artificial intelligence (AI), but the one thing AI is great at is finding patterns in large data sets that are difficult for humans to sift through. Imagine how AI could assist in fighting climate change where you need to understand large atmospheric models, or in medicine where there is so much information about how the human body works.

In the contact center, AI is largely used for chatbots or routing, and identifying complex patterns about the contact center, business, products and customer. However, when you focus on keeping the human in the middle, you’ll find new valuable uses cases for AI.

Creating the Super Agent

Agent assistance is a great example of utilizing AI and keeping humans in the middle. Create Super Agents by giving them the ability to find the right information across any channel. Voice is still the number one channel being used in contact centers, followed by email. If an agent is on a voice call with a customer and searches in a knowledge base, they might come across three different articles on the same topic. AI can quickly read through the articles and offer the agent the correct information, ensuring the customer receives a fast and accurate resolution.

Humans generally have natural intelligence that allows us to know when a customer is happy or upset. But with AI, the sentiment analysis is automated and provides this data to agents and anyone else that might be involved in resolving the customer’s issue, like a supervisor or subject matter expert. They will be able to see how sentiment has changed over time for the specific customer and act accordingly.

There are a lot of great applications that will be used alongside a contact center desktop, including a CRM system like Salesforce or Microsoft Dynamics. But these come with new learning curves. The agent will need to be trained on how to use the CRM as well as the agent desktop and will need to know the ins and outs of how to add records. Or you can use Robotic Process Automation (RPA) to learn and automate those processes for the agent.

Risks of AI and Learning from Others’ Mistakes

The legal risks around AI are mounting. Companies like Air Canada, Facebook, and Patagonia have all faced legal battles around their use of AI.

Air Canada was taken to court when a customer asked their chatbot about bereavement fare. The customer received the wrong information and Air Canada refused to provide a partial refund. The judge ruled in favour of the customer, stating that the chatbot was on the Air Canada webpage, making them liable for what it tells its customers.

A man was trying to transfer his Facebook application to his new phone and searched for a support number. Before calling, he asked Meta’s AI if the phone number was correct. Meta confirmed it was the Facebook support number. So, the man called, and, in the end, he was scammed out of $100 because the phone number did not belong to Meta.

Security is another issue, especially when it comes to generative AI. Pillar Security recently did a study on the security risks of generative AI and found that 20% of generative AI, like chatbots, can be successfully hacked in under 45 seconds. Ninety percent of those hacks reveal personal information.

Between the high risk and the fact that we don’t really know why generative AI hallucinates, when it will hallucinate or what it will hallucinate about, it becomes a game of whack-a-mole to fix some of these guardrail issues around AI. And that’s what makes hackability so high.

A Patagonia customer felt their rights were violated when their call with the contact center was recorded and their information was first fed back to their contact center desktop, Talkdesk, and then given back to Patagonia. The customer says they were not informed of how their information would be used, which violates California privacy laws.

And it’s that pushback that makes AI better used behind the scenes to support human agents. When it comes to direct-to-customer uses of AI, chatbots have a lot of useful use cases. They simply require a lot of testing and ensuring there are guardrails in place. Manufacturer products are more effective at resisting hacks than open-source AI, which we will explore more in part two.