Chatbot company Ada Support Inc. is aiming to automate a greater portion of its clients’ customer service inquiries with generative artificial intelligence, which it says will save costs and boost efficiency.
The Toronto-based firm has released what it calls fully automated software that includes chatbots to correspond with customers and automatically resolve some issues, along with generative voice technology to talk with people over the phone.
“The future of customer service is agent-less,” Ada chief executive and co-founder Mike Murchison said in an interview. “There will be no customer service agents as we know them today.” Globally, hundreds of billions of dollars are invested every year in human-centred customer service, he said. “I believe that in the next half-decade, the vast, vast majority of it will shift to AI platforms.”
Founded in 2016, Ada is a provider of customer service chatbots for companies such as Meta Platforms Inc. META-Q and Shopify Inc. SHOP-T It has previously used generative AI to create potential responses to customer service inquiries, but a human agent would have to select which reply to send. Ada is now aiming to cut the human out of that process.
The company’s software is powered by GPT-4, the large language model developed by San Francisco-based OpenAI. Interest in generative AI, which refers to software that creates text and images, has skyrocketed since OpenAI released ChatGPT in November. But ChatGPT and other bots can make up facts, leading many companies to remain cautious about deploying generative technology.
The risks are high even with seemingly mundane customer service interactions. A chatbot could theoretically provide incorrect information or invent new corporate policies, frustrating and misleading clients. A bot could also be prompted to make malicious and inappropriate comments – exchanges that could cause public-relations nightmares if shared across social media.
Mr. Murchison is confident that Ada’s software will not go rogue. The bots pull from a company’s internal documents and existing customer service transcripts, as well as information about the industry in which a company operates. A chatbot for a finance company would know who heads the Bank of Canada, for example, but it does not need to be able to expound on politics or other potentially fraught topics.
Ada has also rigorously tested the technology, he said, even by attempting to push bots to stray into risky conversational territory. “Humans, especially with new technology, are very quick to try to find every backdoor,” Mr. Murchison said. “That’s why we’ve taken the safety component so seriously.”
While Ada is relying on GPT-4, Mr. Murchison is an early investor in Cohere, a Toronto-based AI company that competes with OpenAI. “We’re seeing the lowest-latency, highest-quality generations possible with GPT-4 today,” he said. “There’s so many different applications for large language models, and I’m confident Cohere is exceptional in a number of them.”
Some customers have already been beta testing Ada’s new platform, but the company was unable to connect The Globe and Mail with one for an interview.
Companies have made claims about automation for years, said Sean O’Brady, an assistant professor at McMaster University in Hamilton, Ont., who studies call-centre labour. Often, the reality has not lived up to the promises.
A lot of routine customer service functions have already been automated, but agents are still necessary to resolve more complex or idiosyncratic issues. Today, AI is more commonly used in call centres to provide assistance and guidance to agents. “I don’t know how much more space there is to automate the rest,” Prof. O’Brady said. When it comes to employment, call-centre workers are still more concerned about job losses due to outsourcing, not AI.
One risk in relying on automation is that customers could face hurdles when they need to reach a fellow human, and by the time they do, they’re deeply frustrated. “Sometimes the technologies that are there to make things simple are also creating more complexity for other calls,” Prof. O’Brady said.
According to Mr. Murchison, Ada’s software is capable of answering a much broader range of questions than previous chatbots and can independently complete some tasks, such as updating subscriptions, creating return labels for shipping and upgrading a seat on a flight. The company is working on automating more complex tasks.
Even so, Mr. Murchison still sees a role for human agents and believes the job will evolve to become one where customer service employees are looking for ways to improve AI. “Customer service agents go from answering customers reactively to pro-actively coaching their AI,” he said.