For years, Forethought and other vendors have been selling chatbots to help answer common customer service questions without a human agent. With the growing popularity of generative AI, the company sees even more potential to expand that capability. Today, Forethought introduced the next logical step in that journey called Autoflows.
“We’re announcing a concept called Autoflows, which are natural language simple prompts, where you specify the goal or the policy and you let the AI figure out the rest,” Deon Nicholas, co-founder and CEO at Forethought, told TechCrunch. The idea behind the Autoflows it not simply to answer the questions, but to undertake a series of tasks, sometimes crossing systems to complete it.
He is careful to point out that this is very different from no-code and low-code workflows, which require customer service teams to more deliberately define the actions in a workflow. “You can basically just enter a block of text that specifies a policy, and in true Forethought fashion, we can [recognize the request] and pull in data from your past conversations or your existing policy docs,” he said.
Nicholas says that customers have been using it in beta, and it has been leading to an increase in resolution rate when using Autoflows, compared to classical workflows. Once the AI is armed with the data for any given task, it can start to figure out what questions to ask the customers and start to build a meaningful automated conversation that sounds close to a human customer service agent, he said.
The whole system is powered by SupportGPT, Forethought’s generative AI model. That is in turn powered by OpenAI models, as well as the company’s own proprietary models.
While much of this is automated, humans can still control the process to add, edit or remove tasks and generally fine-tune it, but the goal is for the AI to figure out as much as possible. It’s not fully iterative yet like ChatGPT that lets you refine your request, but it can connect to systems like Zendesk and start a ticket via a Rest API, but for now at least it requires you to add additional tasks manually.
Over time as they refine the process building part of it, Nicholas expects they will build that level of sophistication into the tool. While Autoflows has been in early beta, the company is opening the beta to the public, starting today, to refine it further.