A little more than six months after San Francisco-based OpenAI released the latest version of its chatbot, ChatGPT, to the public, the number one priority for McKinsey and for customers is to automate mundane jobs and make operations more efficient, Ellencsweig said.
One client is aiming to automate 30% of its calls in the next year, he said. McKinsey itself is working on techniques to automate certain kinds of client engagement events to a self-service model using generative A.I. in a way that doesn’t feel like clients are using a chatbot.
These goals also require a human focus: managing expectations and organizational structures at client companies that are adopting the technology. A.I. is helping with that too. Because so many individuals have tried the chatbot or another buzzy generative A.I. tool, the moment is right to introduce A.I. at work, according to Ellencweig. The faster it’s implemented, the easier it may be for employees to adopt.
That uptake is key, said Alexander Sukharevsky, also a senior partner at McKinsey and global leader of the A.I. group.
“The question has been how to scale from ‘sandboxes’ to enterprise-grade A.I. applications,” said Sukharevsky, who is based in London.
One answer is that companies have to plan to use A.I. sustainably, both in terms of the environment and in terms of their IT spending, said Sukharevsky, since the amount of computing power needed for some algorithms is enormous and expensive.
Another is that they have to be meticulous about data privacy and safety—a major reason for the work with Cohere, which has prioritized protection and customization for enterprise customers. The result is that A.I. models can be trained specifically for individual clients or industries.
Finally, the partners said, firms have to take the time to research and settle on the right A.I. tech stack, since they are likely to form the basis of their innovation strategy for a long time to come.
That means that companies have to move both fast and slowly with the new tech.
“You really have to do both,” said Ellencsweigh. “You have to experiment fast to see the power of it, but in parallel you have to find the right vendors and partners for use cases that are more complex.”