Salesforce to release autonomous AI customer service agents

Salesforce previewed Wednesday its next generation of service bot it calls "autonomous AI agents." They  are currently in pilot and are planned to released later this year.

The autonomous AI bots, called Einstein Service Agent, promise to use generative AI to detect customer intent, invoke Flows and trigger actions within a user's Salesforce instance, said Ryan Nichols, Service Cloud chief product officer. Natural language processing, predictive analytics and generative AI will work together to steer Einstein Service Agents to choose scenarios to find solutions for which a customer is contacting a company.

Current bots must be programmed with rules-based logic to detect the need for and accommodate solutions for individual customer service issues, such as returns, refunds, password changes and subscription renewals.

Autonomous bots aided by generative AI in theory can more precisely detect problems by better understanding a customer issue (i.e., when they ask to "send back" an item, the bot knows they mean "return") and also summarizing the process to accomplish that. Salesforce said Einstein Service Agent will work on WhatsApp, Apple Messages for Business, Facebook Messenger and SMS text. It will understand not just text but also images, video and audio that customers upload when they can't find the words to describe their issues at hand.

These forthcoming bots sound good but they had better work, said Constellation Research analyst Liz Miller. If they don't, all they will do is waste more customer time than today's tech, and potentially worsen customer relationships.

Miller said she believes the key to successful autonomous AI agent implementation will be for organizations to keep humans close by when the bot "fries out," or gets confused by a customer request of a Salesforce user's bot.

Customers, whether they're in B2C or B2B settings, are weary of the current generation of rules-based bots, because they waste time en route to getting what they really want -- to talk to a human.

It's great that the bot can handle complex questions. These are things autonomous agents absolutely have to handle, and they have to resolve.

"It's great that the bot can handle complex questions. These are things autonomous agents absolutely have to handle, and they have to resolve," Miller said. "Advancements like this, where that next [level of complex question] doesn't need to be programmed for a human and if the human didn't program it, the whole bot doesn't fry itself.

"But I think that, with every massive leap in technology, such as we're seeing with GenAI, there also should come a warning label that says, 'People are complex. Language is complex. Thought processes are complex and really hard to map,'" Miller said. "It's one of the big reasons why, in the history of technology, people have to want to talk to people."

Nichols said the forthcoming Einstein Service Agent will take less time for users to set up. Some customers have thousands of "utterances," or potential things a customer can ask for in a service call, mapped to a conversational decision tree.

"That can be quite cumbersome and expensive to get up and running," he said. "And then if you ever need to change it, boy, it's you're dealing with spaghetti [flowchart] conversation design."

Salesforce customers in the Einstein Service agent autonomous AI pilot program have significantly reduced the number of scenarios they need to program, Nichols said. Because generative AI tools can understand many permutations of words behind a service request, each of which needed to be manually programmed before, the technology could potentially reduce the need for developer and data scientist overhead for a Salesforce user.

Pricing for the autonomous Einstein Service Agent will be announced upon release.

Don Fluckinger is a senior news writer for TechTarget Editorial. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.