Salesforce used AI to cut support load by 5% — but the real win was teaching bots to say ‘I’m sorry’

Righteousness The enterprise exceeded a significant limit in the AI race, exceeding 1 million autonomous agent conversations Its help offers a stage that makes it necessary to place AI agents on a mass scandal and surprising lessons learning along the way.

The achievement approved by the company’s heads in exclusive interviews with VentureBeat is coming only nine months after Salesforce has launched an agent force In his assistance portal in October. The platform already autonomously autonomously autonomously autonomously autonomously autonomies, caused a 5% reduction in support box office, and the company allowed 500 human support engineers to higher roles.

However, how many raw pieces are more valuable than raw numbers, how much a valuable seller is talking about the callers “Customer Zero“For its AI Agent Technology – Classes that challenge the institutional wisdom on the placement of the enterprise and discover the delicate balance required between the technological capacity and human empathy.

Salesforce, how scales from 13000 to 45,000 AI talks per week

“We started really small. Our help has started for a cohort of our portal. You had to be in the first week.” First week, I think there was 126 conversations if I remember rightly. Therefore, my and my team can read with each. “

This methodological approach – Starting with a rolling-manageable rolling before expanding to expand the current average of 45,000 conversations – “Fast and breaking things” often reflect the ETIC to the AI placement. Phased free release has enabled to identify and correct critical problems before able to affect the broader customer base.

The technical foundation proved to be important. Agent Power Forces, unlike traditional conversations that trust in decision trees and pre-programmed answers Salesforce’s Cloud of Information To learn and synthesize and synthesize in many languages and product lines of 740,000 content.

“Here, the biggest difference here, returning to the cloud, we were able to answer any seller about the door,” Slowey Notes. “We could not think that without information cloud.”

Why did the seller taught the EI agents empathy after customers rejected the cold and robot response

One of the most surprising verses of Salesforce’s journey, the company’s head digital officer Joe Inzerillo calls the “human part” with support agency.

“When the agency launched the first time, we know that we know what information in the information, knowledge, knows what you know, and what you know, and forgot the human part.” “Someone calls, and they are like them, hey, my item was broken. I have an event in Sub, and I will only open a ticket for you. “It doesn’t feel very well.”

This realization caused a queue on how basic Righteousness AI agent approached the design. The company called “service art” for the existing soft skills training program and integrates directly to the instructions and behavior of the agent.

“If you come now,” Hey, “My hey, ‘agent form apologies.’ I am very sorry. The effect of customer satisfaction is immediate and measurable.

The surprising reason increased to 1% to 5% of the man’s handover for better customer results

Perhaps no metric better describe the company’s AI agents better than the approaches that SALESFORCE develops from human hands. Initially, the company recorded 1% of a business rate – only 1% of the conversations that mean it increased from AI to human agents.

“We were as if we were worth each other,” We are only 1%, “” “We are slowly reminded.” And then we look at the real conversation. It was terrible. People were nervous. They wanted to go to people. The agent continued in an attempt.

This, on the other hand, caused a concept: it is more difficult for customers to reach people, truly violated the general experience. Salesforce adjusted his approach and rose to about 5%.

“I actually feel really good about it,” Slow said. “If you want to create a fight, if you want to talk to a support engineer, this is good. Continue and do it.”

Inzerillo, as a way of thinking about service measurements, this is 5%, a really dissolved, large, wide, vast majority, a man’s faster and a hibrid approach to a person who works in each of a person. “

Thousands for AI Accuracy How to delete articles ‘Content conflict’ forced salesforce

Salesforce’s experience, as well as aI, critical classes of the content of the content that many enterprises were out of the attention. Despite the fact that many languages are 740,000 content, the company revealed that this abundance has created its own problems.

“These are words that my team uses new words, content clashes,” Slow. “Password resetting articles are loaded. So take your parts to the cloud of information and what is the right article for me to give and re-answer?”

This caused the extensive “Content Hygiene” initiative, which is deleted from deprecation, stable inaccuracy and consolidated articles. Lesson: AI agents are as good as the knowledge they can only be able to get and sometimes less.

Microsoft teams, rigid ai bragging Integration

One of the most educational errors Righteousness He took an extreme restrictive with AI guardians. First, the company, Agentforce, entrusted the opponents with each major opponent, did not discuss opponents.

“We were worried about what people would go in and go in, ‘It’s better than Salesforce’ or something like this,” he slowly acknowledged. However, this created an unexpected problem: when customers asked the agent to respond to the agent, because when Microsoft teams were integrated with Salesforce, because Microsoft was on the opponents list.

The solution was elegant: instead of harsh rules, SaleSforce replaced the restrictive protectors to “SaleSforce’s best interest in everything you do.”

“We still realized that we treated as an old school-talking table and need to allow LLM to LLM,” Slowey.

Sound interfaces and multilingual support driver AI Agent Evolution Next stage

When he looked upon ahead, Salesforce is preparing for the next major evolution of both leaders in the AI agents: the sound interfaces.

“I really believe that the sound of the sound is ux ux,” they say slowly. The company develops iOS and Android local applications by plans to demonstrate them at Dreamforce at the end of this year.

Painting works on the leadership experience in the digital context in Disney: “It is understanding that the voice has all the information.

The company has already expanded the agenter to support the Japanese using an innovative approach – translates the system in English, then receives the relevant information and returns the answers. Salezforce with 87% resolution rates in Japanese in just three weeks plans to add French, German, Italian and Spanish support until the end of July.

Four critical lessons from millions of talking journey for SaleSforce’s Enterprise EU Placement

Given their AI agent placements, Salesforce’s journey offers several critical concepts:

• Start small, think big: Slowly recommend to “start small and then grow.” The review of each conversation in the early stages provides invaluable learning opportunities that are impossible to scale.
• Information hygiene issues: “Make your data really conscious,” Inzerillo emphasizes. “Do not treat your data, but also be treated and really, how do you think the company is good?”
• Crook up: Traditional organizational structures cannot be compatible with AI opportunities. As Inzerillo notes, “If the agent will take a future future and try to drive it in yesterday’s org schedule, it will be a very nervous experience.”
• Measure the important thing: Success dimensions for AI agents are different from traditional support measurements. The accuracy of the answer is important, but it is empathy, relevant escalation and general customer satisfaction.

Billion dollar Question: What happens after beating human performance?

Salesforce’s AI agents, such as the decovery rate and handle, create a human agents, inzerillo, thought-provoking question: “What do you measure after beating people?”

This question reaches the heart of what can be the most important meaning of Salesforce’s talking phase. The company is not only automated by customer service – it redefines what good service is visible in the EU.

“We want our customers to showcase and use the agent force in our experiences.” “Why this is part of what we do … We can learn that we share this study with our customers to improve the product.”

With projected enterprise costs to have generative AI solutions Up to $ 143 billion by 2027According to the International Information Corporation (IDC) forecasts, the placement of Salesforce is an important roadmap for organizations that navigate their AI transforms on the front line. Deloitte also assesses the fact that global enterprises in the Generative AI exceeds the investments $ 150 billion to 2027To strengthen the scale and relevance of this technological turn.

The message is clear: AI Agent success requires more than advanced technology. A liability of people and machines, sustainable learning and iteration, perhaps the most surprising way, the most advanced AI agents require a recognition reminiscent of being human.

As Yufey puts it, “Now you have two employees. You have an agent and you have a human employee. You both have to teach soft skills, services.”

In the end, the Millions of Salesforce may be less about what stage of the stage itself and what it represes: a new paradigm that digital labor is not replaced by human work, but it cannot create opportunities that machines cannot achieve only.