AgentForce is “the next step” for Saleforce customers. The market leader in CRM organized a roadshow in the Netherlands for the second time in a year and at this AgentForce World Tour in the Passenger Terminal Amsterdam. In the room: 2000 AI specialists.  

 

Agent-first is AI-agent first for Salesforce. In that respect, Salesforce is already taking a mortgage on the next phase of AI development. The company started with AI in 2014, according to Jeroen Heijmans, followed in 2016 by the launch of Predictive AI in the form of Einstein and in 2022 with Copilot and LLM applications. We have now arrived at the third wave: autonomous agents or AgentForce. They work alongside people, but they also perform all kinds of tasks in the background. In a demo with retailer Saks (located on Fifth Avenue in New York, among other places), Salesforce showed what a customer can arrange via a voicebot. Sophie, the voicebot of Saks, is the customer-facing AI while in the background other specialized AI agents are busy with order statuses, real-time store stocks, discount schemes and product reservations. Loudly proclaiming that AgentForce is the third wave naturally raises the question of what the fourth wave is.

From 'bot' to 'agent'

The abandonment of the word chatbot (the word still appears in the configuration screens of the various AI agents of Salesforce) is also telling. “We are all frustrated about chatbots,” says Heijmans literally, referring to reports on Nu.nl that chatbots only give a correct answer in 8 percent of cases. The AI ​​agents, that is a different matter, they “know your business, can take action and can be scaled.”

The AgentForce team is both broad and deep: bots such as the field service agent, the customer service agent or the sales agent are available for fifteen different industry clouds – from banking to healthcare, including associated certificates. Each bot can be configured for sets of specific tasks based on roles (which are usually already defined in the Salesforce CRM environment), the data they are allowed to work with, the actions (think of capabilities such as initiating API calls), channels where they are present and various guardrails in the field of trust and security.

Final result, such as Reinier van Leuken showed: voicebot Sophie from Saks can arrange, in interaction with the customer, for a product to be set aside for collection at a branch. Of course, this requires checking that the branch has the product in stock and many other actions are required underwater. But without human intervention – although the product must ultimately be prepared by a flesh-and-blood employee.

Drag & drop is passé

Building such a bot in AgentForce is now also done using natural language in the so-called agent builder, which makes drag & drop and puzzling with no-code and low-code flowcharts seem a thing of the past. The underlying engine that provides the reasoning is called Atlas at Salesforce and can work on the basis of various LLMs. Anyone who wants to get started with the agent builder must be well versed in the processes of the organization; another question is how you keep control of the configuration of AI agents within the organization. They can be adjusted on the fly.

That the AI ​​agents function alongside human employees became clear with their presence in an online collaboration tool such as Slack. In addition to a handful of employees in a certain channel, one or more AI agents are also present there. They listen along, think along and make suggestions or carry out actions, depending on their competencies. Think of a discussion about the sales figures of a certain product with an open end: replenish stock or not? It is the AI ​​agent that ultimately arranges for a new order to be placed with the supplier. In this way, the bot can also monitor customers who are looking at a product in the online store - should I reserve the product for you? Marketers can ask the agent to select a segment from the customer base that is invited to come to the store, but then based on the capacity of your own store location.

 

Christian Krebs of the Danish jewelry brand Pandora (3,7 billion euros in turnover, 750 million visitors per year in the offline and online stores) was busy developing its own LLM, for example to be able to automate product recommendations. Because what works in the physical store should also work online, according to Krebs. But after a while of struggling, Pandora opted for a pilot with AgentForce with three use cases: product recommendations that should help customers purchase the right product; questions about orders ("those are expensive conversations, it would be nice if the AI ​​agent could take over") and the interactive handling of the most frequently asked questions.

AgentForce is still relatively new – introduced in september – and Salesforce has every incentive to convince the market with this. CRM is increasingly becoming a commodity that revolves around data and insights; adding intelligent and automated execution is what several AI players are focusing on. Verint announced at the end of 2023 that it would soon be launching a suite of 50 specialized bots. Almost all CCaaS providers now offer one or more forms of AI functionality as part of their platform. In short, when it comes to new technology, the sky is the limit.

What comes after agent assist?

The coming period will make it clear what the next step of AI in (for example) customer service will really be. Because so far, the proven use cases are limited to rather one-dimensional tasks such as making suggestions for the next best action towards an agent, automated summarizing or offering relevant knowledge snippets in real time. In short, we are waiting for the first European or Dutch Salesforce customer who can show how AI agents take over, accelerate and scale human tasks. Salesforce does not call its AI agents from the AgentForce stable 'autonomous agents' for nothing. (Ziptone/editor)

Featured, Knowledge partners, Technology