Getting started with AI isn’t difficult, but effectively getting projects into production is. Salesforce sees where the pain points are and makes it clear how you can sustainably integrate AI. Nothing is more important than starting.
There is a gap between how we use generative AI today and how the technology is integrated into enterprises. That’s what Maarten Laukens, Deputy Country Leader at Salesforce, noted on stage at the Salesforce Innovation Day in Brussels. “95 percent of all AI pilot projects fail,” he says.
Pitfalls and enthusiasm
The reasons are numerous, but Laukens points out some major pitfalls: “Data can be of poor quality or insufficient. Outdated data is also a risk. Too often, AI is also applied as a technology on top of what already exists, instead of being integrated into the processes. One window to work in and one to ask questions: that’s not valuable.”
At the same time, companies are making a significant mistake in how they approach AI. “We see companies starting AI projects because they want AI,” says Laukens. “While AI should be used to specifically enable more efficiency or a better customer experience.”
Companies want to invest in AI, including in Belgium. Country Leader Lien Ceulemans knows this: “Every company knows that a revolution is underway. No one thinks it’s an option not to participate anymore. Of course, there is a big difference in adoption, although we do see companies getting past the experimental phase.”
The good example
Salesforce itself sets a good example. A little less than a year and a half ago, the company put its first agent into production, focused on customer support. “We looked at our data for that,” Laukens clarifies, “but also the business objectives we wanted to achieve, and the underlying processes. We added knowledge and actions to that.”
The result is impressive: at the time of the Innovation Day, the Customer Support agent has already had 1.8 million conversations with customers in production. Salesforce estimates that it has saved around one hundred million dollars on an annual basis. Meanwhile, AI agents support salespeople and have already played a role in the sales pipeline for $60 million in revenue. “We see fifteen percent more conversion through AI-driven personalized messages,” Laukens adds.
We see fifteen percent more conversion through AI-driven personalized messages
Maarten Laukens, Deputy Country Leader Salesforce
With the figures, the Deputy Country Leader makes it clear that Salesforce doesn’t just sell AI solutions, but is also its own first customer and already has a lot of AI experience. In a context where a technology has only exploded for about three years, a year and a half of experience in production counts as an eternity.
Agents in the workflow
During that period, Salesforce has learned some important lessons. Agents must be in the workflow and not exist next to it as a conversation interface in another tool. Furthermore, the capacity for reasoning is crucial. “LLMs will be everywhere,” Laukens predicts. “The art will be to have AI Agents and our human workforce work together optimally and customer-oriented.”
Finally, it is not enough to set agents in motion with a written or spoken prompt. Laukens: “They must kick in when data changes, for example when a customer takes an action or a problem is proactively detected.”
Everyone an agent builder
To arrive at a successful AI product, building and managing AI agents should not be exclusively in the hands of the IT department. According to Salesforce, everyone should get the tools to build an agent-driven company. That requires simplicity.
Laukens takes the opportunity to highlight the Agentforce Builder in that context. That is an environment where AI agents help users build other AI agents. “You describe what you want, and the agent is built.”
That low barrier is important, Ceulemans also sees. “The relatively low cost and simplicity of Agentforce work,” she says. “Companies are getting agents into production that way.”
The relatively low cost and simplicity of Agentforce work.
Lien Ceulemans, Country Leader Salesforce
“I think we can easily demonstrate that anyone can build an agent today,” Ceulemans continues. “Provided the right corporate culture and the right support, of course. However, AI agents are not futuristic things that require enormous time and resources.”
Barco and Callebaut
At the Innovation Day, we see some clear examples. For example, chocolate specialist Barry Callebaut is working on an AI agent that will help customers come up with recipes. That project is possible thanks to the foundation that is already in place with Salesforce and serves as a starting point for further AI innovation. Callebaut started the project after inspiration at Dreamforce this year.
Not all data needs to be ready for AI: look at what you need and connect that. Starting small and building trust is key. Philippe Verlinde, CDIO of Barco, also believes this. “You have to look at data within an end-to-end context,” he says. “Good data orchestration is key. The entire journey towards AI is a long one, but the most important thing is to start now.”
read also
Barry Callebaut: expert in chocolate and the entire Salesforce ecosystem
The most important step: start. The search for a few use cases with high value but low complexity is essential, as are the right expectations. The perfect AI agent will not be there from attempt one. Versions two and three will be needed, and there will probably still be a lot of work to be done, but that’s okay.
New Tools
To support innovation, Salesforce is highlighting some agent-related innovations during the Innovation Day. There is Agent Script, which adds a layer of determinism to the reasoning capabilities of AI agents. Sometimes if-then rules are important, and Salesforce allows agents to reason in a hybrid way.
Then there’s Agentforce Voice, which allows you to talk to an agent in natural language. That functionality is available, we hear, but currently only in English. The fact that Dutch is usually available somewhat later is not a disaster according to Tom Asselman, Head of Enterprise Sales in Belgium.
“In Belgium, we have the luxury of bilingualism,” he says. “French is usually quickly available. Organizations can then start with a use case in French and then roll it out in Dutch relatively easily later.”
Finally, Salesforce is spotlighting Agentforce Vibes on stage. “That is the fastest and most reliable way to build applications,” says Peter Tobac, Manager Solution Engineering. “You simply start by describing what is needed. Because Agentforce Vibes is fully built into the Salesforce platform, it understands all business processes, for example in Sales or Service, and has access to all data. That context ensures that new applications integrate perfectly. It’s as if your best developer knows your business processes from the outside.”
Platform integration
That brings us back to Laukens at the beginning, when he emphasizes how important it is that agents are in the flow of work. Salesforce wants to emphasize that strength of its own platform in Brussels. Because the Salesforce platform with all its integrated components can support the entire business operation, is based on a single source of truth, and can also connect to other data sources via Data 360 (Data Cloud), it is in the right place to enable AI.
“Welcome to the agent-based company,” Ceulemans concludes. “AI Agents must run on a platform, where they get context from connected data and are orchestrated with the right tools. Everything, including MuleSoft and Tableau, comes together within Agentforce 360.”