Salesforce crowdsources its AI roadmap from customers

Artificial intelligence is moving so fast that enterprises are struggling to keep pace, often scrambling to ship new products just to stay relevant. Salesforce believes it has cracked the code by turning to an unexpected source for direction: its own customers. The customer management software giant is now crowdsourcing its AI roadmap in real time, a strategy that lets it pivot quickly even when the future of AI remains uncertain.

Of course, Salesforce isn’t the only company that listens closely to its users. But what sets this approach apart is the scale. With 18,000 customers in the loop, the company is holding weekly check-ins with some clients, not just annual or quarterly surveys. This isn’t about gathering feedback in batches; it’s a continuous, high-frequency dialogue that shapes product development on the fly.

“The 18,000 customers are a wellspring of information and a wealth of information that is really needed to get to customer success,” said Jayesh Govindarajan, executive vice president at Salesforce AI, in a recent interview with TechCrunch. He explained that the stack Salesforce built has resonated with users, and as the company gathers more context over time, its systems improve. “As it gets better, and LLMs get better, agent systems do more and more fully autonomous behaviors. That’s a long running innovation track and we’re going to invest in that.”

Salesforce was an early mover in AI agent management, launching its platform in late 2024 before agentic AI became a headline-grabbing trend the following year. Since then, the company has doubled down, rapidly releasing new products for voice AI and Slack. The secret behind this pace? Customers. By letting users lead the way, Salesforce builds a product roadmap that can react to where AI is headed, not where it was six months ago.

When large language models first emerged, enterprises were eager to adopt them but lacked the last-mile technology to make them useful. That gap sparked the creation of Agentforce, Salesforce’s agent management platform, according to Govindarajan. From there, the company adopted a bottom-up strategy driven by themes like agent context, observability, and deterministic controls, rather than fixed product timelines. This approach relies on rotating groups of customers to validate ideas, assuming that their needs will mirror those of the broader market.

“The innovation that we’ve brought, they are direct result of us working with a vast number of these customers and then classifying the problems they see in the real world,” Govindarajan said. “Then [we break] that down and say, which of this can be solved at the LLM layer, which cannot? And for things that we cannot solve at the LLM layer, we need to build that sort of agentic operating system components around the LLMs to be able to go do that.”

This tight collaboration with customer engineering teams lets Salesforce fix issues before the technology evolves past them. “We can’t wait three months or six months to get feedback, and then go figure out another six months of work,” said Muralidhar Krishnaprasad, president and CTO of Salesforce engineering. “We are literally reacting to it, week by week, month by month. That’s been a big change. Now we push code, pretty fast, and we have various sorts of gates to try out new features, get earlier feedback before we release it broadly as well.”

One company in that feedback loop is Engine, a travel management platform. Its founder and CEO, Elia Wallen, said his operations team meets with Salesforce weekly. Through this partnership, Engine gets early access to AI tools, helping it stay competitive and extract more value from the technology. But the relationship is a two-way street. Wallen recalled instructing an AI voice agent to book a hotel in Chicago, only to find the interaction felt unnatural. After sharing that feedback, Salesforce adjusted the agent, and A/B tests soon showed better results.

“If somebody is willing to actually help curate and build products that we need, they can help us better and really understand our problem and how they can solve it,” Wallen said. “For us, it’s fantastic to actually be invited into a thing like that, because we can influence the product.”

Salesforce also takes successful workflows developed by customers and rolls them out to its broader base. PenFed, a federal credit union, used existing tools and agents in Agentforce to build its own IT service management (ITSM) workflow. Shree Reddy, PenFed’s chief innovation officer and executive vice president, said the company has slimmed down its tech stack by focusing on strategic platforms like Salesforce. “We invest our time, energy into the platforms that are more strategic, and we obviously spend a lot more time on this relationship,” Reddy said. “That investment has yielded good results in terms of strengthening that partnership that’s influencing each other, and what we see is the best value add mutually to both organizations.”

Still, this customer-driven approach has its risks. It relies on the classic assumption that the customer is always right, even when many enterprises are still figuring out what role AI should play in their business. Some may not yet have found value in the technology, making them unreliable guides for long-term product development. And early enthusiasm for beta features doesn’t always translate into lasting usage or future contracts.

To hedge its bets, Salesforce also applies this bottom-up strategy internally. Govindarajan noted that Salesforce employees are the biggest users of its own AI tools. When ChatGPT launched, the company shifted teams and resources to form a new AI team, a tactic it has used successfully during past innovation waves. “As the technology changes, we never know what’s going to come out a month from now,” Krishnaprasad said. “We will adapt to it. And that’s what we did all of last year. If you think about it, agents weren’t even in terminology when you look back a year and a half ago. And then we had to go react to it. We had to go react to all the advances, and we had to react to our customers.”