Salesforce is taking an unusual approach to addressing the breakneck pace of artificial intelligence: it is letting its customers decide where to invest next. The customer relationship management giant has shifted from quarterly product reviews to weekly feedback sessions with select clients, using their real-world challenges to shape its AI strategy in near real time.
Why Salesforce is turning to customers for AI direction
The company’s AI leadership says the traditional product development cycle is too slow for the current environment. Large language models and agentic AI are evolving faster than most enterprises can adapt, creating a gap between what technology offers and what businesses can actually deploy. Salesforce aims to close that gap by building tools that solve specific, recurring problems its customers encounter — not hypothetical future needs.
Also read: Medicare’s quiet bet on AI: A new payment model that most of tech hasn’t noticed
Jayesh Govindarajan, executive vice president at Salesforce AI, told TechCrunch that the company’s 18,000 customers provide a wellspring of information that is essential for building successful products. Rather than guessing what enterprises want, Salesforce is classifying the problems customers face in the real world and deciding which can be solved at the LLM layer and which require additional infrastructure.
How the feedback loop works
Engine, a travel management platform, is one of the companies in Salesforce’s inner circle. Its operations team meets with Salesforce weekly. CEO Elia Wallen said the partnership gives Engine early access to AI tools before public release, helping the startup stay competitive. In return, Salesforce gets immediate, granular feedback on what works and what does not.
Also read: Altman testifies Musk once proposed handing OpenAI to his children during safety dispute
Wallen described an instance where he tested a voice AI agent to book a hotel in Chicago. He felt the interaction was unnatural and shared that observation with Salesforce. Within weeks, the company had adjusted the agent and its A/B tests began showing improved results. This kind of rapid iteration, Wallen said, is rare in enterprise software.
PenFed, a federal credit union with over $30 billion in assets, has also benefited from the close relationship. Chief innovation officer Shree Reddy said PenFed built an IT service management workflow using existing Agentforce tools. Salesforce saw the success, adopted the workflow, and rolled it out to the broader platform for other enterprises.
The risks of letting customers lead
The strategy is not without drawbacks. Many enterprises are still experimenting with AI and have not yet found consistent value from the technology. Relying on them for long-term product direction could lead Salesforce down paths that do not scale. Additionally, early enthusiasm for beta features does not always translate into sustained usage or long-term contracts.
Salesforce acknowledges this tension but argues that the alternative — waiting for clarity before building — is riskier. Muralidhar Krishnaprasad, president and CTO of Salesforce engineering, said the company cannot afford to wait three or six months for feedback. It is pushing code weekly and using gates to test features with customers before broad release.
Internal dogfooding as a complement
Salesforce also uses its own AI tools internally at scale. Govindarajan said employees are the biggest users of the company’s AI products, providing another layer of real-world testing. When ChatGPT launched, Salesforce quickly reassigned teams and resources to form a new AI unit — a strategy it has used successfully during previous technology shifts.
This dual approach — listening to customers and using its own products — gives Salesforce a feedback loop that is both broad and deep. The company launched Agentforce, its agent management platform, in late 2024, before agentic AI became a dominant industry theme. It has since released updates for voice AI and Slack at a rapid pace, crediting customer input for the velocity.
Conclusion
Salesforce’s crowdsourced AI roadmap represents a bet that the fastest path to product-market fit is through continuous, direct customer collaboration. While the approach carries risks — including the possibility that customers may not always know what they need — the company believes the alternative of slower, more traditional development cycles is untenable in the current AI field. For now, Salesforce is betting that its customers’ real-world problems will lead to products that serve the broader market.
FAQs
Q1: How often does Salesforce meet with customers for AI feedback?
Salesforce meets with some customers as often as once a week, a shift from traditional quarterly or annual product reviews.
Q2: What is Agentforce?
Agentforce is Salesforce’s AI agent management platform, launched in late 2024, that allows enterprises to build and deploy AI agents. It was developed based on customer feedback.
Q3: Can other companies use workflows built by Salesforce customers?
Yes. Salesforce rolls out successful workflows and solutions built by individual customers to its broader platform, as happened with PenFed’s IT service management workflow.
Be the first to comment