Salesforce's AI Demos Don't Match Reality, Raising Questions for Customer Support Teams
Salesforce has showcased its AI assistant Agentforce as a solution for customer service automation, but major clients featured in promotional materials have not actually deployed the core features demonstrated. The discrepancy between marketing claims and actual implementation is now drawing scrutiny as the company's stock falls and customers question what they're buying.
Bloomberg's investigation found that Williams-Sonoma, Finnair, and the University of Chicago Medical Center-all highlighted in Salesforce's promotional videos-have not implemented the AI capabilities shown in those demos. For customer support professionals evaluating these tools, the gap matters: you may be considering a product based on capabilities that don't yet exist.
What the Demos Promised vs. What's Actually Running
At a Salesforce user conference in December 2024, the company demonstrated Williams-Sonoma's AI-powered voice customer service system handling calls and completing purchases automatically. Six months later, the retailer's phone system still hasn't integrated Agentforce. The company's chatbot remains unable to process orders without human intervention and currently handles only recipe suggestions and order status checks.
Finnair appeared in a Salesforce video showing Agentforce automatically rebooking flights for passengers. The airline's website labels this feature as "only planned for future development." A spokesperson cited missing capabilities in related software as the barrier.
The University of Chicago Medical Center was featured in a promotional video showing patients refilling prescriptions, scheduling appointments, and receiving parking suggestions through Agentforce. None of these features have gone live, according to people familiar with the implementation. The health system pays millions annually in Salesforce fees.
Salesforce's Defense
CEO Marc Benioff defended the company's approach as standard industry practice. He said Salesforce labels forward-looking content in its marketing and doesn't want clients to discover features don't exist after purchase.
Madhav Thattai, the executive overseeing Agentforce, attributed delays to client-side compliance and processes rather than technology limitations. He said clients typically start with simpler tasks before moving to more complex autonomous scenarios.
Analyst Gil Luria at DA Davidson noted the problem extends beyond Salesforce. "Under pressure to deliver AI, many companies are promising things far beyond their actual capabilities. Salesforce is just the loudest," he said.
Where Agentforce Actually Works
The tool has shown real value in specific, limited scenarios. SharkNinja reduced service calls by 20% after implementing automatic fault diagnosis. Adecco Group processes tens of thousands of initial resume screenings. The city of Kyle, Texas uses it to route resident complaints about potholes and graffiti.
These successes cluster around information processing and sorting-tasks with lower risk. Analyst Rebecca Wettemann said Agentforce struggles with complex operations because companies fear the AI will act autonomously on high-value decisions. "People are afraid the AI will issue refunds on its own at midnight or rack up massive bills," she said.
The Business Pressure Behind the Marketing
Salesforce's annual revenue growth has slowed from 25% in 2022 to 10% recently. The company is betting on Agentforce to reverse that trend. It has signed 29,000 Agentforce contracts generating $800 million in annual recurring revenue.
The company's strategy is shifting. Benioff initially emphasized how easy Agentforce deployment would be. Recently, Salesforce has acknowledged that "building agents is not easy" and is advising clients to focus on narrow, specific use cases. The company is now offering free or discounted technical support and acquired Informatica for $8 billion to strengthen data integration-a prerequisite for AI deployment.
Salesforce is also tying sales compensation to actual customer usage of Agentforce, not just contract signings. This suggests internal recognition that signed deals don't equal implemented solutions.
What This Means for Your Evaluation
If you're evaluating AI for Customer Support, ask vendors directly about live deployments, not pilot programs. Request references from customers using the specific features you need, not general case studies.
Understanding the actual capabilities of Generative AI and LLM technology helps you assess realistic timelines. The gap between what's marketed and what's deployed is not unique to Salesforce, but it's a real cost for teams making purchasing decisions.
Your membership also unlocks:
