Gradient Labs CEO Dimitri Masin is challenging how companies charge for AI agents in customer support. In an interview with The Register, Masin criticized major platforms like Salesforce for charging per conversation, regardless of whether the AI agent solves the issue. If you have a conversation with AI, no matter what it leads to, you pay $2, Masin said, citing Salesforce’s model.
He called it a flawed system that fails to motivate improvement. According to Salesforce’s research, its large language model (LLM)-based agents resolve single-turn queries only 58% of the time and multi-turn requests just 35% of the time. Masin said this approach not only hurts customers but also slows down the development of more capable AI. There’s no incentive for Salesforce to make its agent better, he argued. He instead proposes a model where payment only occurs when an AI agent resolves the issue.
ROI-Driven Billing for AI Agents
Masin believes resolution-based billing gives companies a clearer view of return on investment. A recent F5 survey supports this concern, 62% of enterprises said AI compute cost is their top worry when deploying AI. Gradient Labs’ pricing model aligns payment with outcomes. If we don’t resolve the issue and a human still has to step in, then you don’t need to pay us, Masin said.
This model allows companies to assess whether AI support is worth the cost by comparing it directly to human agent expenses. Masin pointed out that if a human agent costs $10 per issue and an AI agent can resolve it for $3, the company sees an immediate value. But the system must reflect the actual work done. He emphasized that billing needs to scale with query complexity to ensure fairness and real savings.
Tiered Pricing Reflects Query Complexity
Gradient Labs uses a tiered pricing model to account for different types of support queries. Masin explained that some queries are simple and take a minute to solve, while others are much more complex, requiring up to 40 minutes even for trained support staff. To charge a flat $1 per resolution doesn’t make sense, he said.
Gradient Labs sets pricing by resolution success rate; 50% resolution carries one rate, while 70% earns a higher one. The more complex the resolved queries, the more value the AI agent delivers. Masin aims to charge about 30% of what a human would cost, letting companies save 70%. But this depends on which part of the support load the AI handles. If it only tackles easy queries, the value diminishes. If they automate the easiest 50% and only save 20% time, the ROI isn’t clear, he noted.
Real Usage Proves the Model
Gradient Labs claims its AI agents resolve 40–60% of queries without any customization. That alone leads to around 20% savings for clients. With deeper data integration, Masin said, performance rises to 80–90%, typically achieved over a three- to five-month period. In one example, a deployment with Sling Money hit a 78% resolution rate after optimization.
Among five recent fintech clients, AI agents scored higher on customer satisfaction than internal human support teams. Masin said one key factor is speed in escalation. When Gradient’s AI hits a wall, it quickly routes the conversation to a human rather than wasting the customer’s time. Half of our clients disclose the use of AI. The other half don’t, and customers don’t even notice, he said.
Not a One-Size-Fits-All Solution
Masin admitted that AI agents work best in specific domains like customer support, where clear resolution is possible. In contrast, he finds them ineffective in areas like sales and marketing. Most of them are useless there, he said. They need to focus on solving high-value, narrow problems. He believes success lies in building agents that resolve real issues, not just generate dialogue. “The value comes when AI does something, not when it just talks,” Masin added. Gradient Labs plans to continue refining its pricing model as more companies seek AI solutions that deliver measurable ROI, especially amid rising expectations and costs tied to query complexity.
About The Author
