At a time when AI technology is advancing rapidly and companies are pushing forward with intelligent transformation across the board, a once-hidden expense is becoming a financial pressure that major companies cannot ignore. According to Business Insider, after some employees racked up AI token costs of several thousand dollars in a single month last quarter, CRM giant Salesforce decided to take action and began implementing role-based budgets for employee AI usage across the company.

Compute Costs Remain High: From Unrestrained Exploration to Granular Control

In an internal memo sent to employees, Salesforce Chief Operating Officer Miguel Milano announced that starting this month, the company will set monthly token budgets based on the actual business needs of different roles. The core goal is to concentrate valuable AI resources on “work that creates the most value for customers and the business.”

This move deeply reflects the dilemma companies generally face as they embrace the wave of large models: on the one hand, companies strongly encourage employees to use cutting-edge AI to boost productivity; on the other hand, high API call and compute costs force management to control costs precisely.

However, a Salesforce spokesperson also emphasized that the company is not imposing a one-size-fits-all absolute individual cap across the entire group. Instead, it is leaving specific usage permissions for each team to be managed independently by its leaders. The core principle remains ensuring that every dollar invested in AI delivers real returns.

Team Pilot Delivers Strong Results: Spending Drops Notably Without Hurting Efficiency

Before the full rollout, the budget system had already undergone a one-month stress test within one Salesforce team.

The memo shows that after early trials and adjustments to usage habits, by the second month, the team’s actual token spending was already well below the budget ceiling, while the team’s overall output and efficiency were not negatively affected at all. This provides a workable model for companies to optimize costs while opening up AI access.

Detailed Role Budgets Unveiled: Technical and Non-Technical Roles Differ Sharply

According to the officially announced plan, monthly budget amounts vary significantly by department and role:

Sales department: In the sales organization led by President and CRO Alexa Vignone, technical roles have a monthly token budget of $1,000 (about RMB 6,680), while non-technical roles such as sales staff get $100 (about RMB 668).

Customer Success and Professional Services: The Customer Success organization (led by Jim Roth) and the Global Professional Services and Frontier Deployment Engineering organization (led by Lori Steele) each have a monthly budget of $500 (about RMB 3,340) for technical roles and $100 (about RMB 668) for non-technical roles.

Industry Trend: Big Tech Tightens Employees’ AI Purse Strings

In fact, Salesforce is not alone. As enterprise AI applications enter deeper waters, using budgets or caps to regulate employee AI spending has become a common choice among tech giants.

Previously, U.S. crypto exchange Coinbase set a weekly AI spending cap of between $500 and $5,000 based on employees’ specific roles and seniority; financial giant JPMorgan also set a monthly limit of $2,000 for some engineers using Anthropic’s coding tool Claude Code.

Back in May this year, Salesforce CEO Marc Benioff publicly said the company expected its full-year spending on Anthropic tokens to reach about $300 million, with the vast majority going into various coding agents. As this huge expense increasingly shifts toward refined operations, how companies balance innovation speed and financial health is becoming a required question in the AI era.