(Yicai) June 11 -- As token consumption keeps rising without delivering the expected returns, big Chinese technology firms have started to do the math and impose limits on employees’ token use.

After an initial phase of blanket, equal distribution, major tech companies in China are quietly recalibrating token quotas for their staff, testing a more disciplined, efficiency-driven allocation model, Yicai learned from workers at these firms.

Tencent Holdings no longer hands out tokens on an egalitarian basis, but instead distributes them based on employees’ needs, job responsibilities, and other factors, according to an insider at the Shenzhen-based internet giant.

Artificial intelligence has become part of the daily production toolkit, and token quota management is moving from broad-brush allocation to more granular oversight, the person said. Tencent’s leadership has also made clear there will be no token usage leaderboard, because the real test of AI use is whether it lifts efficiency and creates value, not how many tokens an employee burns through, he said.

A worker at a Beijing-based tech titan said the use of AI tools was once an important key performance indicator, with higher usage taken as a sign of greater enthusiasm for innovation. That has now changed, he said. When his own monthly token use reaches half of his quota, his supervisor comes to discuss both his usage and the output it had generated.

The change follows a period in which heavy token consumption failed to produce clear gains, a staffer at another top tech firm said. A department with around 20 employees burned through tokens worth about CNY50,000 (USD7,380) in a single month, but achieved nothing, the person said.

Many companies raised their AI-related budgets over the past two years, but those increases have not always been driven by clear commercial plans, an industry professional noted. More often, they have reflected the fear of missing out (FOMO)on the new technology wave and falling behind competitors, they said.

Companies have realized that token use does not automatically translate into productivity improvements, the person said. If firms turn token consumption into a KPI, equate AI usage with innovation, and invest heavily without first streamlining workflows and organizational structure, even strong AI tools will have their impact diluted, they said.

Some firms only count the explicit cost of application programming interface calls when deploying AI, while overlooking hidden expenses such as manual verification and data governance, Zhang Yi, chief executive of iiMedia Research, discovered when speaking with many of them. Others use AI for low-value routine work, which makes it hard for the technology to show up in revenue or financial performance, leading to higher token usage but insignificant cost reductions, he added.

Using token consumption as an early-stage proxy for AI adoption can encourage employees to try the technology, but AI use still needs to be measured by whether it truly improves efficiency and maximizes corporate value, he pointed out.

Editors: Tang Shihua, Martin Kadiev