One of McKinsey's top AI leaders explains how it's stopping staff from overspending on tokens

Consulting firms moved quickly to make AI portion of employees' moving lives. Now they are confronting the costs of utilizing it astatine scale.

One of McKinsey's strategies for managing rising costs isn't to headdress consultants' usage of AI but to fto them cognize erstwhile they're utilizing excessively much.

The patient tracks depletion connected a user-by-user ground and alerts its labor via email erstwhile their AI usage gets excessively high, said Debasish Patnaik, who leads QuantumBlack — McKinsey's AI, data, and analytics group — successful the UK. The alert strategy was introduced firmwide successful the summer.

"Similar to utilizing mobile information connected a activity phone, we show them this is really you could do things to make it much cost-effective for the firm," said Patnaik.

Companies that initially focused connected encouraging AI experimentation are now confronting the costs of operating the exertion astatine scale.

The property of freewheeling AI token use came to an extremity earlier this twelvemonth arsenic LLM providers moved from subscriptions to consumption-based pricing models. Consulting firms and different endeavor users are now charged based connected the number of tokens — the mini chunks of matter an AI exemplary sounds and produces — they use, alternatively than a level entree rate, making ratio much important arsenic usage grows.

In September, OpenAI announced that its astir prolific users of AI coding agents now consume much than $7,000 worthy of tokens a day.

McKinsey's attack to managing that spending unit rests connected transparency and education, Patnaik said.

By May 2026, the patient was processing astir 5 trillion AI tokens a month, according to a institution blog post. Consumption was highly concentrated: astir 10% of users accounted for astir 65% of the total, pinch consultants and package engineers among the heaviest users.

The AI usage alerts aren't intended to discourage labor from utilizing AI, but to show them what they are consuming and thief them complete the aforesaid activity much efficiently.

"We really judge successful giving autonomy to the consultants," said Patnaik.

Ensuring they person the correct acquisition astir really to usage devices has been cardinal for managing usage, he added, noting that McKinsey has already seen shifts successful really labor usage the devices since giving users much visibility into their AI habits.

Beyond token alerts

McKinsey's soul AI spending has not yet reached a problem stage, Patnaik said.

Most of the firm's AI usage is for individual productivity aliases to present engagement faster, some of which are cases wherever "the cost-benefit is still skewed towards the use side," he said.

Usage levels could beryllium much "egregious" successful different six months, but Patnaik said the patient had already implemented further controls to negociate usage successful summation to alerts.

An soul AI gateway optimizes requests earlier they scope exemplary providers, while circuit breakers temporarily region entree astir peculiarly precocious token usage while the patient assesses whether the usage is productive.

Caching enables answers to repeated questions to beryllium reused and pools depletion costs crossed the business alternatively than locking unused capacity into individual licenses.

Other consulting firms are pursuing akin controls to negociate AI spending.

The Big Four patient EY has set up an "AI Value Realization Office" to negociate AI spending and has installed an "invisible" router down immoderate specialized AI devices that directs labor to the exemplary champion suited to a task. EY told Business Insider that the router, alongside different governance measures, had reduced token depletion by 60% since April.

In June, a elder package technologist astatine Deloitte US told Business Insider that the changes to GitHub's pricing exemplary are "already wreaking havoc" connected expectations for work, pinch developers quickly burning done their caller monthly quotas, which took effect that month.

Taking the strategies to clients

Patnaik's domiciled arsenic leader of QuantumBlack successful the UK chiefly focuses connected providing AI capabilities to McKinsey's clients, alternatively than the firm's soul usage of AI. McKinsey established a general believe this outpouring to counsel clients connected deploying AI cost-effectively.

Patnaik's wide proposal for clients mirrors the firm's soul approach.

Over the past quarter, clients person progressively realized there's a "hidden costs that we haven't wholly thought through," and executives are having to measurement AI's imaginable competitory benefits against its return connected investment, Patnaik said.

Patnaik said companies should measure token costs per result alternatively than per employee. Simply optimizing for less tokens could discourage valuable usage of AI.

"What you don't want is to return costs retired here, but incur costs connected the different broadside without knowing astir it," he said. For now, he added, there's still worth successful encouraging adoption, alternatively than restricting it.