AI is growing explosively, deepening rapidly, and widening the gap between frontrunners and laggards

“The real leap is not in a clever model, but in clever embedding: repeatable tasks, fixed tools, and well-structured systems.”

With The State of Enterprise AI, OpenAI publishes a report that focuses on one central question: what happens when large organizations stop “trying out” AI and start using it as permanent infrastructure? The report explicitly does not look at a single sector or profession. Instead, it focuses on organizations as a whole: teams, functions, processes, and systems.

What makes this report strong is the combination of two sources. On the one hand, OpenAI uses anonymized and aggregated usage data from corporate clients. On the other hand, there is a survey of 9,000 employees at nearly 100 organizations regarding usage, returns, and bottlenecks. An important detail is that the content of messages was automatically classified and not viewed by OpenAI employees for this analysis. Consequently, this article does not provide a collection of future promises, but a realistic picture of the current moment, featuring solid growth figures, clear patterns, and cases that demonstrate where AI is already having a measurable effect.

Key takeaways from the report

• OpenAI now serves more than 7 million ChatGPT “workplace seats” (this is a user license, comparable to a Microsoft 365 license: one seat equals one employee with access), and the number of ChatGPT Enterprise “workplace seats” is growing approximately 9 times year-over-year.
• Since November 2024, the number of weekly Enterprise messages has grown approximately 8 times, and the average employee sends 30% more
• Usage is shifting toward repeatable workflows: users of custom assistants are growing approximately 19 times, and about 20% of all Enterprise messages are handled through such fixed setups.
• Employees report real returns: an average of 40 to 60 minutes of time saved per active day, with peaks of 60 to 80 minutes in fields such as data science, engineering, and communications.

A clear gap is emerging: “frontrunners” use AI much more intensively and therefore derive demonstrably more benefit.

From isolated questions to fixed tools: custom assistants are taking over

OpenAI highlights two shifts that explain this deepening. The first is the rise of custom assistants and projects: configurable environments with instructions, knowledge, and actions, intended for repeatable, multi-step tasks.

The figures:

• Weekly users of these custom solutions are growing by approximately a factor of 19.
• About 20% of all Enterprise messages are processed via such a custom assistant.
• Some organizations are truly building a culture around this. BBVA (a Spanish bank), for example, uses more than 4,000

Why is this such a significant point? Because this is exactly how organizations standardize: not by giving “everyone a smart chat,” but by capturing knowledge and steps in something that can be shared and reused.

The API side is growing just as fast, but under the hood

The second shift is the growth of usage via the API, meaning models that are built directly into products and systems.

This is evident from the figures below:

• 9,000+ organizations collectively process 10 billion+ tokens.
• Nearly 200 organizations even exceed 1 trillion tokens.
• On average, the consumption of “reasoning” tokens per organization increases approximately 320x in 12 months.

The report also shows that API usage is broadening. Customer service and content production already account for about 20% of API activity, and API usage by non-technology companies is growing 5x year-over-year.

Furthermore, Codex (OpenAI’s AI tool specifically built for software development) is gaining traction during this period: in six weeks, there has been 2x growth in weekly active users and approximately 50% growth in weekly messages. This is precisely the point where AI shifts from a tool to a building block.

What employees notice in practice: time savings, quality, and broader skills

The report does not stop at “AI is useful.” It links usage to perceived impact.

Time savings and quality in figures

• 75% of surveyed employees say that AI improves the speed or quality of their work.
• On average, Enterprise users attribute 40 to 60 minutes of time savings per day to AI.
• In data science, engineering, and communications, this increases to 60 to 80 minutes per day.

Impact per function, as reported by employees

• 87% of IT: faster resolution of IT problems.
• 85% of marketing and product: faster execution of campaigns.
• 75% of HR: better employee engagement.
• 73% of engineers: faster delivery of code.

AI opens up technical tasks

Another telling figure: 75% of employees indicate they can perform tasks they previously could not, including programming support, code review, spreadsheet analysis, automation, and even designing custom assistants.

And outside traditional technical departments, coding-related messaging has grown by an average of 36% in six months.

In practice, this means skills spread faster through the organization because the barrier to entry is lowered.

Growth by sector and country: The Netherlands ranks remarkably high

The report shows that growth is not limited to a few early adopters.

• The median sector is growing more than 6x year-over-year, and even the slowest sector is growing more than 2x.
• Fastest growers: technology 11x, healthcare 8x, manufacturing 7x.

Geographically, the picture is equally interesting. Among larger markets, Australia, Brazil, the Netherlands, and France are growing the fastest in terms of paying corporate customers. The Netherlands recorded 153% growth in one year, compared to a global average of 143%.

There is also acceleration internationally under the hood: international growth in API customers has been above 70% in the last six months, and Japan has the largest number of corporate API customers outside the US.

This is a reality check for anyone who still thinks that “it isn’t that significant in the Netherlands.”

The gap becomes measurable: frontrunners get much more out of the same tools

The report introduces a useful distinction: frontrunners are the “frontier” users, roughly the 95th percentile group in terms of intensity.

At the individual level

• Frontrunners send 6x more messages than the median user.
• Within data analysis, frontrunners use the data analysis tool 16x
• In coding, the gap is largest: frontrunners send 17x as many messages as the median user.
• Significant differences are also seen per task type, such as 11x in writing and communication, 10x in analysis and calculations, and 9x in instructions and information gathering.

And then comes the link to impact: users who deploy AI across approximately seven task types report five times as much time savings as users who stick to around four task types.

At the organizational level

• Frontrunner organizations generate 2x more messages per workplace than the median organization.
• And they send 7x more messages to custom assistants, indicating standardization and reuse.

Headroom: many people are not even using the most powerful features

Even among active users, much remains untapped.

• Of monthly active users, 19% have never used data analysis, 14% have never used reasoning models, and 12% have never used search functions.
• For daily active users, those figures drop to 3%, 1%, and 1%.
• The message is clear: routine builds proficiency, and proficiency delivers impact.

Six cases that demonstrate what “measurable impact” means

The report supports the broad trends with examples. Not as marketing stories, but with tangible figures.

Intercom: customer service via telephone, but faster and resolved more often

Intercom builds Fin Voice using OpenAI’s Realtime API.

• Latency has decreased by 48% since March.
• Fin Voice resolves an average of 53% of phone calls from start to finish on its own.
• Calls that still require a human are handled 40% faster after Fin Voice has taken the initial steps.
• Human customer service calls often cost 5 to 20 dollars per call, and Intercom states that this is now saving customers hundreds of millions of dollars per year.

Lowe’s: advice at scale in 1,700 stores

Lowe’s deploys Mylow and Mylow Companion.

• They support consistent advice in more than 1,700
• The systems have answered nearly 1 million questions per month since the launch in March.
• When customers use Mylow during an online visit, conversion more than doubles.
• Mylow Companion runs in 100% of stores and answers hundreds of thousands of questions per week from employees.
• Customer satisfaction rises by 200 basis points when employees use the tool in the store aisles.

Indeed: better matches, faster applications, more hires

• In trials, “Invite to Apply” with AI explanations resulted in 20% more started applications.
• It improves follow-up steps such as interviews and hires by 13% compared to traditional matching.
• Users of Career Scout find and apply for jobs 7x faster and are 38% more likely to be hired.
• 84% rate it as useful.

BBVA: standard questions eliminated, specialized work advanced

BBVA automates recurring questions regarding signing authority.

• The system automates 000+ questions per year.
• As a result, BBVA can shift capacity equivalent to 3 full-time positions to substantive work, accounting for 000+ checks per year.
• It delivers 26% of the relevant department’s annual savings target.
• Oscar Health: resolving many questions directly, less escalation
• The platform answers 58% of reimbursement questions directly.
• It can handle 39% of reimbursement messages without human escalation.

Moderna: compressing weeks of work by processing 300 pages more intelligently

At Moderna, the focus is on drafting Target Product Profiles, often based on evidence dossiers of up to 300 pages. AI assists in extracting facts and assumptions, creating structured drafts, and identifying potential errors, allowing teams to make better decisions sooner.

The silent but firm conclusion: AI maturity is a matter of organizational management

OpenAI does not say it in a single sentence, but it is written between the lines everywhere: technology is increasingly less of a bottleneck. The difference lies in how organizations organize it. This is also reflected in the reference to external performance figures: AI leaders achieve 1.7x revenue growth, 3.6x higher total shareholder return, and 1.6x EBIT margin.

This does not mean that AI automatically causes this. It does mean that mature deployment is correlated with better performance.

And that mature deployment consistently consists of the same building blocks:

• selecting repeatable tasks
• building and sharing fixed tools
• linking to systems and context
• measuring, adjusting, and training
• preventing a small group of frontrunners from carrying everything

The State of Enterprise AI is particularly valuable because it clarifies where we stand now: AI is becoming core infrastructure in many organizations, with measurable impact and clear growth curves. At the same time, the report shows that much capacity remains untapped, simply because people are not using the most powerful capabilities and organizations are not yet standardizing sufficiently. Those who use AI as a standalone text machine gain isolated wins. Those who embed AI into repeatable workflows build momentum, consistency, and scale. This is the movement OpenAI makes visible here, with figures that cannot easily be dismissed.