MAEUM BLOG · AI adoption

Why Buying the Accounts Isn't Enterprise AI Adoption — What OpenAI Found in 1,764 Companies' Data (2026)

2026-09-08 · MAEUM — an AI engineering company that publishes its prices
What decides whether AI adoption works is not which AI you bought, but which task you placed it in. That's the conclusion OpenAI confirmed in its August 2026 paper from the actual usage records of 1,764 enterprise customers. The companies that adopted first were large ones whose workflows were already organized; the heaviest users were not executives but new hires; and there was no single "killer" use case. MAEUM doesn't sell licenses. We build, at published prices, the workflow the AI sits in — SaaS (managed) builds from ₩700K + ₩490K / ₩990K / ₩1.49M/mo, with a working demo before any contract, free.

Logs, not a survey — what this paper is, in 30 seconds

It's a working paper by OpenAI researchers together with professors from Columbia and Wharton: "How Organizations Use AI: Evidence from ChatGPT" (dated August 11, 2026). They didn't ask companies "do you use AI?" — they opened up the actual usage records of companies that adopted ChatGPT Enterprise.

No one had looked at who, at which level, uses it for what, at this scale before. Below are the four findings and how to apply them to your company. This article is Part 0 of the "AI Organization OS" series; the parts that follow each take one design question.

Finding 1 — Usage grew 7× in nine months. But half of that is companies that were already using it, using it more

From June 2025 to March 2026, the AI output (tokens) enterprise customers pulled grew .

Look at where the growth came from. It wasn't only new companies joining. Companies that had adopted before June 2025, taken alone, grew 4×. The paper writes that roughly half the growth came from "deepening among existing adopters."

The meaning is simple. AI adoption isn't an event that ends with "we bought it" — it's a curve that keeps growing, or keeps dying, after adoption. Companies that placed it well use it more six months later. Companies that didn't are left with accounts. If it flashed in the first month and went quiet, that company is on the second curve.

Finding 2 — The companies that bought first were already big, and already had "how we work" organized

Here's the comparison of 2024 medians between adopting and non-adopting US public companies.

ItemAdoptersNon-adoptersRatio
Revenue$2,275M$210M~11×
Market cap$4,997M$316M~16×
Employees2,934424~7×
R&D spend$113M$10M~11×

Big companies bought first — obvious so far. What matters is what comes next. Even controlling for company size, some variables raised the probability of adoption: cumulative SG&A, cumulative R&D, software assets. In other words, "companies that had already invested in people, processes and software" buy AI first. Conversely, companies heavy in plant and equipment buy less at the same size.

The paper calls these "complements." In plain language: AI doesn't work alone. It has to sit on top of an already-organized workflow. If customer inquiries arrive by chat, quotes live in spreadsheets, contracts in email and settlement in the owner's head — there's nowhere for AI to sit.

An honest caveat: this is US public-company data. Don't transfer it directly to a 10-person company. But the direction is the same. The paper itself says: "diffusion may initially widen existing gaps." The difference between companies with organized workflows and those without gets wider after AI.

Finding 3 — The heaviest users aren't executives. They're new hires

Six months after adoption, they looked at who inside the company was using it.

By number of people logging in, managers and team leads were the largest group (about 24%). Executives about 10%, new hires and trainees about 7%. A predictable order.

But by how deeply one person uses it, the order flips.

This isn't "executives are lazy." Executives use AI for briefings before decisions. In the paper too, executives skew toward topic overviews, fact-checking, legal and finance. They ask briefly and decide. New hires, by contrast, produce work directly with AI — they write documents, fix code, draft emails.

For a business owner, one thing to take from this: AI adoption is decided at the top and runs from the bottom. You don't need to use it every day yourself. Your job is to lay the road AI will run on. Without that road, one new hire works hard in a personal account, and when that person leaves, the output leaves with them. Nothing stays in the company's systems.

Finding 4 — There's no "just do this one thing" use case. They use it for everything

OpenAI automatically classified messages into 60 task types.

By message volume, three categories dominate: document writing, technical work, message drafts. Legal, finance and market research are used by many people but shallowly. And the "other" bucket is thick. The paper calls it a "long tail" — people are still finding new uses.

This is why MAEUM doesn't say "we'll put in an AI for you." There isn't one place to put it. Writing quotes, sending booking confirmations, cleaning up inventory spreadsheets, first-line replies to customer inquiries. Every company's places are different, and finding them is 80% of the work.

The two sentences the paper ends with

"Adoption is only the beginning of deployment."

"Firms are not merely deciding whether to use generative AI; they are learning where it belongs in their organizational workflow."

That's a one-line summary of what MAEUM does every day.

Three options — deploy accounts · enterprise contract · plant it in the workflow

CriterionDeploy individual ChatGPT accountsChatGPT Enterprise contractPlant it in the workflow (MAEUM managed)
Cost structurePer-seat monthly subscriptionPer-seat monthly subscription (enterprise rate)Build from ₩700K + ₩490K / ₩990K / ₩1.49M/mo, no per-seat pricing
Where the output ends upPersonal accountsCompany workspaceCompany systems (inside booking, quotes, CRM, settlement)
Who finds the place for itEveryone on their ownEveryone on their own (+ admin dashboard)Found together in scoping (₩250K, 5 sessions)
When the AI model changesAs-isOnly within OpenAIWhichever model wins, the business continues — swappable structure
Right fitYou only want personal productivityHundreds of staff or more, with an IT teamCompanies with repetitive work they want to keep in a system

These three aren't competitors. Individual accounts — let employees use them. MAEUM puts a company-level flow on top of that. If AI isn't needed, we don't add it.

Four questions to apply this to your company

The paper's four findings, turned directly into questions.

These four questions, broken down further, become the next part: The 12 questions a company must answer after adopting AI.

What MAEUM does — not licenses, but the place

In our terms, the paper's "complements" mean this: you buy the ChatGPT license. We build the place it sits.

Customers come in, you consult, contract, do the work, get paid, and data is left behind — every company already has a flow. We organize that flow first, connect where needed, automate what repeats, and leave only the judgment calls to people. AI sits on top.

VAT excluded; final price confirmed after scoping. Write one line about the task that's bothering you right now and send it. It can be small. That's where we start.

Sources and limits — what this article doesn't claim

MAEUM list prices — VAT excluded · starting prices · final price confirmed after scoping
Scoping pass (5 sessions)₩250K (≈ $179) · up to 90 min each · working demo included
SaaS · managed (recommended)Build from ₩700K (≈ $500) + ₩490K / ₩990K / ₩1.49M/mo (≈ $350 / $707 / $1,064) — servers, AI, improvements, new features and a monthly report included. No per-seat pricing.
SI · ownershipBuild from ₩2.5M (≈ $1,786); full build from ₩7M+ (≈ $5,000+) — no monthly fee afterwards, care passes when needed

KRW is authoritative; USD figures are approximate. SEO and AI-search visibility work is included in every build at no charge. The demo is free — try it, and sign only if you like it.

Frequently asked questions

We've already given every employee a ChatGPT account. What more should we do?
Good — that's the "runs from the bottom" part of Finding 3. What's left is laying the road from the top: making sure what employees produce with AI (quotes, replies, documents) ends up in company systems, not personal accounts. In a ₩250K scoping pass we decide together which task goes into a system first.
We're a 10-person company. Does US public-company data apply?
Don't transfer the numbers directly. The direction is the same — companies with organized workflows absorb AI first, and the gap widens. Small companies actually have an advantage: organizing the flow takes far less time.
Do we have to sign a ChatGPT Enterprise contract?
No. MAEUM doesn't sell a specific AI. We pick the model that fits the problem, place it where it's needed, and design it so the business keeps running when the model changes. If a task doesn't need AI, we don't add it.
Is it fine if only new hires use it and executives don't?
By the paper's data, that's normal. Executives use it to ask briefly and decide; new hires use it to produce output. The problem isn't the usage gap — it's whether what the new hires produce stays with the company.
What does AI adoption cost?
On MAEUM's price list, SaaS (managed) is build from ₩700K + ₩490K / ₩990K / ₩1.49M/mo (no per-seat pricing); SI (ownership) is build from ₩2.5M (full build from ₩7M+). VAT excluded; final price confirmed after scoping.
Where do we start?
By writing down one repetitive task you do by hand today. No specification needed. In a ₩250K scoping pass (up to 90 min per session) we look at the actual work together, define the scope, and show you a working demo before any contract — free.

See it working before you decide

Tell us what you do and what you need. We'll show you a working demo first — the demo is free.