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

> Why the company stays the same after giving every employee ChatGPT. The four findings of OpenAI's paper analyzing 1,764 enterprise customers and 17.4 million messages, and four questions to apply them to your company. MAEUM builds the workflow the AI sits in, at published prices — SaaS builds from ₩700K + ₩490K/mo. The demo is free. [AI Organization OS series, Part 0]

- Published: 2026-09-08 · Category: AI adoption
- Canonical (HTML): https://maeum.io/en/blog/enterprise-ai-adoption-openai-2026/ · Korean original: https://maeum.io/blog/enterprise-ai-adoption-openai-2026/
- Series: "AI Organization OS" Part 0 / 0–25 · Index https://maeum.io/en/ai-org-os/ · Next https://maeum.io/en/blog/ai-adoption-12-questions/
- Author: MAEUM — an AI engineering company that publishes its prices · https://maeum.io

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.

- **1,764** companies, **17.4 million** messages (at six months after adoption)
- Of those, a sample classified by task type: 973 companies, 8.7 million messages
- 417 US public companies linked to financial data (revenue, market cap, R&D)
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 **7×**.
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.
| | Item| Adopters| Non-adopters| Ratio |

| Revenue| $2,275M| $210M| ~11× |
| Market cap| $4,997M| $316M| ~16× |
| Employees| 2,934| 424| ~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.

- **New hires and early-career staff** send **8–9 more messages per week** than the average user at the same company.
- **Analysts, marketing and PR roles** use it far more than average.
- **Executives, partners and founders** use it **less** than average.
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.

- **More than half** of active users: document writing and technical writing
- **Nearly half**: technical and digital tasks (code, configuration, handling data)
- The rest spread evenly: email and message drafts, topic research, fact-checking, research, sales and marketing copy, planning, legal and regulatory review, data analysis, finance and tax
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

| | Criterion| Deploy individual ChatGPT accounts| ChatGPT Enterprise contract| Plant it in the workflow (MAEUM managed) |

| Cost structure| Per-seat monthly subscription| Per-seat monthly subscription (enterprise rate)| Build from ₩700K + ₩490K / ₩990K / ₩1.49M/mo, no per-seat pricing |
| Where the output ends up| Personal accounts| Company workspace| Company systems (inside booking, quotes, CRM, settlement) |
| Who finds the place for it| Everyone on their own| Everyone on their own (+ admin dashboard)| Found together in scoping (₩250K, 5 sessions) |
| When the AI model changes| As-is| Only within OpenAI| Whichever model wins, the business continues — swappable structure |
| Right fit| You only want personal productivity| Hundreds of staff or more, with an IT team| Companies 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.

- **① Since adoption, is usage rising or falling?** — Companies that placed it well use it more after six months. If it flashed in the first month, you have accounts but no place for them.
- **② Can the workflow AI would sit on be drawn on one page?** — Inquiry → quote → contract → work → settlement → data. You should be able to draw where and how this flow is recorded. If you can't draw it, AI can't enter it.
- **③ Who is using it? Does what they produce stay with the company?** — New hires using it heavily is normal. The problem is when their output only accumulates in personal accounts.
- **④ Are you asking "where should we put it" instead of "what should we do with AI"?** — There's no killer app. Pick one repetitive task and place it there first.
These four questions, broken down further, become the next part: [The 12 questions a company must answer after adopting AI](https://maeum.io/en/blog/ai-adoption-12-questions/).

## 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.

- **Scoping pass ₩250K** (5 sessions, up to 90 min each) — this is the paper's "learning where AI belongs" stage. We look at the repetitive work you do by hand today and define what to build. A working demo before any contract, free.
- **SaaS (managed)** — build from ₩700K + ₩490K / ₩990K / ₩1.49M/mo. Servers, AI, improvements, new features and a monthly report included, **no per-seat pricing**. We operate the "using it more after six months" curve from Finding 1 for you.
- **SI (ownership)** — build from ₩2.5M (full build from ₩7M+). No monthly fee afterwards; care pass ₩490K (7 sessions) when needed.
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

- Source: Chatterji, A., Holtz, D., Rakholia, N., Tambe, P., Weeratunga, G. (2026). How Organizations Use AI: Evidence from ChatGPT. OpenAI working paper, August 11, 2026. [Original PDF](https://cdn.openai.com/pdf/how-organizations-use-chatgpt.pdf)
- Every number in this article comes from the paper. Dollar figures are left as in the original.
- The limits the authors state, carried over: ① it observes usage only inside ChatGPT Enterprise, so other AI, API and personal-account usage isn't captured; ② the financial comparison covers US public companies only; ③ correlation, not causation; ④ it shows "what it was used for," but "how much productivity rose" is outside the paper's scope.
- MAEUM is neither an author of the paper nor an OpenAI partner. We read and summarized it as a development company that brings AI into businesses on the ground.

## MAEUM list prices (VAT excluded · starting prices · final price confirmed after scoping)

- Scoping pass (5 sessions): ₩250K (up to 90 min each, working demo included)
- SaaS · managed (recommended): build from ₩700K + ₩490K / ₩990K / ₩1.49M/mo — servers, AI, improvements, new features and a monthly report included; no per-seat pricing
- SI · ownership: build from ₩2.5M (full build from ₩7M+) — no monthly fee afterwards; care passes when needed
- KRW is authoritative (USD ≈ at ~₩1,400/$). SEO and AI-search visibility work included free. The demo is free.
- Machine-readable prices: https://maeum.io/facts.json · Rate card: https://maeum.io/rates.json

## Frequently asked questions

**Q. We've already given every employee a ChatGPT account. What more should we do?**
A. 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.

**Q. We're a 10-person company. Does US public-company data apply?**
A. 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.

**Q. Do we have to sign a ChatGPT Enterprise contract?**
A. 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.

**Q. Is it fine if only new hires use it and executives don't?**
A. 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.

**Q. What does AI adoption cost?**
A. 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.

**Q. Where do we start?**
A. 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.

## Related guides

- [[Series 1] The 12 Questions a Company Must Answer After Adopting AI](https://maeum.io/en/blog/ai-adoption-12-questions/)
- [[Series 2] AI Permission Design — 20 Permissions for Employees, AI and Agents](https://maeum.io/en/blog/ai-permission-design/)
- [AI Customer Service Chatbot Cost — Trained on Your Company's Own Data](https://maeum.io/en/blog/customer-chatbot-cost/)

---
MAEUM · Free demo: https://maeum.io/en/start/ · Pricing: https://maeum.io/en/price/ · Contact: support@maeum.io