92% of Fortune 500 companies already use ChatGPT. Yet market research from Gartner and McKinsey estimates that 30 to 50% of artificial intelligence initiatives do not deliver what they should. The enterprise offering promises productivity and security, but it never measures its own usage once deployed. This guide answers the questions leadership asks before investing: features, contract, security, users. Then it asks the one nobody asks: does your company actually use the tool it pays for?
What is ChatGPT Enterprise and how much does it cost?
How it differs from the individual plans
The free plan serves individual needs and keeps access limited. ChatGPT Plus adds priority access, but it is still designed for personal use. The enterprise version plays in a different league: centralised access administration, contractual security commitments, connection to the organisation's internal sources. The difference is not the power of the model, which is the same or nearly so from one plan to the next. It is what leadership can administer, trace and arbitrate once the tool is deployed.
What the enterprise offering covers
The offering is sold per user seat, under a contract negotiated directly with OpenAI. Each organisation gets managed access: seats are added and removed in-house, without any heavy reinstallation. The contract includes dedicated support and stronger security commitments that individual plans do not offer. The company also keeps control over how the tool is integrated into its internal applications.
What company size is this offering right for?
The threshold is not company size but the number of people actively using the tool every day. A few dozen employees who use it for many tasks already justify moving to a dedicated offering. A larger organisation where only a handful of people use the tool will pay for seats that sit idle.
Per-user pricing, quoted on request
OpenAI publishes no public pricing for the enterprise offering. The bill depends on the number of seats, the level of support and the integrations requested. Ballpark figures circulate in market analyses, none confirmed by an official source: the real cost only becomes clear with a quote. What you can steer, on the other hand, is not the list price per seat but the number of seats actually used, the only variable leadership still controls at renewal.
The three plans compared
| Criterion | Individual plan | Team plan | Enterprise offering |
|---|---|---|---|
| Target audience | Individual user | Small team | Entire organisation |
| Access management | None | Basic admin console | Full console, single sign-on |
| Data security | Standard | Enhanced | Dedicated encryption, data not used for training |
| Available context | Standard | Extended | 128,000 tokens, 196,000 in reasoning mode |
The individual plan remains cut out for personal use. The team plan introduces a first level of collective coordination. The enterprise version goes further: it adds full administration and data protection designed for an entire organisation. Demanding leadership will find in it what it needs to answer for the compliance of every solution deployed.
What features does ChatGPT Enterprise offer businesses?
Extended context and performance
The context window extends to 128,000 tokens on the default model and 196,000 tokens in reasoning mode, against a standard window on individual plans. These thresholds go up with every model generation: what matters for leadership is not the exact figure but the order of magnitude, that of a complete file analysed in one go, without splitting it up. The usual usage caps also disappear. For employees who handle large contracts, reports or meeting minutes, this capacity changes their day-to-day work: they get a complete answer without firing off request after request or stitching the pieces back together by hand.
Connectors and company knowledge base
The tool connects directly to the applications the company already uses: SharePoint, Google Drive, GitHub, Slack, Outlook. The knowledge base queries these connected sources to give contextualised answers, without the team having to copy its documents into the interface. Each connected source remains administered by the company: read access, permissions by team and by department. That is what turns the assistant into an internal search engine you can query in natural language, rather than an isolated workspace.
Administration and access control
Each workspace comes with an admin console. Leadership manages single sign-on, roles and permissions by team from there. This level of control directly answers the requirements leadership sets before signing anything, and makes it easier to align the tool with the internal access governance policy.
Use cases by function
The most common professional uses span several functions of the company:
- Customer support: drafting template answers, summarising repetitive tickets
- Analysis: summarising complex financial or operational reports
- Marketing: creating content and adapting messages by audience
- Legal: a first read of contracts before human review
- Project management: meeting minutes and task tracking
Each use case deserves to be measured separately before being rolled out more widely. The best indicator is not the number of connected applications but the quality of the answer on specific, repeated tasks.
What does ChatGPT Enterprise really deliver once deployed?
What the adoption figures show
| Indicator | Value |
|---|---|
| French SMEs and mid-sized companies that have adopted ChatGPT | 32% |
| 12-month retention, enterprise offering | 88% |
| 12-month retention, team plan | 68% |
| 12-month retention, individual plan | 59% |
Adoption is moving fast, including in France, where nearly a third of SMEs and mid-sized companies say they have built ChatGPT into their processes. Twelve-month retention for the enterprise offering is well above that of individual plans, which confirms the contract holds over time. These figures describe contractual adoption. They say nothing about the utilisation rate inside each organisation, or about the quality of the answers obtained.
The gap between paid licences and licences actually in use
Is ChatGPT Enterprise secure and compliant for business use?
Encryption, certification and data not used for training
What the EU AI Act says about using ChatGPT at work
The EU AI Act classifies ChatGPT as a general-purpose model, deemed capable of handling a very wide range of tasks. This classification brings transparency obligations for the company using it, where employment law meets digital law. It notably covers informing the employees who use the tool regularly in their day-to-day work. We cover the topic in detail in our article on AI compliance.
The blind spot after deployment
How do you measure real ChatGPT Enterprise usage in your company?
Map real usage across all AI tools
Measuring a single solution gives a partial view of how work actually gets done. Most companies run several tools in parallel: Copilot, Claude, custom-built internal agents. Mapping real usage means cross-referencing these sources rather than relying on a single vendor's admin console. That is the principle behind our usage analysis across all AI tools. It is also why a vendor dashboard can never be enough.
Compare paid licences with licences actually in use
Our method starts from the logs and exports the company already generates on the platforms it administers, with nothing to install and no re-instrumentation. Whether exports are handed over or a read-only connection is set up is decided in a scoping meeting with the accountable leadership. This factual comparison between paid and active seats reveals what no vendor dashboard can show on its own, by design: what happens outside its own perimeter.
Decide to industrialise, consolidate or cut
| Step | What we do |
|---|---|
| 1. Measure | Map real usage across all AI tools, from existing logs and exports |
| 2. Compare | Set paid licences against licences actually in use, use case by use case |
| 3. Decide | Arbitrate the AI portfolio in the steering committee: industrialise, consolidate or cut |
Measurement is worthless without a decision behind it. First, measure real usage across all tools. Then compare paid and active licences. Finally, decide: industrialise what creates value, consolidate what overlaps between tools, cut what costs money without serving anyone. It is a quantified portfolio arbitration you can present to the steering committee in front of the finance department, not a general usage recommendation.
What should you put in place once usage is measured?
Measurement brings two things to light: the use cases that keep coming back and the seats that sit idle. What remains is to scale up the former and deal with the latter. What works at this stage is tooling, not instructions. Nobody works better because they are told to, whereas usage grows when the right asset is within reach. The three levers below are ranked from least costly to most demanding.
Publish the assets a few teams have already built
In any equipped company, a handful of employees have built their own custom assistants, projects and recurring instructions. These assets usually stay in their own workspace. Publishing them across the organisation costs nothing: measurement has already identified the ones that are actually used. A shared meeting-minutes template or customer-reply assistant gives fifty people what one person had worked out on their own.
Plug the tool in where the work already happens
Usage that forces people to switch applications stays occasional. Connectors to SharePoint, Google Drive, Slack or Outlook remove that break: the answer is built on company documents, in the environment where the question comes up. This lever requires configuration on the IT side and a decision on permissions by team, which makes it a short piece of work rather than a project.
Build agents for recurring use cases
When measurement reveals a task handled every week by several departments, it becomes a candidate for a shared agent: preparing a standard file, qualifying an incoming request, checking a document before it goes out. This is the most demanding lever, because it requires design, an owner and long-term maintenance. It is only launched on cases that measurement has identified, never on a hunch.
