Copilot or ChatGPT Enterprise: what a seat buys, and what it never will

Two products come up in nearly every first conversation we have with a mid-sized company in Germany: Microsoft 365 Copilot, because Microsoft is already in the building, and ChatGPT Enterprise, because someone on the leadership team has been using it personally and wants the company version.

The comparison is worth doing properly. It is also worth saying at the start where it leads: both are reasonable purchases, they solve the same class of problem, and that class is not the one most companies are actually complaining about.

What each one is

Microsoft 365 Copilot puts a model inside the tools people already have open (Word, Excel, Outlook, Teams) with access to the tenant’s own content through Microsoft Graph. Its distinguishing feature is not the model. It is that the model can see your SharePoint, your mail and your meetings without anyone building an integration.

ChatGPT Enterprise is a workspace around the frontier models: unlimited higher-speed access, a longer context window, admin controls, SSO, no training on your data, and connectors to the systems you point it at. Its distinguishing feature is that it is the better chat product, used by people who will actually use a chat product.

The money, as of mid-2026

Microsoft 365 Copilot is $30 per user per month on an annual commitment, and requires a qualifying base licence: E3, E5, Business Standard or Business Premium. That base licence is the part that gets left out of the business case.

After Microsoft’s July 2026 suite price increases, the realistic all-in figure is roughly $69 per seat on E3 and around $90 on E5.

Before any Copilot Studio agent consumption. For organisations under 300 users, Microsoft 365 Copilot Business runs $18 per user per month promotionally against a $21 standard price, through the end of 2026.

ChatGPT Enterprise has no published price. Procurement reports in 2026 cluster between $45 and $75 per seat per month, around $60 typical, with a reported 150-seat minimum and annual prepay. Below that threshold, ChatGPT Business is $20 per user per month billed annually, or $25 monthly, with a two-seat minimum.

Verify both before you budget, these numbers move, and the Copilot base-licence uplift in particular is the one that turns a signed-off business case into an awkward second conversation with finance.

Where each one wins

Copilot wins on reach. If the value you want is “summarise this Teams meeting,” “find what we agreed in that thread,” “draft this reply with the account history in view,” Copilot has the data by default and nothing needs to be built. For a company whose knowledge genuinely lives in SharePoint and Outlook, that is a real advantage and it is unmatched.

Copilot wins on procurement. The agreement exists, the DPA exists, the data stays inside a tenant the security team already accepted. In a German mid-sized company this is often the entire decision, and pretending otherwise is naive.

ChatGPT Enterprise wins on the work itself. For drafting, reasoning through a messy problem, analysing a spreadsheet nobody can explain, or writing and running code against a file, the experience is better and people notice within a week. Adoption follows quality here more than it follows convenience.

ChatGPT Enterprise wins on the people who matter most. The employees who will get real leverage out of a model are the ones who push it, and they will hit Copilot’s edges quickly. Giving those twenty people the better tool is usually a better investment than giving four hundred people the more convenient one.

What neither of them does

Here is the part that gets discovered three months after the rollout.

Both products are assistants. They wait to be asked. Somebody opens a window, types a request, reads the answer and decides what to do with it. That model of work is genuinely useful and it has a hard ceiling: nothing happens when nobody is at the keyboard.

The problems companies describe to us are almost never of that shape. “Inbound enquiries sit for two days before anyone replies.” “Every quarterly review takes an analyst three days to prepare.” “Our CRM data is wrong, so the forecast is fiction.” “New leads never get checked against our ICP before a rep calls them.”

You cannot buy those away by the seat. Each one is a process with a trigger, a sequence of steps across two or three systems, a decision rule, and an owner. A licence gives every employee a better way to do their share of that work by hand. It does not make the work happen without them, and the cost of the work was never the typing.

If the answer to “who runs it?” is a person, you need a seat. If the answer should be “nobody, it just runs,” you need a system.

Most companies need both, buy only the first, and then conclude that AI underdelivered.

What we would actually do

Buy Copilot if the company is deep in Microsoft, the pain is document and meeting work, and there is no appetite for a project. Budget the base-licence uplift honestly, roll out to a department rather than everyone, and measure whether people still use it in month three. Copilot’s failure mode is not bad output. It is licences nobody opens.

Buy ChatGPT Enterprise, or Business below 150 seats, for the group that will push it hard, and do not agonise about whether everyone gets one. Uneven distribution is correct here.

Build a system for the two or three chains where the work is repetitive, the trigger is external, and the delay costs money. That list is short and every company can name it in ten minutes. A single automated chain, an email triage agent, a meeting prep agent, usually returns more than the seat licences did, because it removes the work rather than accelerating it.

The sequence we recommend is boring and it works: seats for the people who want them, one automated chain in production, then decide what the next one is based on what the first one taught you. Companies that start with a company-wide licence and no process work end up with an expensive habit. Companies that start with one chain end up with a system.

Frequently asked questions