Our recommendation
Buy ChatGPT Enterprise or Microsoft Copilot for general writing, research and everyday productivity across the company. Build a custom assistant or agent when it has to act inside your own systems, follow a specific workflow, be measured against an evaluation set, run inside your network, or ship inside a product your customers use.
At a glance
| Custom AI assistant | ChatGPT Enterprise or Microsoft Copilot | |
|---|---|---|
| Best at | One workflow, done reliably | General productivity for everyone |
| Time to value | Weeks | Days |
| Your systems | Reads and acts through your own APIs | Connectors to common apps |
| Quality control | Evaluation set and regression tests | Vendor's general quality |
| Where it runs | Your cloud, or inside your network | Vendor's cloud |
| Model choice | Any; benchmarked on your data | The vendor's models |
| Training on your data | None; you approve each model provider in writing | ChatGPT Enterprise: not by default, under OpenAI's enterprise privacy commitments |
| Inside your product | Yes | No |
| Price | Build cost, then model usage at the provider's rates | Per seat. Microsoft lists the Microsoft 365 Copilot add-on at $30 per user per month paid yearly; OpenAI quotes ChatGPT Enterprise through its sales team |
Choose Custom AI assistant if…
- The assistant must take actions in your CRM, ERP or ticketing system, not just answer questions.
- A wrong answer is expensive, so you need it measured before and after every change.
- Data cannot leave your network, so a hosted workspace is not allowed.
- You want the AI inside your own product, for your customers.
- Usage is concentrated in one high-volume workflow where per-seat pricing is wasteful.
Choose ChatGPT Enterprise or Microsoft Copilot if…
- You want everyone in the company writing, summarising and researching faster.
- The work is varied and nobody can define a single correct answer.
- You do not have an owner for a piece of custom software.
- You want to learn where AI helps before investing in a build.
Most companies need both
A workspace like ChatGPT Enterprise or Copilot raises everyone's baseline. A custom assistant automates one job end to end. They do not compete; the workspace often shows you which job is worth building for.
The difference is the definition of correct
An off-the-shelf assistant is judged on whether it is generally helpful. A custom one is judged against a list of real cases with the answers your team expects. That evaluation set is what lets you trust it with a support queue, a document intake process or a sales follow-up, and it is what makes improvements measurable rather than a matter of opinion.
Acting, not just answering
Connectors let workspaces read company data. A custom agent can do the work: create the ticket, update the record, draft the reply for approval, and log what it did. We put human approval on every step where a mistake would be expensive.
Agents inside the workspaces
Workspaces now let teams build simple agents inside them, and for tasks that stay inside those tools that is often enough. A custom agent is the better choice when it has to act across your own systems, follow your permission model or be tested against your own cases.
Where the model runs
For regulated data, a custom build can run open-weight models inside your own network, or hosted models in your own Google Cloud or Oracle OCI account under the provider's API terms. You choose which providers see your data, in writing.
Questions
Does ChatGPT Enterprise train on our data?
Not by default. OpenAI says it does not train its models on business data from ChatGPT Enterprise, ChatGPT Business or its API unless you opt in. For a custom build the question is the same: which providers see the data, and under which terms.
What do ChatGPT Enterprise and Copilot cost?
Microsoft lists Microsoft 365 Copilot at $30 per user per month on a yearly plan. OpenAI quotes ChatGPT Enterprise through its sales team. A custom assistant costs a build fee and then model usage at the provider's rates, which for one high-volume workflow is often less than seats across a team.
Is there a ChatGPT Enterprise alternative that runs in our own cloud?
Yes. A custom assistant can use models served in your Google Cloud or Oracle OCI account, or an open-weight model in your network. It does one job rather than everything a workspace does, so many companies run both.
Can a custom assistant use GPT, Claude or Gemini?
Yes. We benchmark several models on your evaluation set and design so the model can be swapped later.
Can we start with ChatGPT Enterprise and build later?
Yes, and it is often the right order. The questions people ask the workspace are a good source for the evaluation set of a custom build.
Sources
The vendor pages behind the facts on this page, checked on the date above.