Independent practitioner; not affiliated with, endorsed by or certified by OpenAI. Product details on this page were checked against OpenAI’s help centre in October 2026 and are re-checked at the start of each engagement.
What is different about a ChatGPT rollout
ChatGPT for work is a separate workspace that your staff bring information into. Unlike assistants built into an office suite, it does not start from your documents and mailboxes. It sees what people paste or upload, what is attached to a shared GPT or project, and whatever your administrator allows it to reach through apps and connectors.
That shapes both the risk and the opportunity. The risk is that careful staff avoid it because nobody has said what they may upload, while less careful staff paste whatever is nearest. The opportunity is that a workflow can be packaged deliberately: a shared GPT or project with clear instructions, approved reference material and a defined output, owned by someone who keeps it current.
OpenAI offers more than one business plan, and the administrative controls differ between them. Product names, plans and settings also change frequently, so I check the current position against OpenAI’s help centre and your own admin console rather than assuming.
The readiness exercise: tenant-specific questions
Before any workflow is built, I run a working session with your workspace administrator, your data protection lead and the pilot sponsor. We answer a set of questions about your workspace, and write the answers down:
- Plan and identity. Which plan do you hold? Is single sign-on in place? How are users and groups provisioned, and can feature access differ by group on your plan?
- Upload rules. Which categories of information may staff paste or upload (public, internal, confidential, personal data, client data), and which may they not?
- Apps and connectors. Which are enabled, for whom, and with what access? Where a connector can take actions as well as read, are actions allowed?
- GPTs and projects. Who may create them? How widely may they be shared? Who owns a shared GPT when its creator leaves?
- Retention and sharing. What retention setting applies to conversations? Can chats or GPTs be shared by link, and to whom?
- Measurement. Who can see workspace analytics, and what will the pilot report from them?
Any question nobody can answer is a finding. It usually explains a good part of why adoption has been shallow.
The pilot approach
Choose workflows that suit the platform. The best first workflows in ChatGPT combine a defined set of reference material with a repeatable output. Illustrative examples: first drafts of proposal sections from an approved boilerplate library; summaries of supplier documents against a standard checklist; analysis of a non-personal data export with a fixed set of questions. Tasks that depend on live data in other systems are better left until the connector decision is settled.
Package each workflow. Each becomes one shared GPT or project with written instructions, approved reference files, an output format and a named owner. The owner is responsible for updating the reference files and retiring the GPT when it is no longer needed.
Coach on real work. The pilot team uses each workflow on live tasks. I run short sessions using their own material and then sit alongside people as they work. Managers are asked to request the new workflow in their normal routines, which matters more than any training session.
Review points. Each workflow has a reviewer who checks output before it leaves the team, against an agreed checklist.
The signature deliverable
You receive tenant-specific readiness questions, an approved-workflow pilot and a user coaching plan. Illustrative extract from a workflow record:
| Field | Entry |
|---|---|
| Workflow | Proposal section first drafts, bid team |
| Packaged as | Shared GPT, workspace-only, owner: bid manager |
| Reference files | Approved boilerplate library, current case summaries cleared for reuse |
| Not permitted | Client personal data; unreleased pricing |
| Reviewer | Bid manager, before any draft enters the proposal document |
| Measures | Time to first draft; reviewer edits per section; repeat use by bid writers |
Illustrative example. It shows the format, not a client’s configuration.
How acceptance is judged
The sponsor and your workplace technology owner agree the measures before the pilot starts. Acceptance means the readiness answers are signed off by IT and data protection, each workflow is in use on live work with its owner and reviewer, and results are reported against the baseline. The coaching plan sets out how the next teams will be onboarded, by whom.
Ownership and handover
Your administrator owns the workspace and its settings. Each workflow owner owns their GPT or project. The sponsor owns the rollout decision. For further teams, internal AI champions are usually the cheaper route. If you are still weighing ChatGPT against Microsoft 365 Copilot or Google Workspace with Gemini, start with AI tool rollout and adoption support.