Picture the meeting where the owner or managing partner approves AI for client work. Every question has a written answer: where the work runs, what anyone outside the business sees, what each connection can touch and who signed off. The approval takes one meeting, and your team spends the next month doing work it could not do before.
The short answer to “where does our client information go?” is that it stays where it already is: in the AI workspace your business pays for, under rules your business sets. Our technical article on prompt injection, permissions and MCP covers what to inspect, restrict and test. This article is for the person who approves.
It runs in your own AI workspace.
Every workflow we build runs in your own ChatGPT, Claude, Copilot or Gemini account, the workspace your business already pays for and already governs. Before work starts, your business confirms that workspace is approved for the information the workflow will use. We confirm the plan, permissions and access of the people who will run it. The readiness checklist walks through them.
Sensitive information can be used when your business has approved it for that workspace. Your AI policy, confidentiality obligations and professional rules still apply, and we work within them.
Invented examples for the build, real work on your screen.
To build a workflow we need your way of doing the work, not your records. You share the procedures and examples you choose to share, without production data or sensitive records. Where the know-how lives in people’s heads, we ask the people who do the work, and nothing records or monitors them. We then build and test the workflow on invented companies and cases.
Real work arrives at the end, and it stays in your account. Your person installs the workflow in your workspace with us on the call, then runs real cases while our named builder watches in person or on a screen share. We see the screen. We do not receive a copy, and when the share ends, our view ends.
What we keep afterwards holds none of your records: our general methods, templates and testing tools, the invented examples from the build, and the handover notes written for your owner, which you hold too. Your information and know-how remain yours, and how your team works is never material for another client’s workflow.
Each connection has one job, and your approver says yes.
A workflow does far more when it can pick up what it needs in one system and finish the job in another: the overdue invoices in the accounting system, the last exchange in email, a follow-up drafted and waiting for a person to check and send. That is what Fully Powered Connectors do. What gets connected is agreed when the work is scoped, and we use the system’s own connection first, then a proven third party’s, then one we build.
Three rules hold for every connection. What it can see, and what it can change, is set for each one, in the system that enforces those limits rather than in a line of instructions asking the AI to behave. Nothing is sent, posted or changed until a person says so. And your IT lead, or whoever approves your systems, approves each connection. Read-only access still shows information, so your approver looks at what a connection can read together with where its results can go.
What the Planner keeps.
Fully Powered Planner is where your company decides what each team and role should be able to do with AI. It stores what you enter so your plan is there when you return, and it keeps each workspace separate from every other. Your content is never used to train AI models. Fully Powered receives a summary of your plan only when you choose to send one.
Six questions to ask, with our answers.
Ask these of us, and of anyone else who wants to put AI to work on your client information. The third column is the point: every answer should end with someone in your business deciding.
| Ask | Our answer | Who decides |
|---|---|---|
| Where will the AI run? | In your own ChatGPT, Claude, Copilot or Gemini workspace, the one you already pay for. | Your business approves that workspace for the information the workflow will use. |
| What do you need from us to build it? | The procedures and examples you choose to share, without production data or sensitive records. We build and test with invented cases. | You choose what to share. |
| Will you see our real client records? | On screen only, while your person runs real cases in your account. We receive and keep no copies. | You approve our builder seeing real examples during testing. |
| What do you keep when the work is done? | Our general methods, templates and testing tools, the invented examples and the handover notes, which you hold too. | Your agreement sets out the rights to use and adapt what we deliver. |
| What can a connection see and change? | Only what is set for that connection, in the system that enforces it. Nothing is sent, posted or changed until a person says so. | Your IT lead, or whoever approves your systems, approves each connection. |
| What does the Planner keep, and does it train AI? | What you enter, in a workspace kept separate from every other. Your content never trains AI models. | You choose what to enter, and whether to send us a summary. |
Bring your review requirements to the fit call.
Fully Powered AI holds no security or compliance certification. What this article describes is how we work, not a certification, and we do not interpret your obligations or certify compliance. If your approval process needs a questionnaire or a signed document, bring it to the fit call, and we will confirm what we can provide and sign before you decide to proceed. Our security page sets the same commitments out in one place to share with your approver.
Your team works on the platform it already has, at the pace you set, and your organization keeps control of how the work gets done. That is how you get 10X the ROI of your existing AI without your client information going anywhere new.
Bring the piece of work you want to start with, and the person who approves your systems.