Forward Deployed AI from QuantSpark

ChatGPT Enterprise rollout: a plan for mid-sized firms

Last updated 21 November 2026

A ChatGPT Enterprise rollout needs four decisions before staff are invited: who administers the workspace, who holds a licence, who may share reusable assistants and connected apps, and how use will be measured. ChatGPT is a tool you roll out, and QuantSpark's engineers are model-agnostic: they build on the models you already use. The plan below follows QuantSpark's own method.

Before you invite staff

Name the administrators and keep them few. Decide who holds a licence, and put each in a company workspace rather than a personal account. Confirm from the vendor's current documentation how your plan handles sign-in, groups and roles, and record the answer; this page states no plan limits because they change quickly. Write a short use policy: what data may go into ChatGPT, who approves a new use, and who to ask when unsure.

Reusable assistants and who approves them

If your plan lets staff create and share reusable assistants such as custom GPTs, decide who may publish one before launch. Give each a named owner on your side and a one-page description of what it does and what it must never do. A level 1 engineer builds small automations on tools you already run, and each has a named owner for the same reason.

Connected apps and company data

  • Connect read-only first, with a named approver for any change.
  • Fix files shared too widely before the connection, not after.
  • Treat each connected app as its own decision with its own approver, so adding one later does not skip the review the first one had.
  • Read OpenAI's help page on connectors for what your plan includes today.

This is level 2, Connect.

Driving adoption after launch

Buying a licence gives access, not a habit. The engineer sits with teams and measures who is using what, starting from a baseline of who has a licence, who uses it and how often. Each six-week sprint has a goal with a measure, a baseline and a target, reviewed in week six. Where a team's use does not rise against its baseline, adoption has stalled, and the engineer reads the usage figures and plans how more teams start. Share what works: the prompts and skills that carry your own terms and templates are worth passing between teams, and the engineer writes them with the people who do the work.

Measure use and rank what to build next

Record the baseline per team: who holds a licence, who uses ChatGPT and how often. Ask each team for the jobs it would hand to an assistant, then rank the list by time saved and by how hard each job is to build. The ranked list decides what is built next, and the baseline shows whether use is rising.

Costs and handover

You pay model and tool providers directly on your own accounts. Anything QuantSpark buys for you is passed on at cost, with no mark-up, within a cap agreed in each sprint goal. Everything built is handed over on the last day, so the rollout does not depend on the engineer staying.

Do you need outside help to roll it out?

No rule requires it, and a small team can follow these steps alone. An engineer inside the team is worth considering when people are not yet using ChatGPT every day or adoption has stalled in some teams. Forward Deployed AI is one such option, delivered by QuantSpark on a monthly retainer. The company-wide rollout guide covers the same steps for any assistant, including Microsoft Copilot, and the Claude Enterprise plan covers the other.

Where this sits in the five levels

StepLevelWho does it
Workspace, licences, policy1 UseLicences, prompts, simple automations
Connected apps and company data2 ConnectConnects tools, runs adoption
Reusable assistants and agents3 BuildBuilds simple tools and agents

Questions, answered

What do we set up before inviting staff to ChatGPT?

Name the administrators and keep them few, decide who holds a licence, and put each in a company workspace. Write a short use policy and decide who may share reusable assistants and connected apps. Check your plan's current documentation for sign-in, groups and roles, and record what it says.

How do we drive adoption after launch?

Start from a baseline of who has a licence, who uses it and how often. The engineer sits with teams, measures who is using what, and works to a sprint goal with a measure, a baseline and a target, reviewed in week six. Where use does not rise, the engineer plans how more teams start.

Who approves reusable assistants such as custom GPTs?

Decide before launch. Give each reusable assistant a named owner on your side and a one-page description of what it does and what it must never do, and name who may publish one. An assistant without an owner should not be shared beyond the person who made it.

Do we need outside help to roll out ChatGPT?

No rule requires it, and a small team can follow the steps alone. An engineer inside the team is worth considering when people are not yet using ChatGPT every day or adoption has stalled in some teams. Forward Deployed AI is one option, delivered by QuantSpark on a monthly retainer.