The short answer.
A company centralizes five things about AI: the assistant it pays for, the ground rules, the one plan, the order of what gets built and the owner of each workflow, by name. Teams decide three things: which of their work to propose, the detail of each workflow and the checking of what each workflow prepares. Central decisions are made once and apply to every team; team decisions are made by the people who do the work, inside those five.
The split is the core of our point of view: a company manages how it plans, develops and deploys AI in one place, and its teams do the work. How to manage AI across teams sets out the plan and the record that hold the centre together. This piece draws the line itself, item by item, and gives you a test for every question the list does not cover.
Five things the company decides once.
The company decides five things centrally, because each one goes wrong when every team decides it separately.
- The assistant the company pays for.One company account of ChatGPT, Claude, Copilot or Gemini, approved as the default for everyone. Two assistants means two bills and workflows that do not move between teams. Should we standardize on one AI assistant? covers the choice.
- The ground rules.Which assistant is approved, what information may go into it, and the rule that a person decides before anything is sent, paid or changed. Written once, short, and the same for every team.
- The one plan.One plan for the whole company, with every team, its roles and the workflows each role runs. One plan shows the handoffs between teams, where work gets lost most often.
- The order of what gets built.Teams propose; the plan’s owner sets the order across the company. Without one order, every team’s first workflow starts at once and none of them finishes.
- The owner of each workflow.One person, by name, for every workflow on the plan. The centre makes sure each one has an owner and records who it is.
Three things each team decides.
Each team decides three things, because the team is the only place that knows the work well enough to decide them.
- Which work to propose.The team lead names the pieces of work that repeat every week and start from records the team keeps. The centre cannot see which reply or schedule eats a team’s week.
- The detail of each workflow.The steps, the inputs, what good output looks like and the exceptions. The team’s own people write it, because they know which requests always need a second look.
- The checking.A named person in the team reads what each workflow prepares and decides before it goes out. The workflow prepares; that person decides. Checking at the centre turns one reviewer into a queue.
Where does a new question go?
A new question about AI goes to the centre when a wrong answer in one team causes work or risk for another team, or when two teams would otherwise answer the same question twice. Every other question stays with the team that asked it.
- A team wants a second assistant for one task.Central. The account, the rules and the bill are shared.
- A team wants to change the wording of its own replies.Team. The workflow’s owner changes it and records the change.
- Two teams have each started a workflow for the same piece of work.Central. The plan’s owner keeps one version and names one owner.
- A workflow should pull from a business system the company runs, such as your booking system.Central for whether the connection is allowed and what a person approves; team for which fields the workflow uses.
One place to plan, develop and deploy AI.
The five central decisions live in one place, so that every team finds the same assistant, rules, plan, order and owners. That place is the plan, and the plan covers all three steps: plan it, develop the workflows, deploy them.
FullyPowered Planner is where that plan lives. You set out your teams, their roles and the work between them, and choose what to build first; each team lead adds their own team and proposes its work, so the team decisions land inside the central plan. We develop the workflows with you in a 3-Day Workflow Sprint or a 2-Week Workflow Sprint, and FullyPowered Studio captures the know-how a team already has. The workflows run in your company’s own assistant, on your own account, run by your own people.
A central and team split to copy.
Copy these lines onto one page and fill in a name or a decision on each. Write “Unknown” where nobody knows yet, with who will find out.
- Central: approved assistant.Which one, and who holds the company account.
- Central: ground rules.Where they are written, and who keeps them current.
- Central: owner of the plan.One name.
- Central: the build order.The next three workflows, in order, and when the order is next reviewed.
- Central: owner of each workflow.One name per workflow on the plan.
- Team: proposals.For each team, the lead who proposes its work.
- Team: detail.For each workflow, who in the team writes and updates the steps.
- Team: checking.For each workflow, the person who reads the output and decides, and their stand-in.
- Questions that cross teams.Who a team lead asks when a question fails the test above.
A veterinary group draws the line.
Take a 130-person veterinary group with several clinics. The office, scheduling and client-communication work sits in three teams: the front desk at each clinic, a central scheduling team and a small client-communications team.
Illustrative example.
- Central.The group approved one assistant, on the company account, and wrote its ground rules once: no card details go into the assistant, and a person reads every message before it reaches a client. The operations director owns the plan and set the order: appointment confirmation and reminder messages first, replies to booking questions second, the weekly rebooking list third. Each workflow has one owner: the head of scheduling owns the reminders and the rebooking list, and the client-communications lead owns the booking replies.
- Front desk at each clinic.The front desks proposed the booking replies, because the same questions arrive every day by email and message. They wrote the detail: how each clinic phrases its hours, its parking and its drop-off times. Each clinic’s front desk lead checks the replies.
- Scheduling.Scheduling proposed the rebooking list, built from the booking system’s export of missed and cancelled appointments. A scheduler reads the list each week and decides who gets a call.
- A question that crossed the line.One clinic wanted its own assistant for the booking replies. That question went to the centre, because the account and the rules are shared. The answer was the group’s assistant, with that clinic’s wording in the workflow.
The clinics keep their voice and their checking; the group keeps one assistant, one set of rules and one plan. How to keep AI use consistent across locations covers what each clinic may change inside one workflow.
What is different in a company of under 200 people.
In a company of under 200 people, the centre is one person, not a department. You give the five central decisions to the person who already runs operations, and the plan is the AI office.
- The centre and the team overlap.The person who owns the plan often owns a workflow too. Keep the two jobs on separate lines of the split, so everyone knows which hat a decision comes from.
- The ground rules fit on one page.Write them this week.
- Teams propose in conversation.You hear each lead’s proposals in person, so the order of work is set in one meeting, not through a request process.
Done this way, central control costs your teams nothing: the decisions they need are made once, and the work stays theirs. That is how you 10X the ROI of your existing AI.
Draw your line and put it in one place. FullyPowered Planner holds the plan, the order and the owners, and your colleagues add their own teams. How organizations plan their use of AI covers the rest of the method.