Picture your finance team a year after its first AI pilot. The capability that prepares the weekly receivables review is on its sixth version. It knows this year’s credit terms, the three customers with special arrangements and the exception the team found in March. New hires learn it in their first week. That is 10X what your team can do, and it compounds.
The method that gets you there starts from the outcome, maps the work that crosses teams, lands each capability on the roles that do it, proves it on real work, then rolls it out. What keeps it improving afterwards is small: four responsibilities, each with a named role, written on one page before the pilot team goes back to its other work.
The day the pilot ends.
A pilot has a sponsor, a deadline and everyone’s attention. Then it succeeds, the people who ran it return to their own jobs, and the capability keeps running. Three months later a pricing rule changes, a system export gains a column and a new colleague asks where to send a question. Each of those needs somebody whose job it is.
The rollout checklist covers the first month: introduction, first use, review and the first follow-up. This table covers every month after that. Fill it in at the pilot’s closing meeting, while the people who know the capability best are in the room.
Four responsibilities, four names.
- Business ownership.The role accountable for the result the capability produces. This is the leader of the work, never IT and never whoever was keenest during the pilot. The owner decides who uses it, what good looks like and when it is retired.
- Technical maintenance.The role that keeps it running on your approved platform: access, permissions, the connections and exports it reads, and a check whenever the platform or a source system changes. This is where IT comes in, with a known and bounded job.
- User support.The experienced user who answers questions, shows new colleagues the first run and keeps the exception log. Pick someone who does the work every week.
- Change approval.Who may propose a change, who approves it and what must be retested before it reaches the team. Give this its own row even when the business owner holds it, because a capability edited by whoever has access stops being one standard.
One person can hold two rows in a small company. What matters is that every row has a role, a named cover and a rhythm.
The responsibility table.
Copy it, replace the brackets and keep it with the capability, beside its job aid and its tests.
| Responsibility | Held by | What they do | Rhythm | Cover |
|---|---|---|---|---|
| Business ownership | [role accountable for the result] | Sets the standard, decides who uses it, reads the exception log, reports the result. | [monthly] review; [quarterly] decision to widen, change or retire | [role] |
| Technical maintenance | [IT role] | Manages access and permissions, keeps connections and exports working, checks the capability after any platform or source-system change. | On every joiner, leaver and system change; [quarterly] access review | [role] |
| User support | [experienced user] | Answers questions within [one working day], introduces new users, keeps the exception log and the job aid current. | Daily, as questions arrive | [role] |
| Change approval | [approver], with [second approver] for [named kinds of change] | Receives proposed changes, approves or declines in writing, has the tests run again before release, records the version. | As proposed; released on [a set day] | [role] |
One company’s filled-in table.
An invented wholesale distributor piloted receivables exception review for six weeks. The capability prepares the list of accounts that need a person’s attention; the credit decisions stay with people. This is the table the controller wrote at the closing meeting.
Illustrative example with invented data. An invented 90-person wholesale distributor. Every role and detail is invented.
| Responsibility | Held by | What they do | Rhythm | Cover |
|---|---|---|---|---|
| Business ownership | Controller | Sets what counts as an exception, decides which of the three receivables specialists use it, reads the exception log and reports days-to-review to the finance director. | First Tuesday of each month; a widen, change or retire decision each quarter | Finance director |
| Technical maintenance | IT manager | Grants and removes access in the approved AI workspace, keeps the Monday aging export in the agreed format, reruns two saved test cases after any accounting-system update. | On every joiner, leaver and system update; access review each quarter | Systems administrator |
| User support | Senior receivables specialist | Answers questions the same working day, sits with each new user for their first run, logs every case where a specialist overrode the prepared list. | Daily | Second receivables specialist |
| Change approval | Controller, with the finance director for anything touching credit policy | Takes proposals from any user through the support lead, approves in writing, has all twelve test cases run again and records the version number and date. | As proposed; released on the first Monday of the month | Finance director |
The IT manager knows exactly which change triggers a check. A specialist with a better idea knows where it goes and when it ships.
Keep the table true.
Three checks keep the names current.
- Review it when a role changes hands.A departure, a promotion or a reorganization updates the table in the same week.
- Retest before every release.Change approval means the tests ran again. What we test before we call an AI workflow ready shows the record to keep.
- Keep the decisions with people.Ownership of the capability never moves a decision into it. The capability prepares; the named role decides, as AI prepares, people decide sets out.
If your team runs on EOS, give each capability a place in the structure of roles you already keep: Add AI capabilities to your Accountability Chart shows how. And once the owner is reporting a result, Installed, used or useful? gives them the three questions to answer.
All of this runs on the platform your team already has: ChatGPT, Claude, Copilot or Gemini. You set the pace, your people hold every row, and your organization keeps control of how the work gets done.
Bring the pilot that is about to end and the four roles you would write into its table.