A prompt, a capability template and a capability.

Picture Friday morning in a fully powered finance team. Before nine, the billing coordinator has every overdue invoice sorted. Routine reminders are drafted in the firm’s own wording. The firmer notes wait for the account leads whose names go on them. The disputed invoice is held with its note attached, and an unidentified wire has stopped two chases before they started. By lunch the messages people approved have gone out, sent by the people who approved them.

The team is doing the work it always did, with the consistency of its best person on every invoice. That is how the AI you already pay for starts to 10X what your team can do. Three things get called “AI” on the way there: a prompt, a capability template and a capability. Knowing which one you hold tells you what is still yours to decide.

One task, three ways.

The task is overdue-invoice follow-up preparation at an invented 60-person professional-services firm. Each week the firm decides, invoice by invoice, what happens next: a short reminder, a request for a payment date, a direct note from the account lead, or nothing yet.

Four roles touch the work. The billing coordinator prepares the follow-ups and sends the early reminders. Account leads send the firmer notes to their own clients and resolve disputes. The controller releases anything held and decides any credit or payment plan. The finance lead is copied on the firmest notes. In every version below, AI prepares and a person sends.

A prompt: one person, one request.

A prompt is a request one person types into ChatGPT, Claude, Copilot or Gemini. Here is a good one.

Illustrative example with invented data. A prompt for the overdue-invoice task.

I’ve attached this week’s aging report. Draft a follow-up
email for every overdue invoice. Polite but firm. Include
the invoice number, amount and due date, and ask the client
to confirm a payment date.

Six drafts come back, and all six read well. Four should not go out. Invoice 1035 is disputed, and a chase will make that conversation harder. A $5,600 wire arrived on Wednesday with no reference, and it pays either Invoice 1044 or Invoice 1045. Invoice 1042 shows $200 open because Client A paid short, and nobody has confirmed why. Of the two that should go, Invoice 1031 gets the same polite reminder as the rest, although the client promised payment by 11 September and the note belongs to the account lead. The coordinator knows all of this and typed none of it.

A careful coordinator catches it at review. A colleague covering for a week writes a different prompt and has no reason to look. A prompt belongs to one person, runs once, and is checked only by whoever reads the answer.

A capability template: the reusable starting version.

A capability template is Fully Powered’s reusable starting version of a capability. It writes down what the good coordinator carries in their head: the roles, the inputs, the procedure with its decision rules, the exceptions that stop the work, the output format, the reviewer’s checklist and the tests. Where your firm makes its own call, it leaves a placeholder or a default you can replace.

Illustrative example with invented data. The procedure from a capability template, condensed. Bracketed placeholders are the firm’s to set.

Each gap the prompt left now has a rule. The template also proposes acceptance tests, such as a missed promise or two invoices that could match one wire, each with its expected behavior written before anyone runs it. What it does not yet know is your firm: your brackets, your approved wording, who sends at each stage and who is copied.

A technical aside: on some platforms a capability template is packaged as a skill, a folder of instructions and reference files the AI application loads when the work comes up. What is a SKILL.md file? explains that packaging for the technical reader.

A capability: what your team can now do.

The capability is the template adapted to your rules, proven on your own invoices, owned by a named role and in weekly use. Nobody writes a prompt. The coordinator starts the work, the AI prepares the plan, and each person decides their part.

Illustrative example with invented data. The capability as the team experiences it on a Friday.

That is the destination: the same standard every week, whoever is at the desk. The owner approves each rule change and has the tests run again. The full plan and one of its drafts are on the overdue-invoice follow-up preparation page.

Where the workflow fits.

A workflow is how a piece of work runs in your organization. At this firm, follow-up runs on Friday once aging is refreshed and the exceptions are listed, with firmer notes routed to account leads and holds to the controller. Another firm runs it twice a month, with partner approval above a set balance. Your workflow decides when it starts, which system feeds it, who reviews and where the drafts go.

This work crosses teams. Start from the outcome: cash collected on time, client relationships intact. Map the handoffs between billing, finance and the account leads, and land the capability on each role. Prove it on real invoices, then roll it out.

Side by side.

Illustrative example with invented data. The overdue-invoice task, compared.

Where to start.

This task is one of five in the receivables week shown in What a fully powered finance team knows before month end, and the weekly receivables summary is where its follow-ups are reported. Browse the other capability templates, or read about Fully Powered Capabilities. The guide to programmatic AI workflows covers the engineering. To choose your own first piece of work, use the sheet in The AI subscription is paid for. Here’s how to 10X it.

Fully Powered works across ChatGPT, Claude, Copilot and Gemini, at the pace your team sets. Your people keep control of how the work gets done.

Bring one recurring task and the person who does it best.