A team can draft with a model for months and still freeze the first time someone wants the tool to send the note itself. The draft stayed inside the building. The send is a promise, to a customer, a colleague or a regulator. Write down different owners for those acts before the pilot, or the person keenest on the demo becomes the owner by default.
Deloitte's UK workforce survey found that 31 percent of generative AI users do this without their employer's knowledge. Hayley McKelvey, chief AI officer at Deloitte UK, said the challenge has moved past getting people to use GenAI. Businesses have to meet that demand in a way that is secure, responsible and creates lasting value for both the organisation and its people. An agent that can send removes the pause from that problem.
What people do today
Most of what people do today is still a draft. The same survey says the common uses are search, email drafts and summaries. A person reads the result and decides. An internal agent changes the shape. It can open a thread, file a ticket or reply, once someone has given it the right.
Before that right exists, write four facts. What data can it see? Who receives the output? Can it act, or only propose? What happens on a weekday when it is wrong in public? If nobody in the room can answer, give the agent a draft folder and a named reviewer. Leave the send button off.
Who signs
Put the signature with the person who carries the downside. For a customer email, that is the owner of the relationship, with legal or risk in the room if the note can promise money, a remedy or a regulated statement. For an internal ticket, it is the process owner. For anything that spends money or changes a record, it is the control owner who already signs the manual version.
Learning teams should stay off this signature. They can teach the questions in an AI training workshop. They will be the wrong people to call when the message is wrong. The AI governance workshop is built for that split. Staff get a plain line, and the leaders who approve use get an operating model. What that leadership day has to settle is in AI governance training for leaders.
Write it before the pilot
One page, before the pilot, and well before the first complaint. An agent may draft into a queue. It may send, file or pay only after a named role has said yes for that class of action. Record the yes. A screenshot in a chat will fold under an inquiry. If people are already on unsanctioned accounts, deal with that first. That is the 31 percent. Leave the extra bot off the pile. The pilot has an end date. On that date the permission stays, shrinks or stops.
You cannot show a customer, a regulator or the executive a hope that people will be careful. You can show them the page, the named role and the end date.
Deloitte set the workforce numbers out in its UK workforce survey.
