The British Red Cross lost the case-note pilot when the story could not be trusted

A falling line on a dashboard is a fact about use. It is not a fact about why someone stopped. The chart can be refreshed all morning and the reason can still be sitting with the person who closed the tab, unasked.

Simon Hodgkinson's account is from the Workforce track at aiblLIVE London on 20 October 2026. He is AI Product Owner at the British Red Cross, where he has worked for about five years and leads its work on AI. The page gives three figures as he stated them. About 2,000 of roughly 4,000 staff use Copilot every day. About 2,000 people were trained face to face in AI over the last two years, and the note on the speaker says he has trained about 2,000 people face to face. The page does not say those two lines are the same count, so they should not be collapsed into one. The third figure is a moment, not a headcount. Usage of a frontline case-note pilot began to fall from about week six.

The sequence he describes starts before anyone is asked for a use case. He taught people about the tool first and let the cases come to him. Asking teams to list cases is, in the phrase the page attributes to him, next to impossible. A mass education programme ran for staff and for volunteers. Within six months, the people who knew the jobs were bringing him ideas. That is a claim about where ideas started to arrive from. It is not a list of the ideas, and the page does not print one.

The case-note tool is the episode the account is willing to call a loss of users. It had strong backing. Usage then fell, from about week six. The dashboards could show the drop. They could not say why. Asking the staff could, and the page says the answer changed how the Red Cross approaches adoption. Changed, here, is the verb. The page does not describe a rebuilt tool, a new usage figure or a date when the pilot was restored. Restored would be a later count, or a description of the note as people now write it. Neither is here. A retelling that supplies them has left his account.

What the asking produced is one worker's reason, given in indirect speech. If the tool got a person's story wrong, it was on her. She read everything and rewrote it. That took longer than writing the note herself, so she stopped. The dashboard had the decline. It did not have that sentence. Hodgkinson's line for the wider point is blunt. "Staff and volunteers don't trust us just because we bought some software." The purchase is not the trust. The story, in her hands, was the trust, because a wrong story would be charged to her and not to the vendor.

The measure he offers is not another licence report. Licences show who can open the tool. Dashboards show how much it is opened. His question is whether someone would reach for it on a bad day, with nobody watching. The floor version he gives is direct. If we turned it off tomorrow, what would you miss? Use that appears only when people are watched, he says, is compliance. The page does not report how many people were asked the question, or what they answered. The question is the method. A percentage attached to it later would be a result this account does not contain.

He separates that method from a standard people like to put on a slide. "90% adoption with no changes is not as good as 30% who wouldn't give it back." Read it as the comparison he wants, not as the pilot's score. The pilot's published fact is that usage fell from about week six. Ninety and thirty are not given as counts of Red Cross staff. Treating them as the outcome of the case-note tool would be putting his yardstick into a cell the page left blank. The Copilot figure, about half the staff using it every day, is a different sentence again. Daily use of a general tool is not the same claim as a case-note pilot people would miss if it disappeared.

The governance he describes is short on purpose. He advises against opening with a twelve-page framework. Say what is never allowed, in words everyone understands and will support. Then give people an accountable person they can actually approach when something goes wrong, because that, he says, is a leadership task. The page does not print the list of what is never allowed. It prints the instruction to write that list in plain words, and to put a name beside it. A framework that cannot be said aloud, and a mailbox with no person behind it, fail the test he states even if both exist as documents.

Fear, in the takeaways on the page, is data. The move is to find out what each group thinks it is about to lose. The longer note is described as planning for five fears, in a range from irrelevance to exposure, and as explaining why trusted champions matter. This page does not name the five, and it does not say how a champion is chosen or what they are allowed to stop. Naming the range is not the same as possessing the list. An L&D lead who names the five from imagination has left the account.

Design, as the page puts the last takeaway, means fixing the process first and then giving staff a say in how the tool is used. The case-note episode is the reason that order exists. The page does not say the recording process had been redesigned before the tool arrived. What it shows is the worker doing the note twice, because a wrong story would be hers, and then leaving the tool. A say in the design, offered after people have already stopped, is a different conversation from a say offered while the process can still be changed. The page does not tell us which of those two conversations the Red Cross has since had. It tells us the principle.

Two headcounts sit next to each other and should be kept apart. Roughly 4,000 staff, of whom about 2,000 use Copilot every day. About 2,000 people trained face to face over two years, a line that may or may not be the same 2,000, because the page never joins them. Volunteers are inside the education programme and inside the trust quote, and they are not inside the Copilot staff figure. A single "half the organisation" sentence would have to decide, without evidence, whether volunteers are in the denominator. The account does not decide that. A chart that does has added a definition Hodgkinson did not give.

Week six is a start of a decline, not a full history. "Began to fall" does not say the week it reached zero, or whether some teams never fell. Strong backing is not a number of sponsors. Six months to the first ideas from people who know the jobs is not a count of ideas. Face to face is the form of the teaching he is willing to attach his name to, and the page does not itemise the curriculum. The longer note promises what that teaching covers. This page does not, so a curriculum written from it would be someone else's course of study, not his.

What an L&D or transformation lead can lift, without borrowing a recovery the page does not record, is the order of the questions. They taught before they asked for cases, because a list of cases from people who do not yet understand the tool is, in his phrase, next to impossible. They treated a falling chart as a reason to ask, not as a reason. They heard a worker say the story was on her, and they took the rewrite as evidence that the tool had made the job longer. They then set a test a dashboard cannot pass: would you miss it, on a bad day, with nobody watching. They asked for plain prohibitions and a named person. They did not, on this page, claim the pilot came back.

1. Teach before you collect the cases. A mass programme for staff and volunteers came first. Within six months, ideas arrived from people who knew the jobs. Asking for a list at the start is what he calls next to impossible, and the account does not pretend they did it that way.

2. When the line falls, ask the person who stopped. The case-note pilot had backing, and usage began to fall from about week six. Dashboards had the drop and not the reason. The reason was a story she would not let the tool get wrong, because the error would be hers, and rewriting it took longer than doing the note herself.

3. Keep the yardstick off the scoreboard. Ninety percent with no change in the work, against thirty percent who would not give the tool back, is the comparison he states. It is not a Red Cross result. Daily Copilot use by about half the staff is a third sentence. None of the three is "the pilot recovered."

4. Write the prohibition in words people will say, and put a name on the failure. Not a twelve-page framework as the opening. What is never allowed, said plainly, and an approachable person when it goes wrong, because that is a leadership task. Fix the process, then give staff a say. Fear is data about what a group thinks it will lose. The five fears are not listed here, and they should not be guessed.

A psychological safety session is a practical place to take one tool people have quietly stopped using, ask what they would miss if it disappeared, and put the person who carries the error in the room before the dashboard is treated as the explanation.