A room can leave a prompt workshop with a page of patterns and a good mood, and Thursday's real task can still go out the old way. People use a new pattern when they have had to pull it back out, on work they already owe.
Thursday's job has a reviewer, a deadline and source material the slide never had, and the old method is already in their hands. The notes stay in the folder unless people have to retrieve the new pattern on work they owe.
What the Accenture staff remembered a day later
Kana Okano, Jakub Kaczmarzyk and John Gabrieli reported the comparison in PLOS One. Accenture staff watched a required DevOps video of eight minutes and 20 seconds. The paper describes Accenture as a firm of more than 401,000 people serving clients in more than 120 countries. Accenture paid for the study and helped find volunteers. The authors say the firm had no role in the design, the analysis or the write-up. Of 153 volunteers, 99 came back the next day.
One group watched the way the company usually ran the video. One group talked it through afterwards with no guide, for no more than 10 minutes. A third had a guided discussion of the same length: the key points, the benefits of DevOps, and how those benefits made the practice work. The last group answered multiple-choice questions about once a minute and saw the correct answer before the video went on.
On a memory test 20 to 35 hours later, watching scored 55.45 percent and the unguided chat scored 56.84 percent. Guided discussion scored 69.53 percent. Questions through the video scored 69.70 percent. The authors call that 26 percent better than watching for the questions, and 25 percent better for the guided talk. They measured memory for the video, and they say forgotten material cannot be applied on the job. They did not measure work performance. Fifty-four of the 153 people never came back.
Copy the two conditions that worked
A prompt session is a different room, and the memory problem is the same. If people cannot recall the pattern a day later, they will not use it on the draft that is due. Pause while the example is still on the screen, ask what the pattern was, show the answer, then continue. In the study those questions came about once a minute because the video was short. On a longer day, keep that shape and show the answer before the next idea.
After the demonstration, run a guided conversation and keep it under 10 minutes, which is as long as the discussions in the study ran. Ask what the pattern was, where it changes a task owed this week, and what they will check before sending. An open chat about how capable the tool feels matches the unguided condition, which did not pull ahead of watching.
Have them draft the task before they leave, and say what they would send and what they would still check. Then put the same pattern on a live task the next day, with the reviewer looking at the output. The researchers waited 20 to 35 hours because memory falls fastest in the first nine hours, and a day later stood in for what people still hold a month on.
Bring a task someone still owes
Our AI training workshop is a half day or a day on the team's own work, with the reviewer in the room so the standard is visible. What a one-line prompt does to a draft is in your team types a query and keeps the first draft. What happens when the first week looks busy and the old pack returns is in your AI rollout stalls in week two.
Bring one task a named person will still owe after the session, and the name of the person who reads the result. A fictional brief leaves the patterns in the folder. Send the task and the reviewer, and we will build the session around that job.
