I used to think the briefing problem was a writing problem. Pick the sharper verb. Shorten the paragraph. Lead with the loss figure.
The real problem turned out to be who holds the pen.
The CRO who kept losing the cyber vote
Last month I was prepping a chief risk officer at a European specialty reinsurer. Her annual cyber brief to the board. Three paragraphs, a map, a loss figure. Four years running, the board had voted down her ask.
The deck was fine. The vignette was the kind consultants charge six figures for. I read it twice.
Then I asked her to tell me, out loud, what she would be doing on the Tuesday morning the breach actually hit.
She paused for a long time. Then: "I would be driving my son to school. The incident call would come in on Teams, in the car, from the CISO. I would turn around at the roundabout and go to the office. My son would be late."
The brief she had been reading for four years had never put her inside the Tuesday. Hers did.
What the Michigan paper measured
A team at the University of Michigan built a tool called the As-If Machine. A multi-agent AI that walks you through writing a first-person counterfactual story about your own life under a distant policy future. The AI gets you to write the brief for yourself.
They ran two preregistered Prolific experiments. Study 1 on US science funding, N=1,007. Study 2 on satellite cybersecurity, N=894. Three arms each: the As-If Machine, an expert vignette on the same scenario, and a placebo where participants co-wrote with an AI about an unrelated topic.
The science funding scenario moved psychological closeness 0.80 SD above the placebo. The cyber scenario, 0.90 SD. Policy attitudes shifted by up to 0.67 SD. The expert vignette moved both by much less.
The part that stopped me: the effect held among the AI skeptics. All eight causal-forest moderator coefficients came back non-significant, p≥.12.
The AI did not persuade. Across both studies, 23,496 AI suggestions were shown and 4,068 accepted. When the authors regressed per-participant adoption on outcomes, closeness came back b=-0.002, p=.963. The participants moved themselves by what they wrote.
What this changes in a boardroom
So we rewrote the first twenty minutes of her brief. Before the slides, each board member works a scaffolded counterfactual on their phone. Tuesday morning. Your car. The call. What you would do next. Then she takes the room.
The vote is in January. I will know then if the paper transfers.
Two things keep me up about this. The mechanism works on AI skeptics, which is most of a board. And the same tool is sitting on anyone's laptop. The next proxy fight. The next activist shareholder resolution. The next regulator consultation. Someone will run this on your board before you do.
The slide you built is the one they read. The one that moves them is the one they write.