Caselet · Innovation PM turns shaky AI deck into Δ proof

Week 0 · Hypothesis without proof

The PM team at a global logistics company built a “GenAI exec briefing assistant.” Demos impressed leadership, but the COO asked a simple question: “Show me the over/under compared to manual briefings.” The PMs had anecdotes, screenshots, and a stack of prompts—no shared Δ, no TLX notes, no reviewer minutes.

The PM lead rebooted the effort. Each Thursday she met with one venture PM, one researcher, and one reviewer. They picked a single deliverable—“24-hour expansion brief”—and ran it twice: manual first, AI second. Self-ratings, reviewer scores, and TLX pulses were logged immediately after each run. They also captured reviewer time to flag rework.

Week 1–2 · Instrument the ritual

• Baselines. Manual runs averaged 62 minutes of IC time and 14 minutes of reviewer edits. Self-ratings hovered around 7/10; reviewer scores averaged 6. Δ = +1. TLX pulses: mental demand 68, frustration 44.
• AI-assisted runs. With a locked prompt scaffold, the same deliverable took 37 minutes IC time and 11 minutes reviewer edits. Self-ratings jumped to 9, reviewer scores to 7. Δ = +2 (still safe). TLX dropped to 51/33.

Instead of writing a slide deck, the PM pasted four Analyzer tiles directly into the memo: Δ trend, TLX pulses, reviewer minutes, and annotated outputs. She also called out the guardrails: Fair Trial checklist link, prompt version, reviewer name.

Week 3 · Exec readout

The exec readout changed shape:

1. Slide 1 – Business story: “Expansion briefs now land in under 40 minutes with lower TLX.”
2. Slide 2 – Analyzer tiles: Δ (manual vs. AI) and reviewer rework side-by-side.
3. Slide 3 – TLX snapshot + link to /help/interpretation.
4. Slide 4 – Next run plan: “2 more cohorts, same prompt scaffolds, collect Δ weekly.”

The COO finally had proof: consistent guardrails, honest Δ, fatigue trending down. Funding released.

Make it your own

• Recreate the ritual with the Innovation PM quickstart. It already includes the Task Frontier + System Shift diagrams the team pasted into their deck.
• Drop TLX lanes from the quickstart directly into your memo so everyone knows what “52/35” means.
• Use /help/interpretation as the appendix so execs see how Δ and TLX translate into go/no-go guidance.

When skeptics ask “why trust AI for this ritual,” send them a memo full of tiles—not adjectives.