The RPM Loop

Guardian Life piloted a single AI tool against one specific bottleneck: how long it took to respond to a request for proposal. Five to seven days, most of the time. They ran the pilot small — one process, one team — and watched the number drop to 24 hours. Only after that did they start talking about where else it might go.

I call this an RPM loop in my book: a short cycle, weeks not months, that tests the smallest real version of an idea against a real bottleneck before anyone commits to scaling it. It isn't a circle — you don't run the same experiment forever. It's a spiral. Each loop starts near where the last one ended, but a little higher, carrying real learning instead of a five-year plan's best guess.

Most AI rollouts skip the loop entirely. They start at the top — the strategy deck, the vendor contract, the enterprise license — and only discover what actually works after the rollout is already company-wide and difficult to walk back. Two-thirds of AI initiatives, by one recent estimate, are still stuck in pilot precisely because nobody built a bridge between the pilot and the next-sized version of itself.

Guardian Life didn't skip the loop. They ran it once, honestly, on something small enough to fail safely — and let the result, not the roadmap, decide what came next.

What's the smallest real version of your next AI idea — the one you could test in weeks, not quarters?

From Dancing in the Wind: An Adaptive Leadership Framework in the Age of AI.