The Cadence Curve
The first months of anything new feel like the months where speed doesn't matter. Nothing works yet, there's nothing to show anyone, so why rush? That instinct is exactly backwards — and it's the most expensive mistake on the whole curve. Guess first, then run 24 months and watch what your calendar actually buys you.
Where does a one-week loop pay for itself most?
You have 24 months to take a new capability from nothing to working — a new product line, an AI programme, a skill you don't have yet. A decision loop is when you stop and ask "is this still the right thing?" It costs about a week of the team's capacity every time you run one. Where in those 24 months is a weekly loop worth the most?
Set how often you stop and re-decide.
The top chart is the capability you end up with. The bottom chart is what each loop actually told you — every bar is one decision loop, and its height is how much uncertainty that loop removed. Bars under the dashed line are loops that cost a week and taught you nothing.
In the Experiment phase
You don't know yet whether this is the right thing. Every loop is a keep / kill / pivot.
In the Process phase
The direction is settled. The work is execution, and it compounds if you let it run.
On the plateau
Returns have flattened. Loops here find less and less — the question is changing. A poor cadence earlier never gets this far.
Show the numbers as a table
| Phase | Cadence | Months in phase | Loops | Learned per loop |
|---|
The loop never stops. What it's for changes.
The loop produces a decision
Almost everything you believe is still a guess, so almost every loop overturns something. This is the phase where being wrong is cheapest — there's barely anything built to throw away — and it's the phase people run slowest, because there's nothing to show at a steering meeting yet.
Run it tight. Weeks, not quarters.
The loop produces value
You know what you're building. Re-opening the question now doesn't find new information — it just restarts work that was compounding. The loop becomes routine rather than deliberate.
Keep shipping fast; stop re-deciding fast. Those are two different clocks, and only the second one should slow down.
The loop produces nothing
Effort keeps going in and the curve barely moves. The temptation is to loop harder — more reviews, more tuning — and it buys nothing, because there's no uncertainty left to resolve here.
A flat learning rate is not a problem with your cadence. It's the signal that the question has changed.
The same mismatch, in four different rooms.
The annual pilot
A twelve-month pilot with a readout at the end is one loop. In a field where the tools change every quarter, that's a single guess dressed up as a programme — and the readout arrives describing a world that no longer exists.
The quarter that was already decided
Teams run tight fortnightly reviews on a product whose direction was settled two years ago, and call it discipline. Every one of those reviews is a week that didn't go into the thing that was already working.
Six months before the first check-in
You start something new and give it "a proper go" before judging it — six months head-down, then a review. The whole point of the early phase is that it's the only time changing your mind is nearly free.
The retro that never changes anything
A fortnightly retro on a mature, stable process reliably surfaces the same three items. That flat learning rate isn't a facilitation problem. It's telling you the interesting uncertainty has moved somewhere else.
One sentence: match how often you re-decide to how much you still don't know — not to the calendar, and not to the meeting invite. Early on, uncertainty is the thing you're actually manufacturing progress out of, so loop hard. Later, the uncertainty is gone and the loop is just overhead wearing the costume of rigour. And when the loops stop telling you anything at all, the question stops being how fast and becomes which curve — which is where The S Curve picks the story up.