Controllable Inputs

2026-01-31

In 2001, Jim Collins flew to Seattle and introduced the flywheel concept to Jeff Bezos. Bezos grabbed a napkin and started applying it. “What are the inputs?” he kept asking.¹ Lower prices would attract more customers. More customers would attract more sellers. More sellers would increase selection. Better selection would improve customer experience, which would drive more traffic, which would lower costs through scale, which would allow lower prices. The flywheel.

The diagram was obvious; the discipline was breaking the loop into pieces a single team could move. SKU count. Page load time. Seller signups. Cost per unit shipped. Each input connected directly to a team’s daily work, could be measured without waiting for quarterly earnings, and gave fast feedback. A product manager could wake up on Monday morning and know exactly what to build, how to tell if it was working, and why it mattered.

By 2009, Bezos had codified the logic into Amazon’s annual goal-setting process. From that year’s shareholder letter:

Senior leaders that are new to Amazon are often surprised by how little time we spend discussing actual financial results or debating projected financial outputs. To be clear, we take these financial outputs seriously, but we believe that focusing our energy on the controllable inputs to our business is the most effective way to maximize financial outputs over time.²

Amazon set 452 detailed goals that year. Of those, 360 directly impacted customer experience. The word “revenue” appeared eight times. “Free cash flow” appeared four times. “Net income,” “gross profit or margin,” and “operating profit” appeared zero times.

Most companies set goals like “increase revenue by 20%” or “grow active users by 50%,” then cascade them to individual teams. Product gets revenue targets, engineering gets user-growth targets, and everyone nods in planning, then spends the quarter shipping features that float, untethered, toward outcomes no one can trace.

Alignment Theater

OKRs codify the mistake. “Increase revenue by 15%” is measurable, but it is also a lagging indicator shaped by dozens of variables no single team controls; teams ship features for twelve weeks with nothing moving on any dashboard until the quarter ends. The result is alignment theater: every goal connects on paper, no one’s daily work connects to their KR. Ask yourself whether your team’s KRs this quarter connect to anything they can move by Friday.

Between shipping a feature and seeing it move revenue lies a long chain of confounders: competitive dynamics, market conditions, sales cycles, pricing decisions. In enterprise the problem compounds: new features get rolled into existing contracts at renewal, sales uses them in conversations, deals close, and no one knows whether the feature mattered. By the time the signal reaches the team that built it, it’s noise. Optimizing for an outcome you can’t measure or control is hoping you get lucky, which is most of what quarterly planning rewards.

Every Book in 60 Seconds

A mission says what; a tenet says how. Kindle’s mission was seven words: deliver every book in the world in under 60 seconds with at least 12 hours of reading. The tenets that followed were prioritization decisions frozen in prose, each one a tradeoff that a team could apply on any given Tuesday without escalating.

“Books need to be at least 50% cheaper.” When the marginal cost of delivering bits is zero, a digital book should cost half the physical one; build tools that reduce the cost of getting a book to a reader such that at 50% of the current price, authors still earn more than before.

“Better than physical books.” People have read the same way since paper was invented; changing a behavior that old requires magical experiences, and unless a feature is magical, err on the side of the physical book. It took Amazon four and a half years to ship the first Kindle because of that tenet.

“Bring authors closer to readers.” Remove every barrier between author and reader without compromising value for the reader. That tenet eventually produced Kindle Direct Publishing, now larger by annual revenue than Penguin Random House.

Each tenet is a controllable input in disguise: “50% cheaper” points a team at cost levers, “better than physical” tells the hardware team when to ship and when to wait, and “authors closer to readers” names the intermediaries the platform team should remove. A good tenet makes tradeoffs explicit and points to controllable inputs; a bad one (“build scalable systems”) applies to every team and tells none of them what to do next. The hierarchy matters: when two tenets conflict, the one listed first wins, which means the ordering is itself a prioritization framework the team can apply without a meeting.

Cost Per Kilogram

SpaceX shows what happens when controllable inputs feed a single constraint. The dominant constraint in rocketry is cost per kilogram to orbit. Everything else, thrust-to-weight ratio, booster reusability, test cadence, feeds that one number. In October 2024, a Starship booster flew back to the launch tower and was caught mid-air by mechanical arms. Eighteen months earlier, the same design had exploded four minutes after liftoff. The Merlin 1D already achieves a 184:1 thrust-to-weight ratio, and Starship had flown 11 test flights by late 2025.

Drop cost per kilogram low enough and you stop competing with other launch providers and start unlocking markets that didn’t exist: satellite internet, Mars colonization, economics that were arithmetic on napkins a decade ago. The target is a threshold: get cost to the point where the market restructures around you.

Quarterly Fog

In stable markets, lagging outcomes work as proxies because cause and effect stay close enough to touch. In startups and in AI-accelerated markets where features ship daily and competitors move weekly, the OKR cadence of “set quarterly goals, measure at quarter-end, adjust next quarter” generates almost no signal. Controllable inputs give you daily feedback; most companies never find them because naming the levers is harder than naming the goal.

This framework assumes the causal chain is traceable. Sometimes it is not. A company entering a genuinely new market may have no idea which inputs matter, and no amount of discipline can name a lever that hasn’t been discovered yet; for those companies the honest work is running experiments fast enough to find the inputs, not pretending they already know them. Collins and Bezos drew a flywheel on a napkin, but they drew it in 2001, six years after Amazon’s founding. They had the data to know which arcs were real. I have this tendency myself. I reach for the flywheel too early, naming levers before I’ve run enough cycles to know which ones are real. Most companies do the same: they draw the flywheel on day one, before they’ve earned the right to name the levers. The discipline is knowing when you’re guessing. Nothing moves until you name what moves it, and nothing is harder than admitting you haven’t found it yet.