Input Metrics Over Output Metrics
Stop managing the revenue number. Instrument the controllable customer-experience inputs that produce it.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 95%
Outputs — revenue, active customers, free cash flow, share price — cannot be managed directly, which is why quarter-end shortfalls produce useless fire drills. Inputs are the things you control that touch the customer experience: selection, price, speed, ease of finding and buying. Amazon mapped these onto its growth flywheel, built rigorous metrics for each, and ran its weekly business reviews and operating plans on them, taking it as an article of faith that improving the inputs would take care of the outputs.
Origin
Born from Amazon's 1999–2001 growth wall, and explicitly credited by Carr to two outside sources: the flywheel concept from Jim Collins' Good to Great (which Bezos and the S-team were reading at the time, and which Carr calls the single most influential management book for the company), and the DMAIC cycle — Define, Measure, Analyze, Improve, Control — from Six Sigma.
Core principles
- 01Outputs are lagging and largely uncontrollable; inputs are controllable and causal.
- 02A metric is an input if (a) you control it — you can apply resources to make it better or worse — and (b) it typically touches the customer experience.
- 03No customer will ever wake up wanting less selection, higher prices, or slower delivery — inputs anchored to those truths are durable for decades.
- 04You will get the measurement wrong at first; refining the measurement is part of the work.
- 05Never munge several important metrics into one compound index.
How to run it
- 1
Identify the flywheel
Codify the small set of things that, if improved, compound into growth. Amazon's were broad selection, great customer experience (ease of finding, ease of buying, delivery speed), low prices, more merchants on the platform, and driving out cost.
Pro tip Test each candidate against the ten-year question: all else equal, would a customer ever prefer less of this?
- 2
Map the end-to-end customer experience
Walk every step from the ad click to the completed experience. Carr's Airbnb worked example: ad click, first screen, browse and search, property detail page, reserve, host messaging, then the actual rental. Every step is a place to instrument speed, ease, and quality.
Pro tip Ask 'how are we measuring the speed, ease, and quality of this action?' at every single step.
- 3
Throw a lot of things at the wall
You do not yet know which inputs are most causal to the outputs. Instrument several candidates, including more than one way of measuring the same input, and accept that some will not matter.
Pro tip Where a metric's direction is ambiguous (are more host-guest messages good or bad?), instrument it anyway and let the data resolve it.
- 4
Run DMAIC on each input, one at a time
Define, Measure, Analyze, Improve, Control. Iteratively measure, observe, improve, and watch the effect on the outputs — refining which inputs are truly causal and how accurately you are measuring them.
Pro tip Amazon's most important input — how much selection they had — was measured incorrectly for several years before the measurement was refined.
- 5
Rewrite goals, operating plans, and reviews around inputs
Make the input metrics the substance of your weekly business review, operating plan, and goal-setting. The output number stops being the topic of conversation and becomes the consequence of it.
Pro tip Audit your goal list: on Amazon's ~500-item S-team goal list around 2007–2008, only about 10 items had a financial metric in them.
Watch out Leadership must actually hold the line here; the pull back to output-chasing is constant.
In the wild
Facing a miss with a month to go, Amazon would run around like chickens with their heads cut off — promotions, price cuts, extra emails, extra ads. After several quarters of this they noticed the drills produced no meaningful progress against the number, were a big distraction, and mostly pulled forward revenue that would have arrived next month anyway.
→ The company shifted from short-term revenue-pushing to long-term input improvement, which is what the flywheel work made possible.
Early on, Amazon let autonomous teams run against a 'fitness function' — take the five or six metrics that matter most for a team, weight them, and combine them into a single index to track up and down. Within several months to a year they realised the compound metric obscured exactly what it was supposed to reveal: which actions were producing the good outputs.
→ The fitness function was abandoned and each metric broken out and managed individually. Carr now actively discourages compound metrics anywhere.
Common mistakes
Building a compound metric
Munging several important metrics into one index makes the result meaningless — you can no longer see which action produced which effect. Carr has watched this fail twice, including a variant where the team moved a different contributing metric each quarter and consequently never got good at any of them.
Treating the quarterly miss as a problem to be solved this quarter
Last-minute promotional and pricing levers are near zero-sum, distract the team, and let you avoid the real question — whether you are working on anything that moves the needle for customers over the long term.
Waiting for objective proof before committing
Carr says plainly that he still cannot give objective proof that serving customers well produces revenue and free cash flow. Amazon took it as an article of faith. A leadership team that demands proof first will never make the shift.
Is it for you?
Best for
Executives and product leaders whose weekly reviews are dominated by a lagging revenue number they cannot directly move.
Not ideal for
Very early-stage teams with no instrumented customer journey and too few users for input metrics to be legible.
From the transcript
“we took it as an article of faith that if we can just improve these inputs the outputs will take care of themselves”
“map your end to end customer experience”
“do you control it meaning can you apply resources to make make this thing better or worse”
“this is a concept from uh Six Sigma”
“only 10 of them actually had a financial metric in it”
“these fire drills don't really work”
“I discourage teams and companies from creating any sort of compound metrics”
From the episode
Unpacking Amazon’s unique ways of working
Bill Carr (author of Working Backwards)