The friction audit: find the one stage bleeding energy

The most common mistake after adopting the flywheel is to start improving everywhere at once, which spreads a lean founder's limited effort so thin that nothing moves. The discipline is the opposite: find the single stage bleeding the most energy and fix that one first. A friction audit is how you find it, and it is a one-pass exercise, not a permanent committee.

Instrument every handoff

Walk the loop and put a number on each joint: capture, activation, pipeline, revenue, referral. For each handoff you want a simple conversion reading, what fraction of the thing entering this stage makes it cleanly to the next. How many captured leads actually activate, how many activated users enter the pipeline, how many pipeline deals convert to revenue, how many revenue customers expand or refer. You are not after precision to three decimal places, you are after the shape of where energy disappears, and the shape is almost always lopsided. One joint is usually leaking far more than the rest.

Attack the worst stage, not all of them

Once the numbers are in front of you, resist the urge to tidy up everywhere. Find the single highest-friction stage and put your effort there, because the return on fixing the worst leak dwarfs the return on polishing four joints that were already passing energy fine. The common leaks for lean founders cluster in predictable places: activation that takes too long so new customers never reach value, nurture sequences that went dead months ago and quietly convert no one, a referral ask that simply does not exist anywhere in the loop, and the total absence of an expansion motion so revenue per client never moves after the first sale. Any one of those can be the brake holding your entire wheel, and you will not know which until you instrument it.

Fix the mechanism, then prove it moved the reading

Finding the leak is half the job. The other half is fixing the actual mechanism rather than papering over the symptom, and then proving the fix moved the number. If slow activation was the leak, you change the activation flow and then you watch the activation-to-pipeline reading climb on real data, not on your hope that it worked. A fix you have not verified against the tachometer is a guess wearing a tidy bow. Only once the reading has genuinely moved do you go back to the audit, find the new worst stage, and take the next turn. That sequence, audit, fix the mechanism, prove it moved, repeat, is the entire operating loop of running a flywheel, and it is what turns a static diagram into a wheel that actually accelerates.