Claim
"Long feedback loops" are usually a failure of measurement design, not an inherent property of the question. For any outcome, find an early signal that correlates with it and measure that on a short interval.
Mechanism
Decision velocity is bounded by feedback velocity. If the only signal arrives in 12 months, you get one update per cycle. But almost every long-horizon outcome has earlier proxies — engagement, retention curves, qualitative interview signals. Operationalising a leading indicator gives weekly or monthly checks on a quarterly or yearly bet, raising the rate of learning by an order of magnitude.
Conditions
Holds when:
- The outcome has identifiable leading indicators (most do).
- The team has the discipline to track the indicator without over-fitting to it.
- Leadership accepts proxy signals as input to decisions, not just final outcomes.
Fails when:
- The leading indicator decouples from the long-term outcome (Goodhart's law applied).
- The proxy is too noisy to drive decisions at short intervals.
- Management treats the leading indicator as the goal and forgets the actual outcome.
Evidence
"Did this positioning resonate?" is a long loop. "Did we get 3+ qualified demos from this week's outreach?" is a short loop.
— Annie Duke on Lenny's Podcast, 2026-04-28
Signals
- Decision cadence shifts from quarterly to weekly without losing rigour.
- Teams identify two or three leading indicators per major bet and track them in a shared dashboard.
- Long-term outcome reviews show that the leading indicators were directionally right.
Counter-evidence
Some outcomes are genuinely lagging (retention at 24 months, brand perception, deep behavioural change) and any short-term proxy will mislead. Archie Abrams' Shopify holdout discipline is the pair to this — short signals can lift while long signals stay flat.
Cross-references
- Force intuitions into explicit predictions so you can find out where you are wrong — the prediction discipline that uses these loops
- 30–40% of growth experiments with short-term lift show no incremental value at one year — Archie Abrams' counter-pair on long signals