Push any single number hard enough and a team will find a way to move it — sometimes by quietly breaking the thing the number was supposed to stand for.
’s law: when a measure becomes a target, it stops being a good measure. Tell a team to grow orders 20% and they will — but discounts can torch your margin, notification spam can drive uninstalls, and a sneaky checkout can inflate orders that all come back as refunds.
The fix isn’t to stop measuring growth. It’s to name the counter-metrics that must not degrade while growth climbs — quality, trust, cost, and long-term retention — and to watch them. A guardrail doesn’t create healthy growth; it lets you see unhealthy growth before it compounds.
Growth simulator — hit +20% orders, then survive the quarter
Each tactic adds order growth but also nudges the hidden dials. A guardrail you’re watching lets you catch a tripped counter-metric and pull the offending tactic — you lose that (fake) growth but spare the damage. A counter-metric you didn’t watch ships its damage silently, and it eats into next quarter.
Baselines: refunds 3% uninstalls 2% margin 30% repeat 40%
Tripwires: refunds >6 uninstalls >5 margin <24 repeat <37
Example — you ship a dark-pattern checkout to juice orders:
order growth = +14% (the headline looks great)
refunds 3 -> 12 (tripwire 6, breached by 6)
uninstalls 2 -> 7 (tripwire 5, breached by 2)
repeat 40 -> 34 (tripwire 37, breached by 3)
If none of those are guarded, the damage ships:
survives = 14 - 6*1.3(quality) - 2*1.6(trust) - 3*2.2(retention)
= 14 - 7.8 - 3.2 - 6.6 = -3.6%
You reported +14% and actually went backwards. That is Goodhart's law:
the number moved, the thing it stood for broke. Guard quality/trust/
retention and you'd catch it — and learn this tactic had no real growth
in it at all.| Set guardrails whenever… | The catch |
|---|---|
| You give a team a single headline target (growth, activation, revenue). | You can’t watch everything — naming guardrails means choosing which risks to instrument and which to accept. |
| A metric can be moved by degrading experience, trust, or . | Guardrails detect harm; they don’t prevent it. The real lever is choosing tactics that grow the metric honestly. |
| An experiment could win on the primary metric but lose the business. | Set the tripwires before the test, or you’ll rationalise the breach after seeing the win. |
An interviewer says: “Leadership wants weekly active users up 15%. How would you set this up so we don’t regret it?” A strong answer names the counter-metrics before touching the lever. WAU can be juiced by notification spam (guardrail: uninstall / notification opt-out rate), by counting trivial sessions (guardrail: a meaningful-action or day-7 retention rate), or by discount-driven logins (guardrail: contribution margin). Pick the two or three most gameable, set explicit tripwires, and pre-register them: “we ship only if WAU is up 15% AND uninstalls stay under X AND day-7 retention doesn’t fall.” Then you add: “and I’d expect the honest wins — a genuinely better core loop — to move WAU without tripping anything.” That last line is what separates growth from Goodhart.
Check yourself
A team hits its +20% orders target using a pre-checked add-on at checkout. Orders are up, and no guardrail was tripped — because the only guardrail was “orders.” What went wrong?
Your tactics threaten quality, trust, and retention, but you may only watch three guardrails. Best allocation?