KKindling

The Library / The operating cadence

Kill it or double down: deciding when a marketing play isn’t working

Two clocks decide every play’s fate: a leading-indicator checkpoint at 2–4 weeks, and a verdict clock set by the channel — weeks for paid, months for SEO. When the verdict clock runs out: cut it, double down, or give it more time — once, with a named change. Here are the rules.

Every startup graveyard of marketing budgets has the same two headstones. One says killed too early — the SEO effort abandoned at month three, exactly when the compounding was about to start. The other says kept too long — the ad campaign that limped through four quarters because nobody wrote down what success looked like, so failure never technically happened. Both deaths have the same cause: the decision had no structure, so it defaulted to mood.

The structure below — the Two-Clock System — is how we make these calls inside Kindling, written out so you can run it by hand. It has three parts: the clocks, the verdict rules, and the tiebreakers.

The two clocks

The killed-too-early error and the kept-too-long error are both clock errors — judging a channel on a timescale that isn’t its own. So every play gets two clocks the day it starts.

The leading checkpoint — 2–4 weeks, every channel. Not a verdict; a pulse. Is the leading indicator moving at all? For outbound that’s ICP reply rate; for LinkedIn it’s comments and DMs from target titles (never likes); for paid it’s click-through and cost per signup against a ceiling you wrote before spending; for SEO it’s impressions and ranking movement, which won’t appear before weeks 8–12 and that’s normal.

The verdict clock — set by the channel. Paid retargeting: 2–6 weeks. Outbound and ABM-lite: 6–8 weeks. LinkedIn organic and founder content: 8–12 weeks. Lifecycle email: 2–4 weeks. Referral and customer marketing: 4–8 weeks. SEO, AEO, and content: 6–12 months — budget for a full year or don’t start. The clock is written into the play before it launches, which means the decision date exists before anyone is emotionally invested in the answer.

The verdict rules

Cut it

When the clock has run and any of these is true: leading indicators are flat AND at least one meaningful variation was tried; volume is fine but the leads are off-ICP (the percent-of-pipeline-in-ICP test); or the play is crowding out a channel that’s working. One prohibition, absolute: never kill a long-clock channel early on lagging metrics. “SEO hasn’t driven pipeline by month three” is not evidence of anything except a calendar.

Double down

When leading and lagging indicators improve together across a 3–4 week rolling window and the unit math holds at 2x. Scale on the speedometer principle (Stage 2 Capital’s framing): accelerate while indicators hold, decelerate the moment they degrade — increments, not leaps. And judge segments, not aggregates: an average is usually one strong segment subsidizing three weak ones, and the double-down belongs to the segment.

Give it more time — once

Allowed exactly once per play, only with a named change — new angle, new audience, new offer — and a new deadline. Re-running the same test unchanged is prohibited. This rule exists because “keep testing” with nothing named is how zombie channels survive on hope for quarters at a time. If the named change misses too, the play is cut.

The tiebreakers, in order

Some verdicts are genuinely close. Break ties in a fixed order, so the tiebreak is a rule instead of a mood. First, ICP quality: which play attracts buyers who look like your best customers? A channel delivering fewer-but-right beats one delivering more-but-wrong. Second, compounding: at seed, the tie goes to the channel that builds an asset — rankings, an audience, a list — over the one renting attention. Third, founder energy: a channel the founder hates ships at half quality and dies the first busy month. On a team of one, the founder is the team, and pretending otherwise is how plans meet reality badly.

The honesty clause

At seed-stage volumes, nothing clears statistical significance — twelve replies and three meetings is a quarter’s outbound data. The system doesn’t pretend otherwise. Every verdict is a directional judgment with the reasoning shown, logged where you’ll see it next quarter. The log is the point: calls with written reasoning can be revisited, learned from, and survive being wrong. Verdicts delivered as oracles can’t. Over four quarters, the decision log quietly becomes the most valuable marketing document the company owns — a record of what your market actually rewarded, in your own numbers.

The system assumes plays with written success numbers and dates — that setup lives in the 90-day plan — and it runs inside the weekly cadence as question four. Inside Kindling, this whole system is called the call: every play runs on its clocks, and when one runs out, the call gets made — cut it, double down, or give it more time — with the reasoning logged. It starts, as everything does, with the free audit.

Common questions

How long should you run a marketing campaign before deciding it isn’t working?

Two different clocks answer that. Check leading indicators at 2–4 weeks for every play. But hold the verdict to the channel’s own speed: paid retargeting shows its hand in 2–6 weeks, outbound in 6–8, LinkedIn organic and founder content in 8–12, and SEO or content takes 6–12 months. Killing a slow channel on a fast clock destroys good bets; letting a fast channel limp past its clock burns quarters.

When should you kill a marketing channel?

Three conditions, any one sufficient once the clock has run: leading indicators are flat at checkpoint AND at least one meaningful variation was tried; volume looks fine but the leads are outside your ICP (check percent of pipeline in-ICP); or the play is crowding out a channel that is working. And one prohibition: never kill a long-clock channel early using lagging metrics like pipeline — that is the single most common way startups destroy their best compounding bets.

When should you double down on a marketing channel?

When leading and lagging indicators both improve consistently across a 3–4 week rolling window and the unit math still holds at twice the spend or effort. Then scale incrementally on the speedometer principle: accelerate while indicators hold, decelerate the moment they degrade. Judge segments, not aggregates — an average hides the one audience segment doing all the work.

Is it ever right to keep testing a struggling marketing play?

Exactly once per play, and only with a named change — a different angle, audience, or offer — and a new deadline. Re-running the same test unchanged is prohibited: “let’s give it more time” with nothing named is how zombie channels survive for quarters. If the named change also misses its checkpoint, the play is cut, no further debate.

How do you make marketing decisions without statistically significant data?

You accept that at seed-stage volumes every verdict is a directional judgment, and you compensate with process instead of pretending the sample is bigger: write the reasoning down with every call, use tiebreakers in a fixed order (ICP quality of what the channel attracts, then compounding potential, then founder energy), and keep a decision log so the reasoning improves quarter over quarter. Recommendations with reasoning survive being wrong; oracles don’t.

Every play deserves a fair clock ✨

Kindling writes the clocks and success numbers into every play before it starts — and makes the call, with reasoning shown, when time runs out.

Start with the free audit

Free. About ten minutes. No card required.

Keep reading

The weekly marketing cadence

Where the clocks get checked every Monday.

How to choose marketing channels

Pick plays whose clocks you can afford.

Sources: verdict-clock ranges reflect operator practice and the channel benchmarks in our pipeline math guide; the speedometer principle is Stage 2 Capital’s framing. The Two-Clock System and verdict rules are ours.