Never Track a Metric You Won't Act On

There is a specific kind of monthly report that has trained a generation of founders to feel informed. Twelve charts. Impressions up. Click-through rate up four tenths of a point. Engagement holding steady. Somebody says the word momentum, everyone nods, and the meeting ends on time with nothing moved.

I have replaced agencies over reports like that. Not because the numbers were wrong, usually they were fine. Because nobody in the room had written down, before the report arrived, what any of those numbers would have to do before somebody moved money.

The rule I run is short. Never track a metric you won't act on. A dashboard you do not check is worthless, and the most dangerous state a team can be in is not blindness. It is looking directly at data and doing nothing, on a schedule, with confidence.

A metric earns its place by having a rule attached

The unit of a working dashboard is not the number. It is the sentence you wrote before the number showed up.

I write them in a fixed form: if [metric] moves [how much] for [how long], then [action], and [name] owns it. One I have run for years reads, if CAC is up 20% for two weeks, pause scaling.

Every slot in that sentence is doing work. Twenty percent is high enough to survive normal weekly noise, so you are not chasing a bad Tuesday. Two weeks is long enough to rule out a holiday and short enough that you have not already spent the quarter. Pause scaling is an action a specific person can take on a Monday morning without calling another meeting or asking permission. And the name at the end is the difference between a rule and a wish, because a threshold nobody owns gets discussed rather than executed.

"Monitor CAC closely" fails all four parts. It is a feeling wearing a verb.

Two piles: decision metrics and diagnostics

Once you write rules this way, every number on your dashboard sorts itself. Either you could finish the sentence, in which case it is a decision metric and it stays, or you could not, in which case it is a diagnostic and it belongs in a file you open when you are investigating something specific.

Click-through rate goes in the diagnostics pile, and this is the part people argue with me about.

CTR measures something real. The trouble is that it is a term inside an equation, and no decision hangs on it standing alone. CTR up with cost per outcome flat means you bought cheaper clicks from worse people. CTR down while cost per acquisition improves means your ad started repelling the wrong audience, which is usually good news. In both cases the action is decided by the number underneath. So the movement of CTR by itself never tells you to do anything, and putting it on the wall alongside numbers that do is how a team learns to read charts without changing behavior.

The honest objection is that falling CTR is an early warning of creative fatigue, and it is. Fine. Then write the rule. Name the frequency threshold you consider dangerous, the window you will read it over, and what ships when it trips. If you can write that sentence, promote CTR to the dashboard with my blessing. If you cannot, you were not managing creative fatigue. You were watching it.

That test is the whole discipline, and it is easier to run on your own numbers than to read about. If you want a second read on which of your metrics have rules behind them and which are just familiar, that is a good use of a 20-minute conversation.

The thresholds I actually run

A format is only useful with numbers in it, so here are the rules I put on the board for most accounts. Argue with the thresholds if your business is different. The format is the part I would not change.

If CAC is up 20% for two weeks, pause scaling. Never raise budget more than 20% at a time in the first place, because platform algorithms optimize around your current spend and a jump forces a relearning period you paid for. If LTV to CAC sits under 1 to 1, stop acquiring, since volume only makes a loss larger. Around 3 to 1 is healthy, and above 5 to 1 usually means you are underspending rather than winning. Payback under six months is excellent and twelve to eighteen months is acceptable, past that you are financing customers rather than acquiring them. Monthly churn above 5% means retention gets fixed before another dollar goes into acquisition. Retargeting frequency caps at five a day for high intent audiences and two a day for content, above that you are buying annoyance. And if more than a third of your traffic lands in direct, that is a tagging problem, not your customers typing your URL from memory.

That is nine rules. Most dashboards I inherit carry four times that many, largely because nobody was ever required to write an action next to a number.

When a metric has no rule, rewrite the metric

Cutting is not the only move. Sometimes the number is close to useful and the fix is to restate it until a decision fits.

Email open rates are the cleanest example. There is no action attached to an open rate, and after the privacy changes of recent years there is barely a fact attached to it either. At Cubii we tracked revenue per email subscriber per month instead. That number told us exactly what a new subscriber was worth, which told us exactly what we could afford to spend to acquire one, which is a decision about money that a person can make on a Monday.

The same rewrite applies to anything lagging. Revenue on the P&L is a record of work finished months ago. Pipeline velocity, sales-qualified-lead-to-close rate, time to close and expansion revenue are the numbers that move first, which is why those are the ones I show investors and the ones I put rules against.

What this looks like when it is running

On The RealReal account we built dashboards showing CAC, LTV and ROAS by channel, by audience and by creative variant, refreshed daily rather than assembled monthly by somebody outside the building. When a creative started fatiguing we knew inside 48 hours instead of 30 days. Paid scaled from $300K to $2M+ per month at 7× ROAS on the way to $100M+ per year in attributed revenue, with CAC down 40%. What made that refresh worth building was that every number on it already had a threshold and an owner waiting for it.

At GlacierGrid, back when it was called Therma, we rebuilt a single "Request Demo" retargeting ad into a four tier, thirty day sequence. Click-through rate went up 4x. Cost per demo fell from $450 to $120. Notice which of those two I would have shipped the change for, and which one I would have quoted in a report.

A rule also needs a standing moment or it quietly stops being read. At PartnerSlate the fix was unglamorous: Monday metrics, daily updates in Slack, Friday prep for the week ahead. Execution consistency improved about 40%, mostly because decisions stopped waiting for someone to notice.

What to do first

Print your dashboard on one page. Next to every metric, write the sentence: if this moves this much for this long, then we do this, and this person owns it. Cut everything where you cannot finish it. Half the page usually goes, and the deleted numbers move to a diagnostics file you open only when a decision metric trips.

Then put the survivors on a standing weekly slot with a name against each one. You will notice the meeting gets shorter and something actually changes in it, which is the only test of a dashboard that has ever mattered to me.

If you want the sharper version of this, the free Pain Ladder Diagnostic is the same discipline applied to demand instead of dashboards.