Standing in for a number you cannot measure

The number you actually want doesn’t exist

Some things you want to measure just aren’t measurable. Not “hard to measure with the tools I have” but genuinely not observable as a single number. Customer trust. Whether your product feels premium. How close you are to burning out your best customer. There’s no sensor for any of that.

So you reach for something else. Something you can count, that you believe moves in the same direction as the thing you actually care about. That stand-in is a proxy metric.

I run my businesses on proxy metrics constantly, usually without stopping to name them as such. Net revenue retention stands in for “are my existing customers getting more value over time.” Support ticket volume per active user stands in for “is the product confusing.” Time to first value stands in for “will this signup convert.” None of these are the thing itself. They’re the closest countable shadow of the thing.

This matters because a proxy metric is a bet, not a fact. The bet is: this number and the real thing move together closely enough that I can act on the number instead of the real thing. Sometimes that bet is solid. Sometimes it quietly breaks, and you don’t find out until the business does something odd that the dashboard didn’t predict.

Why you can’t just measure the real thing

Take “customer happiness.” You’d love a direct readout. What you actually have is a handful of imperfect signals: survey scores from the small fraction of people who answer surveys, refund requests, churn, unsolicited praise or complaints, maybe a support rating after a ticket closes. Each one is a proxy. None of them is happiness. They’re behaviors and self-reports that correlate with happiness for most people, most of the time.

The same is true further up the funnel. “Product-market fit” isn’t a number, it’s a judgment about the relationship between what you built and what a market wants. People use retention curves, organic growth rate, or the “how disappointed would you be if this went away” survey as proxies for it. Each one captures a slice. None of them is fit itself.

This isn’t a flaw in your measurement setup. It’s the nature of the thing. Anything abstract, emotional, or judgment-based (trust, satisfaction, brand strength, team morale) has to be approached through proxies, because there’s no direct instrument for it. The skill isn’t finding the perfect measurement. It’s picking a proxy that’s close enough to be useful and staying honest about where it diverges.

Proxies I actually use

A few examples from running things day to day, described plainly rather than with numbers I’d have to make up to sound impressive.

Churn as a proxy for dissatisfaction. Churn is easy to count: a customer stops paying. But churn is a proxy, not the real signal. Some churn is dissatisfaction. Some is a customer’s business closing, a card expiring, a founder switching roles and losing budget authority. If you treat all churn as “we made someone unhappy,” you’ll chase fixes for problems that don’t exist and miss the involuntary churn you could actually reduce with something as boring as a payment retry.

Session length as a proxy for engagement. Longer sessions look good on a dashboard. But a long session can mean someone is deeply engaged, or it can mean they’re lost and can’t find what they need. The number is real. What it means depends on context you have to bring yourself, usually by actually watching a few sessions or reading the support tickets from that week.

MRR as a proxy for business health. Monthly recurring revenue is one of the most load-bearing proxies a subscription business has, and it’s genuinely useful. But it’s a proxy for health, not health itself. MRR can climb while margins erode, while your best customers are quietly souring, while you’re one platform policy change away from losing a third of it. It tells you the size of the engine, not whether the engine is about to seize.

Response time as a proxy for support quality. Fast replies feel like good service, and often they are. But a fast, wrong answer is worse than a slower, right one. If you optimize purely for time-to-first-response, you’ll get faster replies and possibly worse resolutions, because the team starts answering quickly instead of correctly.

In every case, the proxy is legitimate and worth watching. It’s just not the thing. The gap between them is where judgment has to live.

Where proxies quietly stop working

The classic failure mode is that once you start managing to a number, people (including you) start optimizing the number instead of the thing behind it. This isn’t dishonesty, it’s just what happens when a metric becomes the target. A support team measured only on response time will get faster at responding. Whether the customer’s problem actually got solved is a separate question that the metric stopped asking.

I’ve seen this in my own funnels. Optimizing purely for a low-friction conversion metric (like completed checkouts) can quietly reward stripping out the parts of a page that helped serious buyers self-select, at the cost of pulling in customers who never should have converted in the first place and churn fast. The topline number moves the right direction. The thing it was supposed to represent, sustainable paying customers, moves the wrong one.

The second failure mode is drift. A proxy that was tightly linked to reality when you set it up can drift apart from it as the business changes. Session length was a decent proxy for engagement when your product was simple. Add a slow-loading feature and the same metric now partly measures frustration instead. Nobody changed the metric. The world around it changed.

Neither failure means throw the metric out. It means the metric needs periodic reality checks against the thing it’s supposed to represent, not blind trust because the number is easy to pull from a dashboard.

How I try to use a proxy responsibly

A few habits that keep proxy metrics useful instead of misleading, for what they’re worth from someone running these numbers day to day rather than theorizing about them.

Name the proxy out loud. Before acting on a number, I try to finish the sentence “I’m using this as a stand-in for ___.” If I can’t finish it, I don’t actually know what decision the number is supposed to inform.

Watch more than one proxy for anything important. A single number can be gamed or drift without you noticing. Two or three proxies that are supposed to move together, and don’t, are your earliest warning that one of them has stopped meaning what you think it means.

Go look at the raw thing occasionally. Read actual support tickets instead of only the ticket-volume chart. Watch a handful of session recordings instead of only the average session length. The proxy is for scale. The raw material is for calibration, and it’s the only way to catch drift before it costs you.

Hold the number loosely when a decision is big. For a small pricing tweak, act on the proxy and watch what happens. For a decision that’s expensive to reverse, like cutting a feature because its usage proxy is low, spend the extra time to understand why before you trust the number completely.

None of this makes proxy metrics unreliable in some special way you should avoid. Nearly every number you run a business on is a proxy for something you can’t directly see. The goal isn’t to escape that. It’s to know which bet you’re making, and to keep checking whether it still holds.

If you want more of this kind of plain-English breakdown of the numbers behind running a small business, from someone actually using them, come take a look at what else we’ve put together here.

Get new guides and videos first — join the Telegram channel.