Google Analytics Is Useless
A number on a screen looks like a fact. Most analytics numbers aren't. They fail two tests a number has to pass before a decision can rest on it: it has to be checkable, and it has to be movable. A figure that fails both is decoration, not measurement.
The two-counter test
Put two analytics tools on one page and read the visit counts side by side. They will not match. Off by a tenth is common; off by a third is not rare. Two tools, one page, two answers — and no way to tell which is right, because there is nothing underneath to check them against.
The gaps are structural. Events double-fire on single-page apps and fast reloads, counting one visit as several. A slice of traffic never reports at all — blocked by extensions, stripped by privacy settings. Figures land hours late and keep drifting for a day or two while the data "finishes processing." And the raw events — the underlying list, each with a stable identifier — are never handed back. Only the totals are.
A total with no underlying records cannot be recomputed, cannot be de-duplicated, cannot be reconciled against anything real. An estimate that disagrees with itself is not a measurement. It is a guess asking to be trusted.
The metrics it celebrates can't be moved
Set verification aside for a moment. Suppose the counts were exact. Most of them would still be useless, because they carry no money and answer to no lever.
Take click-through rate. It holds no price and no rate of return. A high CTR can be an expensive product bought once in a decade. A low CTR can be a cheap one bought three times a day. The percentage looks the same and means the opposite — it reveals neither what a customer is worth nor how often one arrives.
Bounce rate, average position, time on page, sessions per user — the same shape. Each is a ratio of one intermediate step to another, with no dollars and no time inside it.
And none of them is a control. A bid can be raised. A creative can be swapped. A budget can be moved. A click-through rate can only be observed — it is an after-effect of the real choices, never a choice. Tuning a number that can't be set directly is chasing a shadow across a wall.
What a number has to earn
A figure worth a place on the screen does two things a vanity metric can't: it ties to money, and it ties to time.
Money out per day. Results out per day. Profit per day. How long the budget lasts at the current rate. When the next sale is due. These answer the only question that pays a bill — is this earning, and how fast.
Under each one sits a real event with a stable key — a click, a sign-up, a paid order — read from the source that has no reason to inflate it: the payment system itself, not a tag that models and rounds. Every figure traces back to the events that built it, can be recounted by hand, and reconciles against the bank.
That is the line between an estimate and a measurement. One asks for trust. The other shows its work.
The standard
Two questions decide whether a number belongs on a screen. Can it be checked? Can it be moved? Web analytics fails both — so it isn't installed here, and the budget answers only to figures that pass.