Three tools, three numbers, one signup

It is Tuesday morning and you have three tabs open, all showing last week. The ad account says you got forty-seven signups. Your analytics says thirty-one. The list of actual accounts in your own records says twelve. Same week, same product, same handful of people who you are fairly sure exist. Three answers.

The instinct is to go looking for the bug. Most of the time there isn't one. The three tabs are not three attempts at the same measurement — they are three different questions that happen to use the same word. Until you know which question each one is answering, you will keep re-checking the arithmetic of a disagreement that was never arithmetic.

Why the ad account is always the most generous

An ad platform is not counting signups. It is counting signups it believes it caused, inside a window it chose. If that window is seven days after a click and one day after someone merely saw the ad, then a person who scrolled past your ad on Monday, thought nothing of it, searched your name on Thursday and signed up gets counted as a conversion. From where the platform sits that is a defensible claim. From where you sit it is a person who would probably have arrived anyway.

Two platforms will also both claim the same person, because neither can see the other. Run ads in two places at once and the sum of their conversion columns can exceed the number of humans involved. Nobody is lying. Each one is answering "how many conversions touched my ads", and the answers overlap.

Why analytics lands somewhere in the middle

Analytics counts events that fired in a browser, which means it inherits every strange thing browsers do. Some visitors block it outright and become invisible. Someone who reads on a phone and signs up on a laptop shows up as two people, one of whom did nothing. A confirmation page that reloads can fire the same event twice, and a count that quietly doubles is much harder to notice than one that quietly drops.

The nastiest version is the one where a single real purchase gets counted many times because the same action arrives from more than one source — once from your own page, and again from an imported set of actions you turned on months ago and forgot about. Nothing looks broken. The chart just grows a flattering bulge, and you spend a month explaining a success that never happened.

the tool closest to the money is the one that has to be right

Why your own records are the boring, correct number

Twelve is almost certainly the real one, because it only moves when something irreversible happens: money changes hands, a row gets written, an account exists that a human could log into. It has no window, no attribution opinion, no view of which ad deserves credit. That is exactly why it is useless for deciding which creative to keep, and exactly why it is the only number worth putting in a plan.

So stop trying to reconcile the three to the decimal. You will not get there, and the week you spend failing costs more than the discrepancy. What you do instead is assign each number a job.

One system of record per decision

Money decisions — can I afford to keep spending, did this month pay for itself, what is a customer actually worth — get answered from your own records. Only that.

Comparison decisions — which of two ads is doing better, which audience is cheaper, whether Thursday's creative beat Monday's — can safely use the platform's own numbers, because the inflation is roughly the same on both sides of the comparison. A number can be wrong in absolute terms and still be perfectly good at ranking things.

Behaviour questions — where people land, where they drop, which page they read before they bought — belong to analytics, which is the only one of the three that sees the path.

Write that mapping down somewhere you will see it. One sentence each. The value is not the document, it is that you stop re-litigating the choice every time the numbers diverge.

Then write down the expected gap

Once a month, put the three numbers side by side and record the ratio, along with one plain-language reason for it. "Ad account runs about three times our real count, because both platforms claim the same people and the view window is a week." That sentence is now a baseline, and baselines are what make surprises visible. If the ad account has run at three times reality for four months and this month it runs at eight, something changed — a window got widened, a second source got switched on, a page started firing twice. Now you are debugging a specific change instead of an eternal mystery.

Three patterns mean it is a real problem and not the usual noise. The gap flips direction and your own count is suddenly higher than analytics. The gap widens sharply without any change in spend. Or one number goes flat zero while the others keep moving, which is a different beast entirely — a zero that means nothing happened and a zero that means nothing was measured look identical on a chart and should never be treated the same way.

The one thing worth wiring up

If you do nothing else, make your own site tell you directly when the thing you care about happens — the signup, the purchase, the trial start — as one clean event you control rather than one you infer. It takes an afternoon. We accept those events pushed straight in, alongside the numbers we pull from the tools you have connected, and we would rather show you an honest empty panel than a confident number assembled out of three sources that disagree. A chart that admits it has nothing is a chart you can still trust next quarter.

The rule of thumb: believe the number closest to the money, use the others for ranking and for direction, and know the size of the gap between them well enough to notice the day it changes.