What is Click-through conversion?

A click-through conversion is credited when somebody clicked an ad and converted within the click window.

Compared with the alternative, the evidence is substantially stronger. A click is a deliberate act: the person saw the ad, chose to engage, and arrived somewhere the advertiser controls. There is a chain of behaviour rather than an opportunity to be seen.

That makes it the default currency of performance advertising and the right basis for most reporting. It is also worth being precise about what it establishes, because the word "conversion" in a platform report invites more confidence than the measurement supports.

It establishes sequence. A click happened, a conversion followed, within a period somebody chose.

It does not establish cause. The clearest demonstration is branded search: somebody who has decided to buy from a business types its name, sees the ad above the organic result, clicks it, and buys. The click-through conversion is recorded, the cost is real, and the advertising may have changed nothing except which link was used.

The same logic applies to retargeting against a warm audience, to any campaign serving people who were already in a purchase process, and to a degree to every campaign — because the people most likely to click are the people most interested, and the people most interested are the most likely to convert regardless.

The gap between credit and cause is therefore largest exactly where the reported performance is best, which is the single most consequential fact in performance measurement.

Accounts compare the platform's conversion figure against their own order count and find a difference in both directions.

Orders the platform never saw. Anybody who bought without touching the advertising is absent from its count and present in the order table.

Conversions outside the window. A purchase on day thirty-five under a thirty-day window is credited to nothing.

Journeys that were not observed. Consent refusals, blocked scripts, in-app browsers and cross-device paths all break the link between the click and the purchase.

Modelled conversions. Where the platform estimates conversions it could not observe, those cannot be matched line by line against anything.

Deduplication differences. Two orders from one session, a refunded order, a test order — counted differently by each system.

None of these is a fault. They are a consequence of two systems measuring different things for different purposes. What matters is picking one as the system of record for a given decision, and reconciling on a routine rather than in a panic.

Keep it as the headline. Click-through conversions, reported on a stated model and a stated window, are the comparable figure across channels. View-through belongs in its own column.

State the window. A seven-day and a thirty-day click window produce different numbers from identical delivery.

Wait for maturity. Conversions arrive across the window, so the most recent days are always incomplete, and the incompleteness is worse for slow-converting activity. Judging a campaign on figures that have not finished arriving is the most common cause of a good campaign being paused in its second week.

Separate brand from everything else. Branded search clicks convert at rates no other traffic approaches, and when they are mixed into a campaign with prospecting terms they lift the averages enough to hide a prospecting problem entirely.

Reconcile against the business. The order table, the CRM or the billing system, monthly, as a routine.

For the decisions where being wrong is expensive — whether a channel should exist, whether retargeting is doing anything, whether brand spend pays — the click-through figure is the wrong instrument, however carefully it is read.

Those are causal questions. A holdout test, a geographic split or a clean on-off answers them by withholding advertising from a comparable group and measuring the difference. The answer is coarser, slower and more expensive to obtain, and it is the only one that survives scrutiny from outside the marketing team.

The working arrangement most disciplined accounts reach: steer daily with click-through conversions, control monthly against the blended ratio, and settle the big questions periodically with a test.

Nothing about the measurement makes the underlying evidence stronger, and two things make it less lossy.

Server-side event collection recovers conversions that browser-side tags miss, which narrows the gap against the order table.

Sending back the outcome rather than the event — the lead that closed, the order that shipped, the customer who did not refund — moves the figure closer to what the business actually got, and gives any automated bidding something better to pursue at the same time.

A significant share of journeys begin on one device and finish on another, and the click-through count handles that unevenly.

Where the person is signed in to the platform on both devices, the join is usually made and the conversion is credited. Where they are not, the click on the phone and the purchase on the laptop are two unconnected events, and the campaign that produced the click is credited with nothing.

The effect is not random. It falls hardest on campaigns that reach people during discovery — typically mobile, typically browsing — and lightest on campaigns that catch them at the moment of purchase. So the same bias that favours lower-funnel activity through attribution generally is reinforced by the mechanics of device matching.

There is no fix available inside an ad account. What helps is the same thing that helps everywhere else: a signed-in experience where the business can reasonably ask for one, server-side events that carry an identifier the business already holds, and a periodic test that does not depend on joining anything.

An account handling click-through conversions well has a short, dull checklist in place.

One model and one window, stated on every report. Brand separated from prospecting. Conversions counted from server-side events where the engineering exists. A monthly reconciliation against the order table. Reporting periods held back far enough to have matured. View-through credit in its own column. And a note somewhere of when the last incrementality test ran and what it found.

None of that is sophisticated. Together it is the difference between a number that can be defended and a number that is simply reported.